AI transcript
0:00:03 The content here is for informational purposes only,
0:00:05 should not be taken as legal business tax
0:00:06 or investment advice,
0:00:09 or be used to evaluate any investment or security
0:00:11 and is not directed at any investors
0:00:14 or potential investors in any A16Z fund.
0:00:18 For more details, please see a16z.com/disclosures.
0:00:21 – Hi everyone, welcome to the A6 and Z podcast.
0:00:22 I’m Sonal.
0:00:25 So this week to continue our 10-year anniversary series
0:00:27 since the founding of A6 and Z,
0:00:28 we’re actually resurfacing
0:00:30 some of our previous episodes
0:00:33 featuring founders, Mark Andreessen and Ben Horwitz.
0:00:34 If you haven’t heard the latest episode
0:00:36 with Stuart Butterfield turning the tables
0:00:38 as the entrepreneur interviewing them,
0:00:41 please do check that out and other episodes
0:00:45 in this series on our website at a6andz.com/10.
0:00:48 But this episode was recorded in 2016
0:00:50 on the five-year anniversary
0:00:52 of Mark’s Wall Street Journal op-ed
0:00:55 on why software is eating the world.
0:00:57 And it features me and Scott Cooper
0:01:00 asking Mark and Ben about what’s changed since
0:01:03 and how software is now programming the world.
0:01:05 And we discuss everything from simulations
0:01:09 to distributed systems to other key computing shifts.
0:01:10 Welcome guys.
0:01:11 – Hey, thank you.
0:01:13 – Okay, so let’s just kick things off.
0:01:15 One of the things that I want to understand
0:01:16 is that it’s been since fund one,
0:01:18 which is what, seven years ago?
0:01:20 Yeah, seven years ago.
0:01:21 A lot’s changed in seven years.
0:01:22 And I’ve actually heard you argue, Mark,
0:01:24 that things have accelerated in that time period,
0:01:27 more so than previous decades before.
0:01:29 So what do you think are the biggest shifts now
0:01:31 that are important to us in this newest fund
0:01:34 and what changed in that period, like the biggest things?
0:01:37 – So in fund one, when we started,
0:01:39 we thought that our timing was really good,
0:01:42 despite the fact that I think the world thought
0:01:43 our timing was really bad
0:01:45 and starting a new venture capital fund.
0:01:47 And the reason why we thought that was that
0:01:50 there were three gigantic new platforms
0:01:51 hitting all at the same time,
0:01:56 which was kind of unprecedented in the history of technology.
0:01:58 One was mobile, the second was social,
0:02:00 and the third was cloud.
0:02:02 And that really proved out
0:02:04 through the course of the early history
0:02:06 that the applications on top of those,
0:02:11 particularly mobile and cloud were just spectacular.
0:02:13 And I think we’re coming a little bit
0:02:17 to the end of the first phase of the,
0:02:19 some of the obvious applications
0:02:20 that could be built on those things
0:02:22 and we’re moving into some new areas.
0:02:25 – Yeah, so let me go to the foundations.
0:02:26 So there’s different ways of looking at it.
0:02:27 The foundation levels.
0:02:30 One is Moore’s law has really flipped.
0:02:31 And this actually has happened.
0:02:32 I think this actually has happened
0:02:33 over the last seven or eight years,
0:02:35 actually almost exactly over the life of the fund,
0:02:37 which is for many, many years,
0:02:40 Moore’s law was a process of the chip industry
0:02:42 bringing out a new chip every year and a half
0:02:45 that was twice as fast as the last one at the same price.
0:02:47 And that continued for 40, 50 years.
0:02:49 And that’s, by the way, what resulted in everything
0:02:51 from mainframes, mini computers, PCs,
0:02:55 and then smartphones, about seven, eight, nine,
0:02:57 10 years ago, that process actually started to come in
0:02:59 and the way that it had worked up until then.
0:03:00 So chips have kind of topped out
0:03:02 at a speed of about three gigahertz.
0:03:04 And a lot of people have said, therefore,
0:03:06 like progress in the tech industry is gonna stall out
0:03:07 because the chips aren’t getting faster.
0:03:09 I think what’s actually happened is Moore’s law
0:03:10 has now flipped and the dynamic now,
0:03:12 instead of increased performance has reduced cost.
0:03:15 You now have this dynamic where every year, year and a half
0:03:16 the chip companies come out with a chip
0:03:18 that’s just as fast, but half the price.
0:03:21 And so this is this sort of, just this massive deflationary
0:03:23 force, I think in the technology world.
0:03:27 And I actually also suspect in the economy more broadly,
0:03:28 we’re basically computing is just becoming free.
0:03:31 Basically what we do in this business is we just kind of
0:03:33 chart out the graphs and then just kind of assume
0:03:34 at some point you’re gonna get to the end state
0:03:36 and the end state is gonna be the chips are gonna be free,
0:03:38 which means chips will be embedded in everything.
0:03:40 You’ll be able to use chips for literally everything.
0:03:42 And we’ve never lived in a world before where you can do that.
0:03:44 So that’s the first one.
0:03:46 Second one is just the obvious implication from that,
0:03:48 which is all those chips will be on the network, right?
0:03:50 So all those chips will be connected to the internet.
0:03:52 They’ll all be on wifi or mobile carrier networks
0:03:53 or wired networks or whatever,
0:03:55 but they’ll all fundamentally be on the internet.
0:03:57 You know, that’s something that’s not happening
0:03:59 at a very rapid pace.
0:04:01 And then the third is the continuation of the piece
0:04:03 that I wrote actually five years ago,
0:04:04 which was called software eats the world,
0:04:06 which is basically just say if you’re gonna live in a world
0:04:08 in which there’s gonna be a chip in every physical object.
0:04:11 And if you live in a world in which every physical object
0:04:12 therefore is going to be networked,
0:04:14 it’s gonna be smart because it has a chip
0:04:15 and it’s gonna be connected to the network.
0:04:17 Then basically you can then program the world.
0:04:19 You can basically write software
0:04:20 that applies to the entire world.
0:04:22 So you can write software that all of a sudden applies
0:04:24 to all cars or you can write software that applies
0:04:27 to all, you know, everything flying in the sky.
0:04:29 You can write software that applies to all buildings.
0:04:31 So you can write software that applies to, you know,
0:04:35 all homes or all businesses or whatever, all factories.
0:04:36 And so all of a sudden you can kind of,
0:04:37 you can program the world.
0:04:39 That’s really just starting.
0:04:40 And I think a lot of the,
0:04:42 there’s a number of things that make the entrepreneurs
0:04:44 we’re seeing these days in many ways more interesting
0:04:45 and more aggressive than entrepreneurs
0:04:47 who’ve seen the past and part of it is they just assume,
0:04:49 if there’s something to be done in the world,
0:04:51 there must be a way to write software to be able to do it.
0:04:54 That’s at a new level of power sophistication.
0:04:57 It’s a new scope of what the tech industry can do.
0:04:59 The consequence of that for us as a fund
0:05:01 is that we find ourselves evaluating business plans
0:05:03 and funding companies that are in markets
0:05:04 where I think seven or eight years ago
0:05:06 we would have never anticipated operating.
0:05:09 – So Mark, does that mean that there’s no new innovation
0:05:11 in platforms themselves and everything?
0:05:12 All the innovation will be applications
0:05:14 that ride on that existing infrastructure?
0:05:16 Or do you think there’s also the opportunity
0:05:19 to build a new platform, even given some of those trends?
0:05:20 – I think there are new platforms
0:05:21 and I think there will be new platforms.
0:05:23 I just think there’ll be different kinds of platforms
0:05:24 than we’ve had in the past.
0:05:26 The idea of a platform in the tech industry, as you know,
0:05:29 up until five or 10 years ago was there is a new chip
0:05:31 that has new capabilities is faster
0:05:33 and then therefore you build a new operating system for it.
0:05:36 And that might be Windows or it might be iOS
0:05:37 or whatever it is.
0:05:39 The platforms that we’re seeing getting built these days
0:05:42 are distributed systems, so scale out systems.
0:05:44 Instead of being built on a chip necessarily
0:05:45 with new unique capabilities,
0:05:47 they are platforms that are going to build
0:05:48 across lots of chips.
0:05:49 And so they’re in computer science terms,
0:05:51 they’re distributed systems.
0:05:52 Cloud is one of the first examples, right?
0:05:55 So anybody who uses AWS can now go on
0:05:57 and can program an application on AWS
0:05:59 that will run across 20,000 computers
0:06:02 and they can run it for an hour and it’ll cost 50 bucks.
0:06:06 And that’s a kind of platform that did not exist before.
0:06:08 And by the way, there are many specific elements to that.
0:06:10 So for example, we’ve seen the rise of,
0:06:11 in that category, seen the rise of Doop
0:06:14 and otherwise the spark for distributed data processing.
0:06:16 We’ve seen in financial technology,
0:06:18 we’ve seen the rise of Bitcoin and cryptocurrency,
0:06:20 which is a literally distributed platform for currency
0:06:22 and for exchanging value.
0:06:25 And now we’re seeing the emergence of a major new platform,
0:06:28 which is AI, machine learning and deep learning,
0:06:30 which is inherently, the great thing about machine learning
0:06:33 and deep learning is they’re inherently parallelizable.
0:06:34 They can run across many chips
0:06:36 and they get very powerful as you do that.
0:06:40 And you can do things in AI today as a consequence
0:06:41 of being able to run across many chips
0:06:43 that you just couldn’t even envision
0:06:44 doing five or 10 years ago.
0:06:46 – So let’s talk about the rise of the GPU
0:06:47 as part of this next platform ship.
0:06:49 I mean, I think the biggest surprise people have had
0:06:51 is that this is the graphical processor unit,
0:06:54 which is something that was developed in the gaming industry
0:06:57 for really high resolution graphics processing
0:06:59 and is now finding, I guess, unexpected.
0:07:02 Is it a surprise to us that it’s finding uses
0:07:05 in these new platforms like VR, AR, deep learning?
0:07:06 – It’s actually interestingly,
0:07:08 it’s a new application of an old idea
0:07:09 back when I was getting started 30 years ago
0:07:11 working in physics labs.
0:07:13 If you wanted to run just a normal program,
0:07:16 you just buy a normal computer and run the program.
0:07:19 But if you wanted to do, run a program,
0:07:21 many physics simulations had this property
0:07:24 where you would want to run a very large number
0:07:25 of calculations in parallel, right?
0:07:27 And so you could basically divide up a problem
0:07:29 of simulating anything from a black hole
0:07:32 or to different kinds of biological simulations.
0:07:34 You could basically write these algorithms
0:07:35 in a way that they could run,
0:07:36 you could basically parcel the problem
0:07:38 into many different pieces
0:07:39 and then run them all in parallel.
0:07:41 There was actually in the old days,
0:07:42 there was actually a whole industry
0:07:43 of what we’re called vector processors,
0:07:46 which were literally these kind of sidecar computers
0:07:48 that you would buy and you would hook up to your main computer
0:07:49 and they would let you run these parallel problems
0:07:51 much faster.
0:07:53 And so literally 30 years later, the GPU is a,
0:07:54 it’s basically a vector processor.
0:07:56 It’s basically a sidecar processor that sits along the CPU
0:07:58 and runs these parallel problems much faster.
0:08:01 And graphics are a natural application of that,
0:08:03 but as it turns out, graphics aren’t the only application.
0:08:05 – Yeah, actually interestingly,
0:08:07 and I was at a company making one of these
0:08:11 called Silicon Graphics and the applications then,
0:08:13 whereas Mark was saying a lot of physics applications,
0:08:16 computational fluid dynamics and simulating,
0:08:18 flight simulation and all these kinds of things
0:08:20 that are hard physics to calculate.
0:08:21 When you go into the virtual world
0:08:23 and you’re simulating the physics of the real world,
0:08:24 guess what?
0:08:27 You need the exact same processor to do it.
0:08:31 So it’s a super logical conclusion to what’s been going on,
0:08:35 but I think we’re also in the world of big data,
0:08:39 seeing kind of more reasons to do just lots of math in parallel.
0:08:42 And so it’s an exciting application.
0:08:43 – Yeah, you talk about platforms.
0:08:45 One of the really interesting hardware platforms
0:08:47 is emerging right now is NVIDIA,
0:08:49 which is a very well-established public chip company,
0:08:50 but very successful to your point,
0:08:53 doing graphics chips for a very long time,
0:08:55 has become seemingly overnight.
0:08:57 It’s really, of course, the result of years of work,
0:08:59 but seemingly overnight has become the market leader
0:09:03 in both not just GPUs, but also in chips being used for AI.
0:09:05 And it’s basically extensions of the GPU technology.
0:09:08 And we see this overriding theme,
0:09:09 which is kind of an amazing thing,
0:09:12 which is basically every sharp AI software entrepreneur
0:09:15 that comes in here is now building on top of NVIDIA’s chips,
0:09:17 which is, of course, a very different outcome
0:09:18 than entrepreneurs of previous years
0:09:20 who would have built other kinds of programs
0:09:22 primarily on top of Intel chips.
0:09:24 – We’ve mentioned AI and machine learning a couple of times here.
0:09:25 And one of the interesting things,
0:09:28 at least I think we see in the industry is,
0:09:29 at the same time we’ve got startups doing it,
0:09:32 we also see some of the very large established players
0:09:34 investing significantly in AI and machine learning.
0:09:37 So certainly Facebook and Google, Apple and others
0:09:39 are obviously building big operations.
0:09:41 How do you think about the universe
0:09:42 from an investment perspective?
0:09:44 What are the kinds of things that actually led themselves well
0:09:47 to startup opportunities in the AI space
0:09:49 versus things that actually might make sense,
0:09:52 kind of living inside of one of the larger companies
0:09:53 like a Facebook or Google?
0:09:55 – Yeah, so AI is extremely broad.
0:09:58 And I think one of the challenges that people have with it
0:10:01 is they try to paint it as a narrower thing than it is,
0:10:04 but one can think of it as an entirely new way
0:10:07 to write a computer program.
0:10:12 And so then it’s applicable to the universe of problems.
0:10:15 So there are things that advantage a big company.
0:10:18 If you’re building AI to analyze consumer internet data,
0:10:20 like that’s hard to take Google on at that.
0:10:22 They do have an awful lot of data.
0:10:26 And Facebook with AI, computing power matters
0:10:28 and the data set matters.
0:10:30 Having said that, there are a lot of areas
0:10:32 where nobody has any data yet
0:10:37 in the areas of healthcare and the areas of autonomy.
0:10:40 So there’s lots and lots of opportunities.
0:10:44 And there’s also interesting ideas
0:10:46 about, well, is there a better user interface
0:10:49 than the smartphone using AI techniques
0:10:50 and then what is the form of that?
0:10:51 – What do you mean by that
0:10:53 when you say there’s a better user interface?
0:10:55 – Well, if you think about a smartphone,
0:10:58 it was kind of an advance over what we used to call
0:11:03 the WIMP interface, Windows icons.
0:11:05 What was it?
0:11:06 – Menus.
0:11:07 – Menus.
0:11:07 – What was it, P?
0:11:08 – Pointer.
0:11:09 – Pointer, right.
0:11:12 – Which was like a big advance
0:11:15 over the text-based interface of DOS.
0:11:18 And then the smartphone with a touch interface,
0:11:20 so it was more of a direct manipulation
0:11:21 was an advance over that.
0:11:23 And so you’d go, okay, well,
0:11:27 but that’s not actually what people do in life, right?
0:11:30 Like it’s anthropologically,
0:11:34 it’s a backward step in terms of the natural interface
0:11:36 so that we’ve become accustomed to,
0:11:38 like for example, natural language.
0:11:40 With AI, you get into a world
0:11:44 where things like natural language and natural gestures
0:11:46 and so forth become much more plausible.
0:11:48 So there’s potentially an opportunity
0:11:52 to build interfaces for things that you couldn’t before.
0:11:54 I mean, I think there’s one really interesting thing,
0:11:57 which I’m sure, and I know that Google and Apple
0:11:59 and all the giant companies are very focused on,
0:12:02 which is how do you replace the current set
0:12:03 of user interfaces with it?
0:12:05 But there’s another dimension,
0:12:08 which is what are all the applications
0:12:10 that you just couldn’t have before
0:12:13 because you couldn’t build a workable user interface for it.
0:12:17 And AI seems very promising in those areas.
0:12:18 – You didn’t mention Amazon,
0:12:19 which is sort of the stealth player here
0:12:21 with Echo and Alexa.
0:12:23 I mean, really, George and Horace have been home.
0:12:26 Well, in a way, they’ve got an interesting advantage
0:12:28 in that they’re not tight
0:12:31 to the last generation of user interfaces
0:12:34 so that they don’t have to pay the strategy tax
0:12:39 for shoehorning in their AI into, say, the iPhone.
0:12:40 And that’s something.
0:12:41 – Yeah, that’s worth pointing out.
0:12:43 There’s sort of two kind of classic rules of thumb
0:12:44 in this industry.
0:12:46 One is for major new advances,
0:12:47 especially in things like interfaces,
0:12:49 if you don’t own a platform, you can’t do them.
0:12:52 And so the assumption I think had been up until recently,
0:12:53 that it would have to be Google or Apple
0:12:54 that does these kind of natural language
0:12:57 or interface advances ’cause they own iOS and Android.
0:13:00 The other rule, of course, is the exact opposite rule,
0:13:01 which is the one that Ben mentioned,
0:13:03 which is the problem that big established companies
0:13:06 get into is this, what he referred to as the strategy tax,
0:13:09 which is basically big companies with existing agendas
0:13:11 have to sort of fit their next thing
0:13:12 into their existing agenda,
0:13:14 and they often compromise it in the process.
0:13:16 And so it’s sort of this ironic twist of fate
0:13:18 that Amazon has all of a sudden taken the lead
0:13:19 from Google and Apple.
0:13:22 Even though Amazon famously flopped with their phone,
0:13:23 which is sort of the obvious place
0:13:24 where you would have a voice interface,
0:13:26 it didn’t matter because they came out with this new product,
0:13:29 which is basically the smart speaker called Echo.
0:13:32 And the fact that all of a sudden Amazon didn’t have a phone,
0:13:33 all of a sudden became an advantage
0:13:35 ’cause they could just do the clean,
0:13:36 actual breakthrough product
0:13:38 without worrying about tying it into the existing strategy.
0:13:40 – Right, and those are all still big companies, though,
0:13:42 is I’m not really hearing where startups
0:13:44 can really play in this space,
0:13:46 especially when you’re describing this huge data network effect
0:13:48 that all these big companies have.
0:13:50 – A year ago, we would have probably been sitting here
0:13:52 and say that AI was gonna be likely,
0:13:53 would be a domain of big companies
0:13:55 because of this sort of thing of like,
0:13:57 okay, only big companies can afford
0:13:58 the very large number of engineers
0:13:59 that are required to do AI.
0:14:01 Only big companies can afford the amount of hardware
0:14:03 required to do AI.
0:14:03 And then only big companies
0:14:06 can get the giant data sets required to do AI.
0:14:07 In the last 12 months, what we’ve seen basically
0:14:10 is all three of those changing very fast
0:14:11 into the advantage of startups.
0:14:14 We’ve seen a lot of AI technologies actually,
0:14:16 actually now interestingly standardizing.
0:14:18 So going to open source and then the next step
0:14:20 is gonna be they’re gonna go to cloud.
0:14:22 And we think we’re right on the verge of that.
0:14:23 We think all the major cloud providers
0:14:25 are gonna be providing AI as a service
0:14:27 and they’re gonna really radically reduce
0:14:29 the amount of technical knowledge you need to apply AI.
0:14:30 And so that plays very well to the startups.
0:14:32 – So there will be like an AWS for AI.
0:14:33 – Yeah, exactly.
0:14:34 And that may be literally AWS
0:14:35 or it may be Google or Microsoft
0:14:38 or all three of them in some combination
0:14:40 or it may be other companies yet to emerge.
0:14:42 – An example of the open source like TensorFlow,
0:14:43 like Google releasing TensorFlow.
0:14:44 – Yeah, and this is a big deal.
0:14:45 And of course, yeah, it’s right.
0:14:46 So Google open source to pretty significant part
0:14:47 of how they do deep learning.
0:14:48 And that actually now is something
0:14:50 that other companies can pick up and use directly.
0:14:51 And we see actually a lot of,
0:14:52 not only a lot of companies,
0:14:53 but like a lot of university,
0:14:54 a lot of student projects now
0:14:56 just kind of can pick that up and run with it.
0:14:59 So this technology is kind of trickling down very fast.
0:15:01 – Just this past weekend, we had a hackathon.
0:15:03 And I think most of the teams
0:15:06 had some machine learning AI component into their hacks.
0:15:07 And these are college kids.
0:15:11 – Yeah, if you’re a 21 year old junior in college
0:15:12 and you’re doing some project,
0:15:16 it’s just kind of, it’s becoming rapidly becoming very obvious
0:15:17 that you would have AI be part of it,
0:15:19 which was very much not the case even full months ago.
0:15:20 And that’s a direct to your point.
0:15:22 That’s a direct consequence of the open sourcing
0:15:24 and kind of this knowledge spreading out.
0:15:25 The second thing was the hardware costs.
0:15:27 And there again, the cloud, AI and the cloud,
0:15:29 just the existence of the cloud
0:15:30 is bringing down hardware costs across the board,
0:15:32 but AI and the cloud is gonna bring that down even further.
0:15:34 And by the way, these trends all slam together.
0:15:37 So you get what I think in a year is gonna be very common.
0:15:39 These sort of AI supercomputing chips
0:15:41 with the AI algorithms in the cloud
0:15:43 available to anybody for a dollar, right?
0:15:44 And so there’s gonna be this massive deflation
0:15:47 of hardware cost on that side.
0:15:48 These big data sets are interesting.
0:15:50 Ben made the case that the startups
0:15:51 can assemble big data sets.
0:15:53 And I think that there are certainly examples of that.
0:15:55 We also see another thing happening,
0:15:57 which is the newest generation of experts in deep learning
0:15:59 or many of them are specializing in the idea
0:16:01 of deep learning applied against small data sets.
0:16:03 If you talk to those folks, what they’ll tell you is,
0:16:06 oh, basically, they’ll basically say is
0:16:08 primitive and crude deep learning required big data sets,
0:16:09 but the really good stuff doesn’t.
0:16:11 It is small data sets are fine.
0:16:13 And so that’s still very early,
0:16:14 but it’s extremely enticing.
0:16:16 It’s an extremely enticing idea
0:16:18 because it really brings a lot of these problems
0:16:21 to your point further into being tractable
0:16:22 for small companies.
0:16:24 But actually, one of the things you can do
0:16:25 with these, especially with these GPUs
0:16:27 is you can literally use the same tools
0:16:29 that are used to make video games.
0:16:32 And you can create simulated versions of the real world.
0:16:34 And then you can actually let the AI train
0:16:35 inside the simulation.
0:16:36 And so if you’re building a new self-driving car
0:16:38 or a drone or something like that,
0:16:39 you can actually create simulated worlds
0:16:42 in which there are everything from earthquakes,
0:16:46 to floods, to thunderstorms, hail storms.
0:16:48 You can create birds, swarms of birds.
0:16:51 You can literally simulate the real world environment.
0:16:53 And then you can let the AI actually train inside that world.
0:16:54 And actually, it’s funny,
0:16:56 the AI actually has no idea it’s training
0:16:57 in the virtual world.
0:16:58 It’s learning just the same
0:16:59 as if it were learning in the physical world.
0:17:02 And so again, for startups with access to cloud-based AI,
0:17:05 you could potentially run basically millions of hours
0:17:07 of simulated training at a very low cost
0:17:08 and all of a sudden catch up to big companies.
0:17:12 – Interestingly, the very famous AI project
0:17:14 that Google did with DeepMind,
0:17:17 that whole dataset came from the game “Playing Itself”.
0:17:19 So, there wasn’t some dataset
0:17:21 that Google had collected over 20 years.
0:17:24 It was the game “Playing Itself”.
0:17:26 – So you guys have both mentioned simulations a few times.
0:17:27 Why are they so important?
0:17:28 ‘Cause I feel like there was this period,
0:17:29 like maybe even a decade ago
0:17:31 where simulations were almost frowned upon
0:17:33 as this promised thing
0:17:35 that didn’t really actually deliver
0:17:37 in what you needed to be able
0:17:40 to navigate complex environments in real life.
0:17:41 – Yeah, well, it’s interesting.
0:17:45 So was AI, was frowned upon 10 years ago,
0:17:48 saying it was all, it didn’t work.
0:17:50 And particularly neural nets and deep learning
0:17:53 were the most frowned upon area.
0:17:55 And there’s been similar kind of breakthroughs
0:17:56 with simulation.
0:17:58 And first of all, so if you think about the field
0:18:02 of data science and what you do with data,
0:18:04 there’s you have a giant set of data,
0:18:07 which is always historical in nature,
0:18:08 and you can analyze that.
0:18:10 And maybe it’s predictive of the future,
0:18:12 but oftentimes it’s not.
0:18:15 And we see this in particular in things like,
0:18:17 really dynamic things where the past affects the future,
0:18:20 like say stockpicking or the weather
0:18:23 or other kinds of things where data analysis
0:18:24 doesn’t get you an accurate answer.
0:18:26 Simulation is a flip side of that
0:18:28 where you can say, okay, here are all the entities
0:18:32 in the world and let’s generate their behavior over time.
0:18:34 And then their actual behavior feeds back
0:18:37 into the simulation, which is critical,
0:18:39 a critical component.
0:18:42 Historically, that’s been difficult at scale,
0:18:45 but there have been some really important breakthroughs
0:18:47 lately, particularly from a company
0:18:49 that we’re invested in called Improbable,
0:18:52 which is able to do very large scale,
0:18:55 scale out simulation using cloud computing techniques
0:18:59 and some very important new technology
0:19:00 that they’ve developed.
0:19:03 And so you can get a really complete picture of the world.
0:19:06 And as Mark was saying, you can actually generate
0:19:09 your own dataset rather than collecting it
0:19:10 for certain kinds of situations.
0:19:12 Yeah, let me add the thing to that.
0:19:14 So one way to think about it is it’s expensive
0:19:15 to make things happen in the real world.
0:19:17 Like it’s expensive to change things in the real world
0:19:18 because the real world is physical
0:19:20 and causing physical changes to happen.
0:19:21 I mean, everything from building roads to flying planes,
0:19:23 all these things are very expensive.
0:19:24 And then things in the real world,
0:19:26 changes have serious consequences, right?
0:19:28 And so if you, depending on where you put the dam
0:19:29 or where you put the airport
0:19:31 or what your evacuation plan you have for the city
0:19:32 and if something bad happens,
0:19:35 like these decisions have huge consequences.
0:19:36 Which banks you bail out.
0:19:38 Which banks you bail out.
0:19:39 Which banks you don’t bail out.
0:19:41 And so you always have these consequences
0:19:43 and people who have to make these decisions
0:19:44 are often flying blind
0:19:45 ’cause they don’t have any real sense
0:19:46 of what’s gonna happen
0:19:47 as a consequence of their decisions.
0:19:49 In contrast, if you can simulate a world
0:19:50 and if you can run an experiment,
0:19:53 if you can simulate the real world or some portion of it
0:19:55 like the highway system or the banking system or whatever.
0:19:58 And then you can basically introduce change
0:19:59 into that simulation
0:20:00 and you can see what the consequences are.
0:20:02 It’s very cheap to do that because Moore’s Law
0:20:04 and the collapse of chips and the rise of cloud computing
0:20:06 and all these other things we’ve been talking about
0:20:08 all of a sudden make it very cheap to run the simulations.
0:20:10 It’s much cheaper to do it in the simulated world.
0:20:11 And then there are no consequences.
0:20:13 You run a simulation and everything goes wrong
0:20:15 and everybody dies
0:20:16 or the entire financial system collapses or whatever.
0:20:17 It doesn’t matter.
0:20:19 You just erase it and you run it again.
0:20:20 – Yeah, you have infinite testability.
0:20:22 – Well, I went and challenged that.
0:20:25 There is Elon Musk simulation in which case
0:20:27 the consequences are quite dire for us.
0:20:27 – There is, yes.
0:20:29 There is a scenario that we’re all living in a simulation.
0:20:30 – Right, if we’re living in one place.
0:20:32 – In which case I would argue it’s gone badly awry
0:20:34 as evidenced by the current political situation.
0:20:36 – There is no do over button in this simulation.
0:20:38 – Yes, and then you basically, again,
0:20:40 you look at the progress of Moore’s Law
0:20:41 and the rise of these new technologies
0:20:43 and you say, okay, how about instead of running
0:20:44 one simulation, let’s run a million simulations
0:20:46 or let’s run a billion simulations
0:20:47 and let’s try every conceivable thing
0:20:49 we can possibly think of and let’s imagine,
0:20:51 let’s literally model all potential future states
0:20:54 of the world and then let’s decide which one of those,
0:20:57 which path is the one that leads to the best consequences.
0:21:00 And so we can then make these very big real world decisions
0:21:04 with a lot more foreknowledge of what will unfold afterwards.
0:21:06 – Maybe just to get concrete on some opportunities,
0:21:08 what are the other areas and maybe it’s life sciences
0:21:10 or what are some of the other kind of more tangible areas
0:21:11 that you think in your term,
0:21:13 as you think about kind of deploying this fund
0:21:15 or beyond over the next, you know, five to 10 years
0:21:17 that might be interesting for, you know, people to think
0:21:19 about in the context of real world applications
0:21:20 of this technology.
0:21:21 – Yeah, so as Mark was saying,
0:21:24 we’re coming into this era of new platforms
0:21:27 and with the intersection of health and computer science,
0:21:30 what we’re saying is really exciting new platforms
0:21:34 around data and around basically you being able
0:21:36 to get much more information about someone’s health
0:21:40 from a variety of techniques that have been developed,
0:21:43 you know, based on the kind of historic breakthroughs
0:21:47 and sequencing the genome, but beyond that as well,
0:21:51 where we can get really, really powerful data about people
0:21:54 and understand them better.
0:21:55 And once you have that data about people
0:21:58 when you can be predictive of diseases
0:22:00 that they might get or things that are wrong
0:22:02 and you aggregate that into a platform,
0:22:05 then you can actually make new scientific discovery
0:22:06 off it as well.
0:22:08 So that’s one interesting area.
0:22:11 There’s, if you think about the AI platform itself,
0:22:14 one of the things about it is the hardware
0:22:17 that’s been built for it or that’s been built historically
0:22:20 is for a completely different kind of computer programming.
0:22:23 And we’ve seen Google already announced a chip
0:22:25 to power their deep learning cloud.
0:22:28 And, you know, similarly there’s new breakthroughs
0:22:31 and quantum computing, which at least on the surface
0:22:33 look like they may be very promising
0:22:37 for much more powerful deep learning systems and so forth.
0:22:39 So there’s a lot of things
0:22:41 that are coming out of these platforms.
0:22:43 And then, you know, as we get to a chip and everything,
0:22:47 the platforms to run and manage and understand those chips
0:22:50 are equally as exciting.
0:22:52 – So, you know, one of the themes that’s come up through here
0:22:55 is that tech is reaching into places it never did before.
0:22:57 I mean, every company is becoming a tech company
0:23:00 or they have tech inside, or as Benedict likes to say,
0:23:01 growing the tech industry,
0:23:04 the reality is it’s permeating everywhere.
0:23:06 And the question I have for us
0:23:08 is that we are founded on this thesis
0:23:10 that software is eating the world, that’s our premise.
0:23:12 And yet we’ve seemed to have been making
0:23:14 a lot of hard investments, you know,
0:23:18 if you count things like Soylent, Oculus, Nutribox.
0:23:21 So are we changing our thesis about hardware
0:23:23 as a result of this software eating the world?
0:23:24 – No, I don’t think so.
0:23:26 I mean, I think that what we see
0:23:29 with the companies that you’ve named are interesting.
0:23:31 So, Oculus, I think we would all agree
0:23:33 that the software component of Oculus
0:23:37 is both more complex, has many more people working on it
0:23:39 and is kind of the core of the investment.
0:23:42 Sometimes if you have a breakthrough technology,
0:23:45 then you require a new hardware to actually support it.
0:23:46 And that’s the case there.
0:23:49 And I think that Soylent and Nutribox,
0:23:51 both of them apply computer science techniques
0:23:54 and information technology to get people to optimal health.
0:23:56 And that’s what we’re doing there.
0:24:00 So I think we’re big, big believers that, you know,
0:24:03 in the last 100 years, the great breakthroughs
0:24:05 and knowledge have been the breakthroughs
0:24:08 of people like Alan Turing and Claude Shannon
0:24:10 who gave us a new model of the world
0:24:12 and how to understand it.
0:24:16 And companies that build on that fundamental knowledge
0:24:18 breakthrough are what we’re about
0:24:20 and will continue to be about that.
0:24:22 – Even if some of them may ship their products in a box.
0:24:25 – Yes, the package is not a technology.
0:24:27 Let’s talk a little bit about SaaS.
0:24:27 As you’ve probably seen,
0:24:29 there’s been actually a bunch of acquisitions
0:24:31 in this space recently, but what’s left to do there?
0:24:34 So is the new platform, the salesforce.com
0:24:36 and others of the world or are there actually
0:24:37 both kind of vertical applications
0:24:39 and or are there other platforms
0:24:41 that actually might exist over time in that market?
0:24:43 – So there’s SaaS as the metaphorical
0:24:45 in the cloud version of all the stuff
0:24:49 that we had built over the previous, you know, 30, 40 years.
0:24:54 So that’s like workday, salesforce,
0:24:58 success factors, you know, the kind of big categories.
0:25:00 The thing that we believe that’s changed
0:25:02 as you go from on premise to the cloud
0:25:05 is the technology is so much easier to adopt
0:25:08 that we’re now seeing software applications
0:25:10 for things that you just would never do
0:25:13 as a software application because the cost of,
0:25:15 as we used to say in the old days, screwing it in
0:25:19 and paying the army of Accenture consultants
0:25:21 to get it going just wasn’t worth it
0:25:24 for say expense reporting, which, you know,
0:25:28 Concur, of course, built a really powerful product in that.
0:25:31 But like there was no packaged software
0:25:34 for expense reporting in the same way that there is now.
0:25:37 And I think there’s a gigantic number of categories
0:25:39 in everything that you do in business
0:25:41 that can be automated in that way.
0:25:43 In addition to that, you can scale down
0:25:45 to very, very small companies.
0:25:47 Companies below thousands of employees
0:25:49 never bought Oracle financials.
0:25:51 It would have been insane to do so,
0:25:53 but they’re absolutely buying, you know,
0:25:55 NetSuite and things like that.
0:25:59 And then beyond that, you now it becomes economical
0:26:02 and very interesting to build vertical applications
0:26:03 for industries.
0:26:06 So to build an application that revolutionizes,
0:26:09 say the real estate industry or something like that,
0:26:13 or the construction industry is becoming extremely viable
0:26:15 and not just as a niche business,
0:26:18 but as a real venture capital based kind of activity.
0:26:20 – One of the consequences that will be interesting
0:26:22 to watch play out is that historically,
0:26:24 enterprise software has been described
0:26:27 as represented by companies like Oracle SAP IBM.
0:26:29 Like that stuff was really only accessible
0:26:31 to the largest companies,
0:26:33 the top 500 or 1000 companies in a country.
0:26:36 And then in particular, only in a handful of countries,
0:26:39 those businesses, their revenue and their customer base
0:26:40 have always been dominated by, you know,
0:26:42 two or 3000 companies globally that are these,
0:26:44 you know, these giant multinational companies
0:26:45 that we’ve all heard of.
0:26:48 So big companies had this sort of inherent advantage
0:26:50 versus a lot of mid-sized small companies.
0:26:51 And then companies in the US and Western Europe
0:26:53 had this big advantage versus companies
0:26:54 in other parts of the world
0:26:56 where the companies, the large companies
0:26:58 and the large companies in the US and Western Europe
0:27:00 could just afford to make technology investments
0:27:01 that small and mid-sized companies
0:27:02 all over the world couldn’t make.
0:27:05 The sort of changes in SaaS that Ben described,
0:27:07 they lead you to an interesting conclusion
0:27:08 which is it may actually be interesting
0:27:12 for a smaller company or a company not in the US
0:27:13 or Western Europe to be able to adopt
0:27:16 the next generation of SaaS and cloud technology.
0:27:18 It’s almost like the folks who’ve been able to skip
0:27:19 landline telephones or just go straight to mobile phones.
0:27:21 You can just leapfrog the old stuff
0:27:22 ’cause you never had it
0:27:24 and you can just start using the new stuff out of the box.
0:27:26 And then the big established companies
0:27:27 might have a harder time adapting
0:27:29 ’cause they’ve made these giant investments
0:27:30 in the old systems and it’s hard to just jump
0:27:31 to the new thing.
0:27:34 And so there may be a power shift happening
0:27:36 from on the one hand large companies
0:27:37 to small and medium companies
0:27:40 that can now more aggressively adopt technology faster.
0:27:42 And then from companies in the US and Western Europe
0:27:43 to companies all over the world
0:27:45 that can also do the exact same thing.
0:27:48 And so at the very least a leveling of the playing field
0:27:49 and possibly even a national shift in balance
0:27:51 where small and mid-sized companies all over the world
0:27:53 may all of a sudden get a lot more competitive.
0:27:55 – So you’ve got kind of democratization at one point
0:27:56 and then to your point,
0:27:58 there’s one version of internationalization
0:28:00 which is adoption across international communities.
0:28:02 So how do you think about then
0:28:03 the other aspect of internationalization
0:28:04 which is company formation?
0:28:07 Should we then expect to see more new company formation
0:28:10 outside the US partly as a result of some of these trends
0:28:14 and why won’t we see or will we see 50 Silicon Valley’s
0:28:15 over the next 20, 30, 40 years
0:28:17 and how do you all think about what the strategy
0:28:19 should be vis-a-vis those opportunities?
0:28:21 – That would be probably the most amazing thing
0:28:23 for the world that could happen
0:28:25 in the realm of business and economics.
0:28:28 So we’re hoping for it
0:28:31 and certainly building kind of help trying
0:28:33 to build technologies that would facilitate it.
0:28:37 And I think the world has never been kind of more ripe
0:28:39 for that kind of thing.
0:28:40 Having said that, look,
0:28:42 there are real network effects,
0:28:44 geographical network effects
0:28:46 and Silicon Valley obviously has the biggest one
0:28:50 in technology and you always have to keep in mind
0:28:52 and this is something that gets lost
0:28:55 is there are no local technology companies, right?
0:28:58 There’s nobody who sells, you know,
0:29:01 internet search to Wyoming.
0:29:02 That’s not like a viable thing.
0:29:07 So when you’re competing globally, it does matter.
0:29:08 You know, do you have the best people?
0:29:09 Do you have the best executives?
0:29:10 Do you have the best engineers?
0:29:12 Do you have access to money?
0:29:15 Like all these things become real competitive things.
0:29:17 So we still are believers in Silicon Valley
0:29:20 and we’re very hopeful that the rest of the world grows
0:29:23 and that we can participate in that as well,
0:29:25 but that’s TBD.
0:29:27 – This is an interesting macro kind of thing
0:29:28 that’s happening in a lot of the, you know,
0:29:30 one of the really kind of negative stories
0:29:32 is that there’s basically the world is starved
0:29:34 for innovation and growth.
0:29:36 One of the data points you point to on that
0:29:39 is there’s now $10 trillion of money
0:29:41 in being held in government bonds,
0:29:42 governments all over the world,
0:29:44 trading at what’s called negative yield.
0:29:46 This is literally like the equivalent of a savings account
0:29:48 where instead of the bank paying you interest,
0:29:51 you have to pay the bank interest to hold your money.
0:29:53 And so there’s literally $10 trillion of capital parked
0:29:56 around the world that is actually losing money
0:29:58 as it sits there, which means people cannot find
0:30:00 enough productive places to deploy capital.
0:30:02 The conventional view, if you just pick up the newspaper
0:30:03 and read the economic section,
0:30:04 the horrible this is and how it means
0:30:06 the world is start for growth.
0:30:10 The optimistic side of it is there’s $10 trillion of money
0:30:12 sitting on the sidelines waiting for something productive
0:30:13 to be done with it.
0:30:15 What could be productively done with it?
0:30:17 New kinds of healthcare, new kinds of education,
0:30:20 new kinds of consumer products, new kinds of media,
0:30:21 new kinds of art, new kinds of science,
0:30:24 new kinds of self-driving cars, new kinds of housing,
0:30:27 all of these things that need to be done all over the world.
0:30:29 And so the world has never been more ripe
0:30:32 for a very large wave of innovation
0:30:34 that would actually be quite easy to finance.
0:30:36 A lot of the times you just can’t get things done
0:30:37 ’cause you don’t have money.
0:30:38 That’s just kind of the constant state of the world
0:30:39 for a very long time.
0:30:41 And now ironically, we live in a world
0:30:42 where the opposite is true.
0:30:44 There’s actually, quote unquote, too much money.
0:30:46 And more money than ideas.
0:30:46 More money than ideas.
0:30:48 Which really can’t be true.
0:30:49 It can’t be true, right?
0:30:50 Unlock the ideas.
0:30:52 Human creativity is boundless.
0:30:54 And so if you can get more smart people
0:30:55 around the world educated
0:30:57 and with the skills required to do these things
0:30:58 and if you can get them in environments,
0:31:00 either create new environments to do that
0:31:01 or figure out how to get more of the people
0:31:03 from other places in environments
0:31:04 where they can do new things,
0:31:06 we could do all kinds of new things globally.
0:31:08 And it’s something that we hope to contribute to
0:31:10 but I think is a very big opportunity for the world.
0:31:11 So do you think we’re getting to the point
0:31:13 where it’s kind of geopolitical risk
0:31:16 and rule of law issues that limit adoption
0:31:17 or deployment of some of these new technologies
0:31:19 in other countries outside the US?
0:31:23 It sounds like it’s less so technological advancement.
0:31:25 Well, I would say there’s bad news and good news.
0:31:27 So the bad news is we frequently have delegations
0:31:29 of folks coming into the Valley from all over the US
0:31:31 and all over the world.
0:31:32 And they basically come in
0:31:33 and it’s economic delegations of different kinds
0:31:35 or politicians or whatever.
0:31:36 And they come in and they’re like, okay,
0:31:38 what can we do to have our own Silicon Valley?
0:31:39 And then you kind of sit down with them
0:31:42 and you kind of go through ABCDEF, all these things.
0:31:44 Well, you don’t want rule of law,
0:31:46 you want ease of migration, you want ease of trade,
0:31:49 you want deep investments in scientific research,
0:31:51 you want no non-competes,
0:31:52 you want fluid labor laws to let companies
0:31:54 very easily both hire and fire.
0:31:55 You want the ability for entrepreneurs
0:31:57 to be able to start companies very quickly.
0:31:59 You want bankruptcy laws that make it very easy
0:32:01 to move on and start another company.
0:32:04 And at some point, the visitors get this stricken look
0:32:05 on their face and they’re like, well,
0:32:07 at the end of it, they’re like, okay,
0:32:08 but what if we want Silicon Valley
0:32:10 but we can’t do any of those things.
0:32:12 And so that’s the bad news.
0:32:14 – Well, they can hire Donald Trump to run their country.
0:32:15 – It’s sad that’s ironic that we have this guy running
0:32:17 for president who would seriously move us backwards
0:32:19 on a number of those topics.
0:32:21 So even we struggle with these things, right?
0:32:23 Like I would argue the formula is fairly well known.
0:32:25 It’s just people do not want to apply it
0:32:27 for reasons that have a lot to do with politics
0:32:29 and have a lot to do with other issues.
0:32:30 The good news is it can be done.
0:32:32 And then the other good news is it is happening.
0:32:35 And there are very, very, very exciting things happening
0:32:36 throughout much of the world.
0:32:38 They’re very active now startup scenes
0:32:41 all through South America, Brazil, Argentina, Buenos Aires.
0:32:42 Amazing things are happening in India.
0:32:44 There’s all kinds of startup activity
0:32:45 throughout the Middle East.
0:32:47 There’s startup activity now throughout Africa.
0:32:50 There’s, obviously China’s been a gigantic success story.
0:32:52 Korea has all kinds of interesting things happening.
0:32:55 So there are lots and lots of extremely positive
0:32:57 early indications of what’s possible
0:32:59 in many places all over the world.
0:33:00 That said, there are very big political questions
0:33:02 about whether or not those founders
0:33:03 are gonna be able to operate an environment
0:33:05 that’s really gonna let them succeed to the level
0:33:06 that they should be capable of doing.
0:33:08 The big reason that we raise the fund
0:33:12 and are excited about the fund is it is a backing
0:33:14 of our core belief system here,
0:33:18 which is we believe in the creativity
0:33:21 and genius and intelligence of human beings
0:33:23 and the entrepreneurs that we see
0:33:26 and come to Silicon Valley and around the world.
0:33:29 And we believe that these people absolutely have
0:33:32 the ability to change things and are changing things.
0:33:36 And there’s plenty of room to improve the world.
0:33:38 And there’s plenty of ideas to do so.
0:33:42 And that’s really what we’re about with fund five.
0:33:44 So let’s talk a little bit about kind of company building
0:33:45 and founders in particular.
0:33:49 So, undoubtedly you had a very distinct view
0:33:50 of what types of founders you wanted to back
0:33:53 when you started the firm now seven years ago.
0:33:54 How has that evolved?
0:33:56 If at all over time, what has changed either
0:33:57 in terms of the types of founders you see
0:33:59 or the types of qualities you see
0:34:01 that actually make founders successful
0:34:02 that’s caused you to either augment
0:34:05 or rethink some of the initial foundations for the firm.
0:34:08 You know, I think a lot of the things,
0:34:10 we had this great advantage when we started the firm
0:34:13 that we ourselves were founders.
0:34:16 I think that we’ve probably gotten,
0:34:18 I would say more risk tolerant
0:34:20 in our view of founders over time,
0:34:21 even though sometimes the risk–
0:34:22 – What do you mean by that?
0:34:24 What do you mean by getting risk more risk tolerant?
0:34:25 – Well, we have this thing we say at the firm,
0:34:27 which is we’re much more interested
0:34:29 in the magnitude of the strength
0:34:31 than the number of the weaknesses.
0:34:33 We always believe that intellectually,
0:34:36 I think that some of the number of weaknesses
0:34:39 were fairly terrifying early on,
0:34:43 just ’cause you do have a lot of founders
0:34:46 with a very small amount of experience these days,
0:34:47 which is also part of their strength
0:34:50 in that it’s hard to rewrite the world
0:34:52 if you’re too steeped in the world.
0:34:56 And so I think over time, we’ve kind of doubled down on that.
0:34:59 And really, the founders who have figured out
0:35:02 something really important or who are true geniuses
0:35:07 or have will to power that we can’t even contain in the room,
0:35:10 when they bring those things to the table,
0:35:11 whatever is wrong with them,
0:35:14 we tend to overlook and work with them on that.
0:35:16 And if they’re strong enough in those areas,
0:35:18 the really interesting thing for us has been,
0:35:21 those weaknesses do go away pretty quickly.
0:35:23 And that’s probably the biggest learning is,
0:35:25 I’d say we went in thinking that,
0:35:28 but we’ve gotten even more extreme
0:35:31 in our commitment to that kind of philosophy.
0:35:33 So almost in financial terms,
0:35:35 you’re buying volatility to a certain extent.
0:35:38 – Well, I think buying volatility in the sense
0:35:40 that we’re buying people have world-class strengths
0:35:42 where we care about them,
0:35:45 and regardless of whatever else.
0:35:46 There’s volatility in that,
0:35:49 but you can have a different kind of volatility.
0:35:51 You can have people who have gigantic weaknesses
0:35:54 that are spectacular without having the strengths.
0:35:57 And we’re not trying to buy that kind of volatility.
0:35:58 – How do you know, though,
0:35:59 that they’re going to be the ones
0:36:01 to actually build the companies at scale?
0:36:03 Because there seems to be this inflection point
0:36:05 where the very thing that makes you a founder
0:36:07 that’s going to punch through this tough industry
0:36:10 is also the thing that’s pretty much going to hold you back
0:36:11 from really building your company
0:36:13 in a really meaningful way
0:36:15 if you think you can do everything your way.
0:36:17 And there seems to be an inherent contradiction in that.
0:36:19 – I think that that would be right
0:36:21 if founders did not evolve.
0:36:23 So I think-
0:36:23 – And some don’t.
0:36:24 – And some don’t.
0:36:26 Like some don’t and they get stuck
0:36:28 and they can’t get past that point.
0:36:32 But it’s a real common characteristic in great founders
0:36:34 that they want to know absolutely everything
0:36:35 about the company and how it works
0:36:38 and every knob and every button.
0:36:43 And they really would like, have a strong desire
0:36:45 to actually be able to do every job
0:36:47 in the company themselves if it came down to it.
0:36:51 But those kind of founders also have great ambition
0:36:54 and it’s very logical and easy to understand
0:36:58 that there’s never actually been a gigantic long,
0:36:59 really important long lasting company
0:37:02 that had like five employees that those just don’t exist.
0:37:05 And so if you’re gonna have to have a bigger company
0:37:10 than that, you have to think about the company,
0:37:12 not only from the scale perspective,
0:37:15 but from the perspective of the people working there.
0:37:18 And how are you gonna get great people to work with you
0:37:21 if you’re literally making a redecision in the company?
0:37:22 And I think that, look,
0:37:24 not every founder can let go of that.
0:37:26 And sometimes it’s a psychological flaw
0:37:29 rather than a desire for greatness.
0:37:32 And if it’s a psychological flaw that they can’t overcome,
0:37:36 then it’s just like any flaw that any of us have.
0:37:37 We can’t stop eating ice cream or whatever.
0:37:40 And there’s nothing we can do at that point,
0:37:42 like we can give them the logical explanation,
0:37:44 but they’ve got to fix themselves.
0:37:46 One of the things that we’ve seen even in the short time
0:37:49 that the firm has been in business is companies
0:37:51 staying private longer, taking longer times to IPO.
0:37:53 What are some of the implications of that
0:37:54 on the company building process?
0:37:57 How do you kind of balance that new reality?
0:37:59 If it is a new reality around how companies stay private
0:38:01 with how you think about building management teams
0:38:03 and other issues around the company.
0:38:05 – Yes, I think this gets back to probably
0:38:09 one of the more neglected parts of company building,
0:38:10 which is like, what is the company culture?
0:38:11 What does it believe?
0:38:13 What’s our way of doing things?
0:38:14 When we come to work every day,
0:38:16 what does quality mean?
0:38:18 How do we prosecute an opportunity?
0:38:23 And the kind of philosophy onboarding,
0:38:25 training into that culture and so forth.
0:38:27 And so you kind of have to develop a philosophy
0:38:29 like what kind of employees do you want?
0:38:31 How do you want them to behave when they get there?
0:38:32 How do people contribute?
0:38:34 – As we’re getting close to wrapping up here,
0:38:36 what would be one piece of advice that you might give
0:38:37 either from a management perspective,
0:38:39 from a go-to-market perspective?
0:38:40 What would be a takeaway for people
0:38:42 listening to this podcast?
0:38:44 – From a management perspective,
0:38:46 I think the most common mistake that founders make
0:38:50 is they make decisions based on management decisions
0:38:52 and organizational design decisions
0:38:56 based on very kind of proximate perspective.
0:38:57 So what’s my perspective?
0:39:00 What’s a person I’m talking to perspective?
0:39:03 What’s my HR person’s perspective?
0:39:05 Without like taking the time to go,
0:39:07 okay, like how does everybody in the entire company
0:39:08 see this decision?
0:39:10 And how will they see it once it’s made?
0:39:13 Is it motivating people in the way that I think it will?
0:39:16 And let’s look past the person I’m talking to
0:39:18 feeling good about what I’m saying
0:39:20 and really make this for the long-term health
0:39:22 of the organization.
0:39:24 – Single biggest strategic piece of advice
0:39:25 we just see across all of our companies
0:39:27 is literally people just need to raise prices.
0:39:29 People need to charge more for their products and services.
0:39:31 The good news is you have all these new founders
0:39:33 with many different backgrounds who have come in.
0:39:35 Many of them have never run companies before
0:39:36 or run sales forces before.
0:39:38 And so they have these extremely sophisticated views
0:39:40 on things like products and design and engineering.
0:39:42 And then I think in some cases,
0:39:45 relatively naive views on how to actually prosecute
0:39:47 a campaign to be able to get the world to use your product.
0:39:50 And so the temptation we see from many founders
0:39:53 is to have a one-dimensional view of what I call
0:39:54 a one-dimensional view of the relationship
0:39:56 between price and volume.
0:39:58 Which is if I price my product cheap,
0:39:59 then I sell more of it.
0:40:01 ‘Cause the assumption is just that people just make
0:40:03 purchase decisions based on cost.
0:40:05 And so you drive down prices, you drive up volume.
0:40:07 And by the way, a lot of the history of the tech industry,
0:40:08 like the chip industry is,
0:40:10 drive down prices, drive up volume.
0:40:14 But a lot of startups really suffer from having that view.
0:40:16 Instead, we encourage companies to adopt
0:40:17 what I call kind of the two-dimensional view,
0:40:19 which is the advantage of raising prices.
0:40:21 Actually, there’s a couple of advantages.
0:40:22 So one big advantage, if you raise prices,
0:40:25 you can afford a bigger sales and marketing effort.
0:40:28 A lot of companies have prices that are actually too low
0:40:30 to be able to mount the kind of sales and marketing campaign
0:40:32 required to get people to ever actually buy the product.
0:40:34 And I call this the too hungry to eat problem, right?
0:40:38 I’m not selling enough, but I’m not selling enough
0:40:39 ’cause I don’t have the sales and marketing coverage required
0:40:41 to actually get the product out there.
0:40:43 And I don’t have that ’cause I’m charging too little.
0:40:44 And as a consequence, I’m not selling any,
0:40:46 despite my low prices.
0:40:47 The other really interesting thing
0:40:49 is that for a very large number of products,
0:40:50 it turns out if you charge higher prices,
0:40:52 the customers take the product more seriously.
0:40:53 They impute more value into it
0:40:55 when they’re making their purchase decision.
0:40:56 And then once they’ve purchased,
0:40:57 they’ve made a bigger commitment to it.
0:41:00 And in particular, anybody selling anything to businesses,
0:41:02 businesses will take something that they had to pay
0:41:03 a lot of money for a lot more seriously
0:41:05 than something that they didn’t have to pay
0:41:06 very much money for.
0:41:08 So you can get a much higher level of engagement
0:41:10 and stickiness and actual use of your product
0:41:11 if you charge more.
0:41:12 – Going through this,
0:41:13 this definitely has felt like swimming upstream
0:41:14 for the last several years.
0:41:16 We see some glimmers that more folks
0:41:17 are starting to figure this out.
0:41:19 – Okay, well, that’s all we have time for.
0:41:20 I think this is the first time
0:41:22 I’ve actually had all you guys together on the podcast
0:41:24 since we did our fifth anniversary podcast
0:41:25 a couple of years ago.
0:41:26 Kind of amazing how much has changed
0:41:28 even in that short amount of time.
0:41:29 So thank you.
0:41:30 Thanks everyone.
0:00:05 should not be taken as legal business tax
0:00:06 or investment advice,
0:00:09 or be used to evaluate any investment or security
0:00:11 and is not directed at any investors
0:00:14 or potential investors in any A16Z fund.
0:00:18 For more details, please see a16z.com/disclosures.
0:00:21 – Hi everyone, welcome to the A6 and Z podcast.
0:00:22 I’m Sonal.
0:00:25 So this week to continue our 10-year anniversary series
0:00:27 since the founding of A6 and Z,
0:00:28 we’re actually resurfacing
0:00:30 some of our previous episodes
0:00:33 featuring founders, Mark Andreessen and Ben Horwitz.
0:00:34 If you haven’t heard the latest episode
0:00:36 with Stuart Butterfield turning the tables
0:00:38 as the entrepreneur interviewing them,
0:00:41 please do check that out and other episodes
0:00:45 in this series on our website at a6andz.com/10.
0:00:48 But this episode was recorded in 2016
0:00:50 on the five-year anniversary
0:00:52 of Mark’s Wall Street Journal op-ed
0:00:55 on why software is eating the world.
0:00:57 And it features me and Scott Cooper
0:01:00 asking Mark and Ben about what’s changed since
0:01:03 and how software is now programming the world.
0:01:05 And we discuss everything from simulations
0:01:09 to distributed systems to other key computing shifts.
0:01:10 Welcome guys.
0:01:11 – Hey, thank you.
0:01:13 – Okay, so let’s just kick things off.
0:01:15 One of the things that I want to understand
0:01:16 is that it’s been since fund one,
0:01:18 which is what, seven years ago?
0:01:20 Yeah, seven years ago.
0:01:21 A lot’s changed in seven years.
0:01:22 And I’ve actually heard you argue, Mark,
0:01:24 that things have accelerated in that time period,
0:01:27 more so than previous decades before.
0:01:29 So what do you think are the biggest shifts now
0:01:31 that are important to us in this newest fund
0:01:34 and what changed in that period, like the biggest things?
0:01:37 – So in fund one, when we started,
0:01:39 we thought that our timing was really good,
0:01:42 despite the fact that I think the world thought
0:01:43 our timing was really bad
0:01:45 and starting a new venture capital fund.
0:01:47 And the reason why we thought that was that
0:01:50 there were three gigantic new platforms
0:01:51 hitting all at the same time,
0:01:56 which was kind of unprecedented in the history of technology.
0:01:58 One was mobile, the second was social,
0:02:00 and the third was cloud.
0:02:02 And that really proved out
0:02:04 through the course of the early history
0:02:06 that the applications on top of those,
0:02:11 particularly mobile and cloud were just spectacular.
0:02:13 And I think we’re coming a little bit
0:02:17 to the end of the first phase of the,
0:02:19 some of the obvious applications
0:02:20 that could be built on those things
0:02:22 and we’re moving into some new areas.
0:02:25 – Yeah, so let me go to the foundations.
0:02:26 So there’s different ways of looking at it.
0:02:27 The foundation levels.
0:02:30 One is Moore’s law has really flipped.
0:02:31 And this actually has happened.
0:02:32 I think this actually has happened
0:02:33 over the last seven or eight years,
0:02:35 actually almost exactly over the life of the fund,
0:02:37 which is for many, many years,
0:02:40 Moore’s law was a process of the chip industry
0:02:42 bringing out a new chip every year and a half
0:02:45 that was twice as fast as the last one at the same price.
0:02:47 And that continued for 40, 50 years.
0:02:49 And that’s, by the way, what resulted in everything
0:02:51 from mainframes, mini computers, PCs,
0:02:55 and then smartphones, about seven, eight, nine,
0:02:57 10 years ago, that process actually started to come in
0:02:59 and the way that it had worked up until then.
0:03:00 So chips have kind of topped out
0:03:02 at a speed of about three gigahertz.
0:03:04 And a lot of people have said, therefore,
0:03:06 like progress in the tech industry is gonna stall out
0:03:07 because the chips aren’t getting faster.
0:03:09 I think what’s actually happened is Moore’s law
0:03:10 has now flipped and the dynamic now,
0:03:12 instead of increased performance has reduced cost.
0:03:15 You now have this dynamic where every year, year and a half
0:03:16 the chip companies come out with a chip
0:03:18 that’s just as fast, but half the price.
0:03:21 And so this is this sort of, just this massive deflationary
0:03:23 force, I think in the technology world.
0:03:27 And I actually also suspect in the economy more broadly,
0:03:28 we’re basically computing is just becoming free.
0:03:31 Basically what we do in this business is we just kind of
0:03:33 chart out the graphs and then just kind of assume
0:03:34 at some point you’re gonna get to the end state
0:03:36 and the end state is gonna be the chips are gonna be free,
0:03:38 which means chips will be embedded in everything.
0:03:40 You’ll be able to use chips for literally everything.
0:03:42 And we’ve never lived in a world before where you can do that.
0:03:44 So that’s the first one.
0:03:46 Second one is just the obvious implication from that,
0:03:48 which is all those chips will be on the network, right?
0:03:50 So all those chips will be connected to the internet.
0:03:52 They’ll all be on wifi or mobile carrier networks
0:03:53 or wired networks or whatever,
0:03:55 but they’ll all fundamentally be on the internet.
0:03:57 You know, that’s something that’s not happening
0:03:59 at a very rapid pace.
0:04:01 And then the third is the continuation of the piece
0:04:03 that I wrote actually five years ago,
0:04:04 which was called software eats the world,
0:04:06 which is basically just say if you’re gonna live in a world
0:04:08 in which there’s gonna be a chip in every physical object.
0:04:11 And if you live in a world in which every physical object
0:04:12 therefore is going to be networked,
0:04:14 it’s gonna be smart because it has a chip
0:04:15 and it’s gonna be connected to the network.
0:04:17 Then basically you can then program the world.
0:04:19 You can basically write software
0:04:20 that applies to the entire world.
0:04:22 So you can write software that all of a sudden applies
0:04:24 to all cars or you can write software that applies
0:04:27 to all, you know, everything flying in the sky.
0:04:29 You can write software that applies to all buildings.
0:04:31 So you can write software that applies to, you know,
0:04:35 all homes or all businesses or whatever, all factories.
0:04:36 And so all of a sudden you can kind of,
0:04:37 you can program the world.
0:04:39 That’s really just starting.
0:04:40 And I think a lot of the,
0:04:42 there’s a number of things that make the entrepreneurs
0:04:44 we’re seeing these days in many ways more interesting
0:04:45 and more aggressive than entrepreneurs
0:04:47 who’ve seen the past and part of it is they just assume,
0:04:49 if there’s something to be done in the world,
0:04:51 there must be a way to write software to be able to do it.
0:04:54 That’s at a new level of power sophistication.
0:04:57 It’s a new scope of what the tech industry can do.
0:04:59 The consequence of that for us as a fund
0:05:01 is that we find ourselves evaluating business plans
0:05:03 and funding companies that are in markets
0:05:04 where I think seven or eight years ago
0:05:06 we would have never anticipated operating.
0:05:09 – So Mark, does that mean that there’s no new innovation
0:05:11 in platforms themselves and everything?
0:05:12 All the innovation will be applications
0:05:14 that ride on that existing infrastructure?
0:05:16 Or do you think there’s also the opportunity
0:05:19 to build a new platform, even given some of those trends?
0:05:20 – I think there are new platforms
0:05:21 and I think there will be new platforms.
0:05:23 I just think there’ll be different kinds of platforms
0:05:24 than we’ve had in the past.
0:05:26 The idea of a platform in the tech industry, as you know,
0:05:29 up until five or 10 years ago was there is a new chip
0:05:31 that has new capabilities is faster
0:05:33 and then therefore you build a new operating system for it.
0:05:36 And that might be Windows or it might be iOS
0:05:37 or whatever it is.
0:05:39 The platforms that we’re seeing getting built these days
0:05:42 are distributed systems, so scale out systems.
0:05:44 Instead of being built on a chip necessarily
0:05:45 with new unique capabilities,
0:05:47 they are platforms that are going to build
0:05:48 across lots of chips.
0:05:49 And so they’re in computer science terms,
0:05:51 they’re distributed systems.
0:05:52 Cloud is one of the first examples, right?
0:05:55 So anybody who uses AWS can now go on
0:05:57 and can program an application on AWS
0:05:59 that will run across 20,000 computers
0:06:02 and they can run it for an hour and it’ll cost 50 bucks.
0:06:06 And that’s a kind of platform that did not exist before.
0:06:08 And by the way, there are many specific elements to that.
0:06:10 So for example, we’ve seen the rise of,
0:06:11 in that category, seen the rise of Doop
0:06:14 and otherwise the spark for distributed data processing.
0:06:16 We’ve seen in financial technology,
0:06:18 we’ve seen the rise of Bitcoin and cryptocurrency,
0:06:20 which is a literally distributed platform for currency
0:06:22 and for exchanging value.
0:06:25 And now we’re seeing the emergence of a major new platform,
0:06:28 which is AI, machine learning and deep learning,
0:06:30 which is inherently, the great thing about machine learning
0:06:33 and deep learning is they’re inherently parallelizable.
0:06:34 They can run across many chips
0:06:36 and they get very powerful as you do that.
0:06:40 And you can do things in AI today as a consequence
0:06:41 of being able to run across many chips
0:06:43 that you just couldn’t even envision
0:06:44 doing five or 10 years ago.
0:06:46 – So let’s talk about the rise of the GPU
0:06:47 as part of this next platform ship.
0:06:49 I mean, I think the biggest surprise people have had
0:06:51 is that this is the graphical processor unit,
0:06:54 which is something that was developed in the gaming industry
0:06:57 for really high resolution graphics processing
0:06:59 and is now finding, I guess, unexpected.
0:07:02 Is it a surprise to us that it’s finding uses
0:07:05 in these new platforms like VR, AR, deep learning?
0:07:06 – It’s actually interestingly,
0:07:08 it’s a new application of an old idea
0:07:09 back when I was getting started 30 years ago
0:07:11 working in physics labs.
0:07:13 If you wanted to run just a normal program,
0:07:16 you just buy a normal computer and run the program.
0:07:19 But if you wanted to do, run a program,
0:07:21 many physics simulations had this property
0:07:24 where you would want to run a very large number
0:07:25 of calculations in parallel, right?
0:07:27 And so you could basically divide up a problem
0:07:29 of simulating anything from a black hole
0:07:32 or to different kinds of biological simulations.
0:07:34 You could basically write these algorithms
0:07:35 in a way that they could run,
0:07:36 you could basically parcel the problem
0:07:38 into many different pieces
0:07:39 and then run them all in parallel.
0:07:41 There was actually in the old days,
0:07:42 there was actually a whole industry
0:07:43 of what we’re called vector processors,
0:07:46 which were literally these kind of sidecar computers
0:07:48 that you would buy and you would hook up to your main computer
0:07:49 and they would let you run these parallel problems
0:07:51 much faster.
0:07:53 And so literally 30 years later, the GPU is a,
0:07:54 it’s basically a vector processor.
0:07:56 It’s basically a sidecar processor that sits along the CPU
0:07:58 and runs these parallel problems much faster.
0:08:01 And graphics are a natural application of that,
0:08:03 but as it turns out, graphics aren’t the only application.
0:08:05 – Yeah, actually interestingly,
0:08:07 and I was at a company making one of these
0:08:11 called Silicon Graphics and the applications then,
0:08:13 whereas Mark was saying a lot of physics applications,
0:08:16 computational fluid dynamics and simulating,
0:08:18 flight simulation and all these kinds of things
0:08:20 that are hard physics to calculate.
0:08:21 When you go into the virtual world
0:08:23 and you’re simulating the physics of the real world,
0:08:24 guess what?
0:08:27 You need the exact same processor to do it.
0:08:31 So it’s a super logical conclusion to what’s been going on,
0:08:35 but I think we’re also in the world of big data,
0:08:39 seeing kind of more reasons to do just lots of math in parallel.
0:08:42 And so it’s an exciting application.
0:08:43 – Yeah, you talk about platforms.
0:08:45 One of the really interesting hardware platforms
0:08:47 is emerging right now is NVIDIA,
0:08:49 which is a very well-established public chip company,
0:08:50 but very successful to your point,
0:08:53 doing graphics chips for a very long time,
0:08:55 has become seemingly overnight.
0:08:57 It’s really, of course, the result of years of work,
0:08:59 but seemingly overnight has become the market leader
0:09:03 in both not just GPUs, but also in chips being used for AI.
0:09:05 And it’s basically extensions of the GPU technology.
0:09:08 And we see this overriding theme,
0:09:09 which is kind of an amazing thing,
0:09:12 which is basically every sharp AI software entrepreneur
0:09:15 that comes in here is now building on top of NVIDIA’s chips,
0:09:17 which is, of course, a very different outcome
0:09:18 than entrepreneurs of previous years
0:09:20 who would have built other kinds of programs
0:09:22 primarily on top of Intel chips.
0:09:24 – We’ve mentioned AI and machine learning a couple of times here.
0:09:25 And one of the interesting things,
0:09:28 at least I think we see in the industry is,
0:09:29 at the same time we’ve got startups doing it,
0:09:32 we also see some of the very large established players
0:09:34 investing significantly in AI and machine learning.
0:09:37 So certainly Facebook and Google, Apple and others
0:09:39 are obviously building big operations.
0:09:41 How do you think about the universe
0:09:42 from an investment perspective?
0:09:44 What are the kinds of things that actually led themselves well
0:09:47 to startup opportunities in the AI space
0:09:49 versus things that actually might make sense,
0:09:52 kind of living inside of one of the larger companies
0:09:53 like a Facebook or Google?
0:09:55 – Yeah, so AI is extremely broad.
0:09:58 And I think one of the challenges that people have with it
0:10:01 is they try to paint it as a narrower thing than it is,
0:10:04 but one can think of it as an entirely new way
0:10:07 to write a computer program.
0:10:12 And so then it’s applicable to the universe of problems.
0:10:15 So there are things that advantage a big company.
0:10:18 If you’re building AI to analyze consumer internet data,
0:10:20 like that’s hard to take Google on at that.
0:10:22 They do have an awful lot of data.
0:10:26 And Facebook with AI, computing power matters
0:10:28 and the data set matters.
0:10:30 Having said that, there are a lot of areas
0:10:32 where nobody has any data yet
0:10:37 in the areas of healthcare and the areas of autonomy.
0:10:40 So there’s lots and lots of opportunities.
0:10:44 And there’s also interesting ideas
0:10:46 about, well, is there a better user interface
0:10:49 than the smartphone using AI techniques
0:10:50 and then what is the form of that?
0:10:51 – What do you mean by that
0:10:53 when you say there’s a better user interface?
0:10:55 – Well, if you think about a smartphone,
0:10:58 it was kind of an advance over what we used to call
0:11:03 the WIMP interface, Windows icons.
0:11:05 What was it?
0:11:06 – Menus.
0:11:07 – Menus.
0:11:07 – What was it, P?
0:11:08 – Pointer.
0:11:09 – Pointer, right.
0:11:12 – Which was like a big advance
0:11:15 over the text-based interface of DOS.
0:11:18 And then the smartphone with a touch interface,
0:11:20 so it was more of a direct manipulation
0:11:21 was an advance over that.
0:11:23 And so you’d go, okay, well,
0:11:27 but that’s not actually what people do in life, right?
0:11:30 Like it’s anthropologically,
0:11:34 it’s a backward step in terms of the natural interface
0:11:36 so that we’ve become accustomed to,
0:11:38 like for example, natural language.
0:11:40 With AI, you get into a world
0:11:44 where things like natural language and natural gestures
0:11:46 and so forth become much more plausible.
0:11:48 So there’s potentially an opportunity
0:11:52 to build interfaces for things that you couldn’t before.
0:11:54 I mean, I think there’s one really interesting thing,
0:11:57 which I’m sure, and I know that Google and Apple
0:11:59 and all the giant companies are very focused on,
0:12:02 which is how do you replace the current set
0:12:03 of user interfaces with it?
0:12:05 But there’s another dimension,
0:12:08 which is what are all the applications
0:12:10 that you just couldn’t have before
0:12:13 because you couldn’t build a workable user interface for it.
0:12:17 And AI seems very promising in those areas.
0:12:18 – You didn’t mention Amazon,
0:12:19 which is sort of the stealth player here
0:12:21 with Echo and Alexa.
0:12:23 I mean, really, George and Horace have been home.
0:12:26 Well, in a way, they’ve got an interesting advantage
0:12:28 in that they’re not tight
0:12:31 to the last generation of user interfaces
0:12:34 so that they don’t have to pay the strategy tax
0:12:39 for shoehorning in their AI into, say, the iPhone.
0:12:40 And that’s something.
0:12:41 – Yeah, that’s worth pointing out.
0:12:43 There’s sort of two kind of classic rules of thumb
0:12:44 in this industry.
0:12:46 One is for major new advances,
0:12:47 especially in things like interfaces,
0:12:49 if you don’t own a platform, you can’t do them.
0:12:52 And so the assumption I think had been up until recently,
0:12:53 that it would have to be Google or Apple
0:12:54 that does these kind of natural language
0:12:57 or interface advances ’cause they own iOS and Android.
0:13:00 The other rule, of course, is the exact opposite rule,
0:13:01 which is the one that Ben mentioned,
0:13:03 which is the problem that big established companies
0:13:06 get into is this, what he referred to as the strategy tax,
0:13:09 which is basically big companies with existing agendas
0:13:11 have to sort of fit their next thing
0:13:12 into their existing agenda,
0:13:14 and they often compromise it in the process.
0:13:16 And so it’s sort of this ironic twist of fate
0:13:18 that Amazon has all of a sudden taken the lead
0:13:19 from Google and Apple.
0:13:22 Even though Amazon famously flopped with their phone,
0:13:23 which is sort of the obvious place
0:13:24 where you would have a voice interface,
0:13:26 it didn’t matter because they came out with this new product,
0:13:29 which is basically the smart speaker called Echo.
0:13:32 And the fact that all of a sudden Amazon didn’t have a phone,
0:13:33 all of a sudden became an advantage
0:13:35 ’cause they could just do the clean,
0:13:36 actual breakthrough product
0:13:38 without worrying about tying it into the existing strategy.
0:13:40 – Right, and those are all still big companies, though,
0:13:42 is I’m not really hearing where startups
0:13:44 can really play in this space,
0:13:46 especially when you’re describing this huge data network effect
0:13:48 that all these big companies have.
0:13:50 – A year ago, we would have probably been sitting here
0:13:52 and say that AI was gonna be likely,
0:13:53 would be a domain of big companies
0:13:55 because of this sort of thing of like,
0:13:57 okay, only big companies can afford
0:13:58 the very large number of engineers
0:13:59 that are required to do AI.
0:14:01 Only big companies can afford the amount of hardware
0:14:03 required to do AI.
0:14:03 And then only big companies
0:14:06 can get the giant data sets required to do AI.
0:14:07 In the last 12 months, what we’ve seen basically
0:14:10 is all three of those changing very fast
0:14:11 into the advantage of startups.
0:14:14 We’ve seen a lot of AI technologies actually,
0:14:16 actually now interestingly standardizing.
0:14:18 So going to open source and then the next step
0:14:20 is gonna be they’re gonna go to cloud.
0:14:22 And we think we’re right on the verge of that.
0:14:23 We think all the major cloud providers
0:14:25 are gonna be providing AI as a service
0:14:27 and they’re gonna really radically reduce
0:14:29 the amount of technical knowledge you need to apply AI.
0:14:30 And so that plays very well to the startups.
0:14:32 – So there will be like an AWS for AI.
0:14:33 – Yeah, exactly.
0:14:34 And that may be literally AWS
0:14:35 or it may be Google or Microsoft
0:14:38 or all three of them in some combination
0:14:40 or it may be other companies yet to emerge.
0:14:42 – An example of the open source like TensorFlow,
0:14:43 like Google releasing TensorFlow.
0:14:44 – Yeah, and this is a big deal.
0:14:45 And of course, yeah, it’s right.
0:14:46 So Google open source to pretty significant part
0:14:47 of how they do deep learning.
0:14:48 And that actually now is something
0:14:50 that other companies can pick up and use directly.
0:14:51 And we see actually a lot of,
0:14:52 not only a lot of companies,
0:14:53 but like a lot of university,
0:14:54 a lot of student projects now
0:14:56 just kind of can pick that up and run with it.
0:14:59 So this technology is kind of trickling down very fast.
0:15:01 – Just this past weekend, we had a hackathon.
0:15:03 And I think most of the teams
0:15:06 had some machine learning AI component into their hacks.
0:15:07 And these are college kids.
0:15:11 – Yeah, if you’re a 21 year old junior in college
0:15:12 and you’re doing some project,
0:15:16 it’s just kind of, it’s becoming rapidly becoming very obvious
0:15:17 that you would have AI be part of it,
0:15:19 which was very much not the case even full months ago.
0:15:20 And that’s a direct to your point.
0:15:22 That’s a direct consequence of the open sourcing
0:15:24 and kind of this knowledge spreading out.
0:15:25 The second thing was the hardware costs.
0:15:27 And there again, the cloud, AI and the cloud,
0:15:29 just the existence of the cloud
0:15:30 is bringing down hardware costs across the board,
0:15:32 but AI and the cloud is gonna bring that down even further.
0:15:34 And by the way, these trends all slam together.
0:15:37 So you get what I think in a year is gonna be very common.
0:15:39 These sort of AI supercomputing chips
0:15:41 with the AI algorithms in the cloud
0:15:43 available to anybody for a dollar, right?
0:15:44 And so there’s gonna be this massive deflation
0:15:47 of hardware cost on that side.
0:15:48 These big data sets are interesting.
0:15:50 Ben made the case that the startups
0:15:51 can assemble big data sets.
0:15:53 And I think that there are certainly examples of that.
0:15:55 We also see another thing happening,
0:15:57 which is the newest generation of experts in deep learning
0:15:59 or many of them are specializing in the idea
0:16:01 of deep learning applied against small data sets.
0:16:03 If you talk to those folks, what they’ll tell you is,
0:16:06 oh, basically, they’ll basically say is
0:16:08 primitive and crude deep learning required big data sets,
0:16:09 but the really good stuff doesn’t.
0:16:11 It is small data sets are fine.
0:16:13 And so that’s still very early,
0:16:14 but it’s extremely enticing.
0:16:16 It’s an extremely enticing idea
0:16:18 because it really brings a lot of these problems
0:16:21 to your point further into being tractable
0:16:22 for small companies.
0:16:24 But actually, one of the things you can do
0:16:25 with these, especially with these GPUs
0:16:27 is you can literally use the same tools
0:16:29 that are used to make video games.
0:16:32 And you can create simulated versions of the real world.
0:16:34 And then you can actually let the AI train
0:16:35 inside the simulation.
0:16:36 And so if you’re building a new self-driving car
0:16:38 or a drone or something like that,
0:16:39 you can actually create simulated worlds
0:16:42 in which there are everything from earthquakes,
0:16:46 to floods, to thunderstorms, hail storms.
0:16:48 You can create birds, swarms of birds.
0:16:51 You can literally simulate the real world environment.
0:16:53 And then you can let the AI actually train inside that world.
0:16:54 And actually, it’s funny,
0:16:56 the AI actually has no idea it’s training
0:16:57 in the virtual world.
0:16:58 It’s learning just the same
0:16:59 as if it were learning in the physical world.
0:17:02 And so again, for startups with access to cloud-based AI,
0:17:05 you could potentially run basically millions of hours
0:17:07 of simulated training at a very low cost
0:17:08 and all of a sudden catch up to big companies.
0:17:12 – Interestingly, the very famous AI project
0:17:14 that Google did with DeepMind,
0:17:17 that whole dataset came from the game “Playing Itself”.
0:17:19 So, there wasn’t some dataset
0:17:21 that Google had collected over 20 years.
0:17:24 It was the game “Playing Itself”.
0:17:26 – So you guys have both mentioned simulations a few times.
0:17:27 Why are they so important?
0:17:28 ‘Cause I feel like there was this period,
0:17:29 like maybe even a decade ago
0:17:31 where simulations were almost frowned upon
0:17:33 as this promised thing
0:17:35 that didn’t really actually deliver
0:17:37 in what you needed to be able
0:17:40 to navigate complex environments in real life.
0:17:41 – Yeah, well, it’s interesting.
0:17:45 So was AI, was frowned upon 10 years ago,
0:17:48 saying it was all, it didn’t work.
0:17:50 And particularly neural nets and deep learning
0:17:53 were the most frowned upon area.
0:17:55 And there’s been similar kind of breakthroughs
0:17:56 with simulation.
0:17:58 And first of all, so if you think about the field
0:18:02 of data science and what you do with data,
0:18:04 there’s you have a giant set of data,
0:18:07 which is always historical in nature,
0:18:08 and you can analyze that.
0:18:10 And maybe it’s predictive of the future,
0:18:12 but oftentimes it’s not.
0:18:15 And we see this in particular in things like,
0:18:17 really dynamic things where the past affects the future,
0:18:20 like say stockpicking or the weather
0:18:23 or other kinds of things where data analysis
0:18:24 doesn’t get you an accurate answer.
0:18:26 Simulation is a flip side of that
0:18:28 where you can say, okay, here are all the entities
0:18:32 in the world and let’s generate their behavior over time.
0:18:34 And then their actual behavior feeds back
0:18:37 into the simulation, which is critical,
0:18:39 a critical component.
0:18:42 Historically, that’s been difficult at scale,
0:18:45 but there have been some really important breakthroughs
0:18:47 lately, particularly from a company
0:18:49 that we’re invested in called Improbable,
0:18:52 which is able to do very large scale,
0:18:55 scale out simulation using cloud computing techniques
0:18:59 and some very important new technology
0:19:00 that they’ve developed.
0:19:03 And so you can get a really complete picture of the world.
0:19:06 And as Mark was saying, you can actually generate
0:19:09 your own dataset rather than collecting it
0:19:10 for certain kinds of situations.
0:19:12 Yeah, let me add the thing to that.
0:19:14 So one way to think about it is it’s expensive
0:19:15 to make things happen in the real world.
0:19:17 Like it’s expensive to change things in the real world
0:19:18 because the real world is physical
0:19:20 and causing physical changes to happen.
0:19:21 I mean, everything from building roads to flying planes,
0:19:23 all these things are very expensive.
0:19:24 And then things in the real world,
0:19:26 changes have serious consequences, right?
0:19:28 And so if you, depending on where you put the dam
0:19:29 or where you put the airport
0:19:31 or what your evacuation plan you have for the city
0:19:32 and if something bad happens,
0:19:35 like these decisions have huge consequences.
0:19:36 Which banks you bail out.
0:19:38 Which banks you bail out.
0:19:39 Which banks you don’t bail out.
0:19:41 And so you always have these consequences
0:19:43 and people who have to make these decisions
0:19:44 are often flying blind
0:19:45 ’cause they don’t have any real sense
0:19:46 of what’s gonna happen
0:19:47 as a consequence of their decisions.
0:19:49 In contrast, if you can simulate a world
0:19:50 and if you can run an experiment,
0:19:53 if you can simulate the real world or some portion of it
0:19:55 like the highway system or the banking system or whatever.
0:19:58 And then you can basically introduce change
0:19:59 into that simulation
0:20:00 and you can see what the consequences are.
0:20:02 It’s very cheap to do that because Moore’s Law
0:20:04 and the collapse of chips and the rise of cloud computing
0:20:06 and all these other things we’ve been talking about
0:20:08 all of a sudden make it very cheap to run the simulations.
0:20:10 It’s much cheaper to do it in the simulated world.
0:20:11 And then there are no consequences.
0:20:13 You run a simulation and everything goes wrong
0:20:15 and everybody dies
0:20:16 or the entire financial system collapses or whatever.
0:20:17 It doesn’t matter.
0:20:19 You just erase it and you run it again.
0:20:20 – Yeah, you have infinite testability.
0:20:22 – Well, I went and challenged that.
0:20:25 There is Elon Musk simulation in which case
0:20:27 the consequences are quite dire for us.
0:20:27 – There is, yes.
0:20:29 There is a scenario that we’re all living in a simulation.
0:20:30 – Right, if we’re living in one place.
0:20:32 – In which case I would argue it’s gone badly awry
0:20:34 as evidenced by the current political situation.
0:20:36 – There is no do over button in this simulation.
0:20:38 – Yes, and then you basically, again,
0:20:40 you look at the progress of Moore’s Law
0:20:41 and the rise of these new technologies
0:20:43 and you say, okay, how about instead of running
0:20:44 one simulation, let’s run a million simulations
0:20:46 or let’s run a billion simulations
0:20:47 and let’s try every conceivable thing
0:20:49 we can possibly think of and let’s imagine,
0:20:51 let’s literally model all potential future states
0:20:54 of the world and then let’s decide which one of those,
0:20:57 which path is the one that leads to the best consequences.
0:21:00 And so we can then make these very big real world decisions
0:21:04 with a lot more foreknowledge of what will unfold afterwards.
0:21:06 – Maybe just to get concrete on some opportunities,
0:21:08 what are the other areas and maybe it’s life sciences
0:21:10 or what are some of the other kind of more tangible areas
0:21:11 that you think in your term,
0:21:13 as you think about kind of deploying this fund
0:21:15 or beyond over the next, you know, five to 10 years
0:21:17 that might be interesting for, you know, people to think
0:21:19 about in the context of real world applications
0:21:20 of this technology.
0:21:21 – Yeah, so as Mark was saying,
0:21:24 we’re coming into this era of new platforms
0:21:27 and with the intersection of health and computer science,
0:21:30 what we’re saying is really exciting new platforms
0:21:34 around data and around basically you being able
0:21:36 to get much more information about someone’s health
0:21:40 from a variety of techniques that have been developed,
0:21:43 you know, based on the kind of historic breakthroughs
0:21:47 and sequencing the genome, but beyond that as well,
0:21:51 where we can get really, really powerful data about people
0:21:54 and understand them better.
0:21:55 And once you have that data about people
0:21:58 when you can be predictive of diseases
0:22:00 that they might get or things that are wrong
0:22:02 and you aggregate that into a platform,
0:22:05 then you can actually make new scientific discovery
0:22:06 off it as well.
0:22:08 So that’s one interesting area.
0:22:11 There’s, if you think about the AI platform itself,
0:22:14 one of the things about it is the hardware
0:22:17 that’s been built for it or that’s been built historically
0:22:20 is for a completely different kind of computer programming.
0:22:23 And we’ve seen Google already announced a chip
0:22:25 to power their deep learning cloud.
0:22:28 And, you know, similarly there’s new breakthroughs
0:22:31 and quantum computing, which at least on the surface
0:22:33 look like they may be very promising
0:22:37 for much more powerful deep learning systems and so forth.
0:22:39 So there’s a lot of things
0:22:41 that are coming out of these platforms.
0:22:43 And then, you know, as we get to a chip and everything,
0:22:47 the platforms to run and manage and understand those chips
0:22:50 are equally as exciting.
0:22:52 – So, you know, one of the themes that’s come up through here
0:22:55 is that tech is reaching into places it never did before.
0:22:57 I mean, every company is becoming a tech company
0:23:00 or they have tech inside, or as Benedict likes to say,
0:23:01 growing the tech industry,
0:23:04 the reality is it’s permeating everywhere.
0:23:06 And the question I have for us
0:23:08 is that we are founded on this thesis
0:23:10 that software is eating the world, that’s our premise.
0:23:12 And yet we’ve seemed to have been making
0:23:14 a lot of hard investments, you know,
0:23:18 if you count things like Soylent, Oculus, Nutribox.
0:23:21 So are we changing our thesis about hardware
0:23:23 as a result of this software eating the world?
0:23:24 – No, I don’t think so.
0:23:26 I mean, I think that what we see
0:23:29 with the companies that you’ve named are interesting.
0:23:31 So, Oculus, I think we would all agree
0:23:33 that the software component of Oculus
0:23:37 is both more complex, has many more people working on it
0:23:39 and is kind of the core of the investment.
0:23:42 Sometimes if you have a breakthrough technology,
0:23:45 then you require a new hardware to actually support it.
0:23:46 And that’s the case there.
0:23:49 And I think that Soylent and Nutribox,
0:23:51 both of them apply computer science techniques
0:23:54 and information technology to get people to optimal health.
0:23:56 And that’s what we’re doing there.
0:24:00 So I think we’re big, big believers that, you know,
0:24:03 in the last 100 years, the great breakthroughs
0:24:05 and knowledge have been the breakthroughs
0:24:08 of people like Alan Turing and Claude Shannon
0:24:10 who gave us a new model of the world
0:24:12 and how to understand it.
0:24:16 And companies that build on that fundamental knowledge
0:24:18 breakthrough are what we’re about
0:24:20 and will continue to be about that.
0:24:22 – Even if some of them may ship their products in a box.
0:24:25 – Yes, the package is not a technology.
0:24:27 Let’s talk a little bit about SaaS.
0:24:27 As you’ve probably seen,
0:24:29 there’s been actually a bunch of acquisitions
0:24:31 in this space recently, but what’s left to do there?
0:24:34 So is the new platform, the salesforce.com
0:24:36 and others of the world or are there actually
0:24:37 both kind of vertical applications
0:24:39 and or are there other platforms
0:24:41 that actually might exist over time in that market?
0:24:43 – So there’s SaaS as the metaphorical
0:24:45 in the cloud version of all the stuff
0:24:49 that we had built over the previous, you know, 30, 40 years.
0:24:54 So that’s like workday, salesforce,
0:24:58 success factors, you know, the kind of big categories.
0:25:00 The thing that we believe that’s changed
0:25:02 as you go from on premise to the cloud
0:25:05 is the technology is so much easier to adopt
0:25:08 that we’re now seeing software applications
0:25:10 for things that you just would never do
0:25:13 as a software application because the cost of,
0:25:15 as we used to say in the old days, screwing it in
0:25:19 and paying the army of Accenture consultants
0:25:21 to get it going just wasn’t worth it
0:25:24 for say expense reporting, which, you know,
0:25:28 Concur, of course, built a really powerful product in that.
0:25:31 But like there was no packaged software
0:25:34 for expense reporting in the same way that there is now.
0:25:37 And I think there’s a gigantic number of categories
0:25:39 in everything that you do in business
0:25:41 that can be automated in that way.
0:25:43 In addition to that, you can scale down
0:25:45 to very, very small companies.
0:25:47 Companies below thousands of employees
0:25:49 never bought Oracle financials.
0:25:51 It would have been insane to do so,
0:25:53 but they’re absolutely buying, you know,
0:25:55 NetSuite and things like that.
0:25:59 And then beyond that, you now it becomes economical
0:26:02 and very interesting to build vertical applications
0:26:03 for industries.
0:26:06 So to build an application that revolutionizes,
0:26:09 say the real estate industry or something like that,
0:26:13 or the construction industry is becoming extremely viable
0:26:15 and not just as a niche business,
0:26:18 but as a real venture capital based kind of activity.
0:26:20 – One of the consequences that will be interesting
0:26:22 to watch play out is that historically,
0:26:24 enterprise software has been described
0:26:27 as represented by companies like Oracle SAP IBM.
0:26:29 Like that stuff was really only accessible
0:26:31 to the largest companies,
0:26:33 the top 500 or 1000 companies in a country.
0:26:36 And then in particular, only in a handful of countries,
0:26:39 those businesses, their revenue and their customer base
0:26:40 have always been dominated by, you know,
0:26:42 two or 3000 companies globally that are these,
0:26:44 you know, these giant multinational companies
0:26:45 that we’ve all heard of.
0:26:48 So big companies had this sort of inherent advantage
0:26:50 versus a lot of mid-sized small companies.
0:26:51 And then companies in the US and Western Europe
0:26:53 had this big advantage versus companies
0:26:54 in other parts of the world
0:26:56 where the companies, the large companies
0:26:58 and the large companies in the US and Western Europe
0:27:00 could just afford to make technology investments
0:27:01 that small and mid-sized companies
0:27:02 all over the world couldn’t make.
0:27:05 The sort of changes in SaaS that Ben described,
0:27:07 they lead you to an interesting conclusion
0:27:08 which is it may actually be interesting
0:27:12 for a smaller company or a company not in the US
0:27:13 or Western Europe to be able to adopt
0:27:16 the next generation of SaaS and cloud technology.
0:27:18 It’s almost like the folks who’ve been able to skip
0:27:19 landline telephones or just go straight to mobile phones.
0:27:21 You can just leapfrog the old stuff
0:27:22 ’cause you never had it
0:27:24 and you can just start using the new stuff out of the box.
0:27:26 And then the big established companies
0:27:27 might have a harder time adapting
0:27:29 ’cause they’ve made these giant investments
0:27:30 in the old systems and it’s hard to just jump
0:27:31 to the new thing.
0:27:34 And so there may be a power shift happening
0:27:36 from on the one hand large companies
0:27:37 to small and medium companies
0:27:40 that can now more aggressively adopt technology faster.
0:27:42 And then from companies in the US and Western Europe
0:27:43 to companies all over the world
0:27:45 that can also do the exact same thing.
0:27:48 And so at the very least a leveling of the playing field
0:27:49 and possibly even a national shift in balance
0:27:51 where small and mid-sized companies all over the world
0:27:53 may all of a sudden get a lot more competitive.
0:27:55 – So you’ve got kind of democratization at one point
0:27:56 and then to your point,
0:27:58 there’s one version of internationalization
0:28:00 which is adoption across international communities.
0:28:02 So how do you think about then
0:28:03 the other aspect of internationalization
0:28:04 which is company formation?
0:28:07 Should we then expect to see more new company formation
0:28:10 outside the US partly as a result of some of these trends
0:28:14 and why won’t we see or will we see 50 Silicon Valley’s
0:28:15 over the next 20, 30, 40 years
0:28:17 and how do you all think about what the strategy
0:28:19 should be vis-a-vis those opportunities?
0:28:21 – That would be probably the most amazing thing
0:28:23 for the world that could happen
0:28:25 in the realm of business and economics.
0:28:28 So we’re hoping for it
0:28:31 and certainly building kind of help trying
0:28:33 to build technologies that would facilitate it.
0:28:37 And I think the world has never been kind of more ripe
0:28:39 for that kind of thing.
0:28:40 Having said that, look,
0:28:42 there are real network effects,
0:28:44 geographical network effects
0:28:46 and Silicon Valley obviously has the biggest one
0:28:50 in technology and you always have to keep in mind
0:28:52 and this is something that gets lost
0:28:55 is there are no local technology companies, right?
0:28:58 There’s nobody who sells, you know,
0:29:01 internet search to Wyoming.
0:29:02 That’s not like a viable thing.
0:29:07 So when you’re competing globally, it does matter.
0:29:08 You know, do you have the best people?
0:29:09 Do you have the best executives?
0:29:10 Do you have the best engineers?
0:29:12 Do you have access to money?
0:29:15 Like all these things become real competitive things.
0:29:17 So we still are believers in Silicon Valley
0:29:20 and we’re very hopeful that the rest of the world grows
0:29:23 and that we can participate in that as well,
0:29:25 but that’s TBD.
0:29:27 – This is an interesting macro kind of thing
0:29:28 that’s happening in a lot of the, you know,
0:29:30 one of the really kind of negative stories
0:29:32 is that there’s basically the world is starved
0:29:34 for innovation and growth.
0:29:36 One of the data points you point to on that
0:29:39 is there’s now $10 trillion of money
0:29:41 in being held in government bonds,
0:29:42 governments all over the world,
0:29:44 trading at what’s called negative yield.
0:29:46 This is literally like the equivalent of a savings account
0:29:48 where instead of the bank paying you interest,
0:29:51 you have to pay the bank interest to hold your money.
0:29:53 And so there’s literally $10 trillion of capital parked
0:29:56 around the world that is actually losing money
0:29:58 as it sits there, which means people cannot find
0:30:00 enough productive places to deploy capital.
0:30:02 The conventional view, if you just pick up the newspaper
0:30:03 and read the economic section,
0:30:04 the horrible this is and how it means
0:30:06 the world is start for growth.
0:30:10 The optimistic side of it is there’s $10 trillion of money
0:30:12 sitting on the sidelines waiting for something productive
0:30:13 to be done with it.
0:30:15 What could be productively done with it?
0:30:17 New kinds of healthcare, new kinds of education,
0:30:20 new kinds of consumer products, new kinds of media,
0:30:21 new kinds of art, new kinds of science,
0:30:24 new kinds of self-driving cars, new kinds of housing,
0:30:27 all of these things that need to be done all over the world.
0:30:29 And so the world has never been more ripe
0:30:32 for a very large wave of innovation
0:30:34 that would actually be quite easy to finance.
0:30:36 A lot of the times you just can’t get things done
0:30:37 ’cause you don’t have money.
0:30:38 That’s just kind of the constant state of the world
0:30:39 for a very long time.
0:30:41 And now ironically, we live in a world
0:30:42 where the opposite is true.
0:30:44 There’s actually, quote unquote, too much money.
0:30:46 And more money than ideas.
0:30:46 More money than ideas.
0:30:48 Which really can’t be true.
0:30:49 It can’t be true, right?
0:30:50 Unlock the ideas.
0:30:52 Human creativity is boundless.
0:30:54 And so if you can get more smart people
0:30:55 around the world educated
0:30:57 and with the skills required to do these things
0:30:58 and if you can get them in environments,
0:31:00 either create new environments to do that
0:31:01 or figure out how to get more of the people
0:31:03 from other places in environments
0:31:04 where they can do new things,
0:31:06 we could do all kinds of new things globally.
0:31:08 And it’s something that we hope to contribute to
0:31:10 but I think is a very big opportunity for the world.
0:31:11 So do you think we’re getting to the point
0:31:13 where it’s kind of geopolitical risk
0:31:16 and rule of law issues that limit adoption
0:31:17 or deployment of some of these new technologies
0:31:19 in other countries outside the US?
0:31:23 It sounds like it’s less so technological advancement.
0:31:25 Well, I would say there’s bad news and good news.
0:31:27 So the bad news is we frequently have delegations
0:31:29 of folks coming into the Valley from all over the US
0:31:31 and all over the world.
0:31:32 And they basically come in
0:31:33 and it’s economic delegations of different kinds
0:31:35 or politicians or whatever.
0:31:36 And they come in and they’re like, okay,
0:31:38 what can we do to have our own Silicon Valley?
0:31:39 And then you kind of sit down with them
0:31:42 and you kind of go through ABCDEF, all these things.
0:31:44 Well, you don’t want rule of law,
0:31:46 you want ease of migration, you want ease of trade,
0:31:49 you want deep investments in scientific research,
0:31:51 you want no non-competes,
0:31:52 you want fluid labor laws to let companies
0:31:54 very easily both hire and fire.
0:31:55 You want the ability for entrepreneurs
0:31:57 to be able to start companies very quickly.
0:31:59 You want bankruptcy laws that make it very easy
0:32:01 to move on and start another company.
0:32:04 And at some point, the visitors get this stricken look
0:32:05 on their face and they’re like, well,
0:32:07 at the end of it, they’re like, okay,
0:32:08 but what if we want Silicon Valley
0:32:10 but we can’t do any of those things.
0:32:12 And so that’s the bad news.
0:32:14 – Well, they can hire Donald Trump to run their country.
0:32:15 – It’s sad that’s ironic that we have this guy running
0:32:17 for president who would seriously move us backwards
0:32:19 on a number of those topics.
0:32:21 So even we struggle with these things, right?
0:32:23 Like I would argue the formula is fairly well known.
0:32:25 It’s just people do not want to apply it
0:32:27 for reasons that have a lot to do with politics
0:32:29 and have a lot to do with other issues.
0:32:30 The good news is it can be done.
0:32:32 And then the other good news is it is happening.
0:32:35 And there are very, very, very exciting things happening
0:32:36 throughout much of the world.
0:32:38 They’re very active now startup scenes
0:32:41 all through South America, Brazil, Argentina, Buenos Aires.
0:32:42 Amazing things are happening in India.
0:32:44 There’s all kinds of startup activity
0:32:45 throughout the Middle East.
0:32:47 There’s startup activity now throughout Africa.
0:32:50 There’s, obviously China’s been a gigantic success story.
0:32:52 Korea has all kinds of interesting things happening.
0:32:55 So there are lots and lots of extremely positive
0:32:57 early indications of what’s possible
0:32:59 in many places all over the world.
0:33:00 That said, there are very big political questions
0:33:02 about whether or not those founders
0:33:03 are gonna be able to operate an environment
0:33:05 that’s really gonna let them succeed to the level
0:33:06 that they should be capable of doing.
0:33:08 The big reason that we raise the fund
0:33:12 and are excited about the fund is it is a backing
0:33:14 of our core belief system here,
0:33:18 which is we believe in the creativity
0:33:21 and genius and intelligence of human beings
0:33:23 and the entrepreneurs that we see
0:33:26 and come to Silicon Valley and around the world.
0:33:29 And we believe that these people absolutely have
0:33:32 the ability to change things and are changing things.
0:33:36 And there’s plenty of room to improve the world.
0:33:38 And there’s plenty of ideas to do so.
0:33:42 And that’s really what we’re about with fund five.
0:33:44 So let’s talk a little bit about kind of company building
0:33:45 and founders in particular.
0:33:49 So, undoubtedly you had a very distinct view
0:33:50 of what types of founders you wanted to back
0:33:53 when you started the firm now seven years ago.
0:33:54 How has that evolved?
0:33:56 If at all over time, what has changed either
0:33:57 in terms of the types of founders you see
0:33:59 or the types of qualities you see
0:34:01 that actually make founders successful
0:34:02 that’s caused you to either augment
0:34:05 or rethink some of the initial foundations for the firm.
0:34:08 You know, I think a lot of the things,
0:34:10 we had this great advantage when we started the firm
0:34:13 that we ourselves were founders.
0:34:16 I think that we’ve probably gotten,
0:34:18 I would say more risk tolerant
0:34:20 in our view of founders over time,
0:34:21 even though sometimes the risk–
0:34:22 – What do you mean by that?
0:34:24 What do you mean by getting risk more risk tolerant?
0:34:25 – Well, we have this thing we say at the firm,
0:34:27 which is we’re much more interested
0:34:29 in the magnitude of the strength
0:34:31 than the number of the weaknesses.
0:34:33 We always believe that intellectually,
0:34:36 I think that some of the number of weaknesses
0:34:39 were fairly terrifying early on,
0:34:43 just ’cause you do have a lot of founders
0:34:46 with a very small amount of experience these days,
0:34:47 which is also part of their strength
0:34:50 in that it’s hard to rewrite the world
0:34:52 if you’re too steeped in the world.
0:34:56 And so I think over time, we’ve kind of doubled down on that.
0:34:59 And really, the founders who have figured out
0:35:02 something really important or who are true geniuses
0:35:07 or have will to power that we can’t even contain in the room,
0:35:10 when they bring those things to the table,
0:35:11 whatever is wrong with them,
0:35:14 we tend to overlook and work with them on that.
0:35:16 And if they’re strong enough in those areas,
0:35:18 the really interesting thing for us has been,
0:35:21 those weaknesses do go away pretty quickly.
0:35:23 And that’s probably the biggest learning is,
0:35:25 I’d say we went in thinking that,
0:35:28 but we’ve gotten even more extreme
0:35:31 in our commitment to that kind of philosophy.
0:35:33 So almost in financial terms,
0:35:35 you’re buying volatility to a certain extent.
0:35:38 – Well, I think buying volatility in the sense
0:35:40 that we’re buying people have world-class strengths
0:35:42 where we care about them,
0:35:45 and regardless of whatever else.
0:35:46 There’s volatility in that,
0:35:49 but you can have a different kind of volatility.
0:35:51 You can have people who have gigantic weaknesses
0:35:54 that are spectacular without having the strengths.
0:35:57 And we’re not trying to buy that kind of volatility.
0:35:58 – How do you know, though,
0:35:59 that they’re going to be the ones
0:36:01 to actually build the companies at scale?
0:36:03 Because there seems to be this inflection point
0:36:05 where the very thing that makes you a founder
0:36:07 that’s going to punch through this tough industry
0:36:10 is also the thing that’s pretty much going to hold you back
0:36:11 from really building your company
0:36:13 in a really meaningful way
0:36:15 if you think you can do everything your way.
0:36:17 And there seems to be an inherent contradiction in that.
0:36:19 – I think that that would be right
0:36:21 if founders did not evolve.
0:36:23 So I think-
0:36:23 – And some don’t.
0:36:24 – And some don’t.
0:36:26 Like some don’t and they get stuck
0:36:28 and they can’t get past that point.
0:36:32 But it’s a real common characteristic in great founders
0:36:34 that they want to know absolutely everything
0:36:35 about the company and how it works
0:36:38 and every knob and every button.
0:36:43 And they really would like, have a strong desire
0:36:45 to actually be able to do every job
0:36:47 in the company themselves if it came down to it.
0:36:51 But those kind of founders also have great ambition
0:36:54 and it’s very logical and easy to understand
0:36:58 that there’s never actually been a gigantic long,
0:36:59 really important long lasting company
0:37:02 that had like five employees that those just don’t exist.
0:37:05 And so if you’re gonna have to have a bigger company
0:37:10 than that, you have to think about the company,
0:37:12 not only from the scale perspective,
0:37:15 but from the perspective of the people working there.
0:37:18 And how are you gonna get great people to work with you
0:37:21 if you’re literally making a redecision in the company?
0:37:22 And I think that, look,
0:37:24 not every founder can let go of that.
0:37:26 And sometimes it’s a psychological flaw
0:37:29 rather than a desire for greatness.
0:37:32 And if it’s a psychological flaw that they can’t overcome,
0:37:36 then it’s just like any flaw that any of us have.
0:37:37 We can’t stop eating ice cream or whatever.
0:37:40 And there’s nothing we can do at that point,
0:37:42 like we can give them the logical explanation,
0:37:44 but they’ve got to fix themselves.
0:37:46 One of the things that we’ve seen even in the short time
0:37:49 that the firm has been in business is companies
0:37:51 staying private longer, taking longer times to IPO.
0:37:53 What are some of the implications of that
0:37:54 on the company building process?
0:37:57 How do you kind of balance that new reality?
0:37:59 If it is a new reality around how companies stay private
0:38:01 with how you think about building management teams
0:38:03 and other issues around the company.
0:38:05 – Yes, I think this gets back to probably
0:38:09 one of the more neglected parts of company building,
0:38:10 which is like, what is the company culture?
0:38:11 What does it believe?
0:38:13 What’s our way of doing things?
0:38:14 When we come to work every day,
0:38:16 what does quality mean?
0:38:18 How do we prosecute an opportunity?
0:38:23 And the kind of philosophy onboarding,
0:38:25 training into that culture and so forth.
0:38:27 And so you kind of have to develop a philosophy
0:38:29 like what kind of employees do you want?
0:38:31 How do you want them to behave when they get there?
0:38:32 How do people contribute?
0:38:34 – As we’re getting close to wrapping up here,
0:38:36 what would be one piece of advice that you might give
0:38:37 either from a management perspective,
0:38:39 from a go-to-market perspective?
0:38:40 What would be a takeaway for people
0:38:42 listening to this podcast?
0:38:44 – From a management perspective,
0:38:46 I think the most common mistake that founders make
0:38:50 is they make decisions based on management decisions
0:38:52 and organizational design decisions
0:38:56 based on very kind of proximate perspective.
0:38:57 So what’s my perspective?
0:39:00 What’s a person I’m talking to perspective?
0:39:03 What’s my HR person’s perspective?
0:39:05 Without like taking the time to go,
0:39:07 okay, like how does everybody in the entire company
0:39:08 see this decision?
0:39:10 And how will they see it once it’s made?
0:39:13 Is it motivating people in the way that I think it will?
0:39:16 And let’s look past the person I’m talking to
0:39:18 feeling good about what I’m saying
0:39:20 and really make this for the long-term health
0:39:22 of the organization.
0:39:24 – Single biggest strategic piece of advice
0:39:25 we just see across all of our companies
0:39:27 is literally people just need to raise prices.
0:39:29 People need to charge more for their products and services.
0:39:31 The good news is you have all these new founders
0:39:33 with many different backgrounds who have come in.
0:39:35 Many of them have never run companies before
0:39:36 or run sales forces before.
0:39:38 And so they have these extremely sophisticated views
0:39:40 on things like products and design and engineering.
0:39:42 And then I think in some cases,
0:39:45 relatively naive views on how to actually prosecute
0:39:47 a campaign to be able to get the world to use your product.
0:39:50 And so the temptation we see from many founders
0:39:53 is to have a one-dimensional view of what I call
0:39:54 a one-dimensional view of the relationship
0:39:56 between price and volume.
0:39:58 Which is if I price my product cheap,
0:39:59 then I sell more of it.
0:40:01 ‘Cause the assumption is just that people just make
0:40:03 purchase decisions based on cost.
0:40:05 And so you drive down prices, you drive up volume.
0:40:07 And by the way, a lot of the history of the tech industry,
0:40:08 like the chip industry is,
0:40:10 drive down prices, drive up volume.
0:40:14 But a lot of startups really suffer from having that view.
0:40:16 Instead, we encourage companies to adopt
0:40:17 what I call kind of the two-dimensional view,
0:40:19 which is the advantage of raising prices.
0:40:21 Actually, there’s a couple of advantages.
0:40:22 So one big advantage, if you raise prices,
0:40:25 you can afford a bigger sales and marketing effort.
0:40:28 A lot of companies have prices that are actually too low
0:40:30 to be able to mount the kind of sales and marketing campaign
0:40:32 required to get people to ever actually buy the product.
0:40:34 And I call this the too hungry to eat problem, right?
0:40:38 I’m not selling enough, but I’m not selling enough
0:40:39 ’cause I don’t have the sales and marketing coverage required
0:40:41 to actually get the product out there.
0:40:43 And I don’t have that ’cause I’m charging too little.
0:40:44 And as a consequence, I’m not selling any,
0:40:46 despite my low prices.
0:40:47 The other really interesting thing
0:40:49 is that for a very large number of products,
0:40:50 it turns out if you charge higher prices,
0:40:52 the customers take the product more seriously.
0:40:53 They impute more value into it
0:40:55 when they’re making their purchase decision.
0:40:56 And then once they’ve purchased,
0:40:57 they’ve made a bigger commitment to it.
0:41:00 And in particular, anybody selling anything to businesses,
0:41:02 businesses will take something that they had to pay
0:41:03 a lot of money for a lot more seriously
0:41:05 than something that they didn’t have to pay
0:41:06 very much money for.
0:41:08 So you can get a much higher level of engagement
0:41:10 and stickiness and actual use of your product
0:41:11 if you charge more.
0:41:12 – Going through this,
0:41:13 this definitely has felt like swimming upstream
0:41:14 for the last several years.
0:41:16 We see some glimmers that more folks
0:41:17 are starting to figure this out.
0:41:19 – Okay, well, that’s all we have time for.
0:41:20 I think this is the first time
0:41:22 I’ve actually had all you guys together on the podcast
0:41:24 since we did our fifth anniversary podcast
0:41:25 a couple of years ago.
0:41:26 Kind of amazing how much has changed
0:41:28 even in that short amount of time.
0:41:29 So thank you.
0:41:30 Thanks everyone.
with Marc Andreessen (@pmarca), Ben Horowitz (@bhorowitz), Scott Kupor (@skupor), and Sonal Chokshi (@smc90)
Continuing our 10-year anniversary series since the founding of Andreessen Horowitz (aka ”a16z”), we’re resurfacing some of our previous episodes featuring Andreessen Horowitz founders Marc Andreessen and Ben Horowitz.
This episode was actually recorded in 2016 — on the 5-year anniversary of Marc’s Wall Street Journal op-ed on “Why software is eating the world” — and features Sonal Chokshi and Scott Kupor interviewing Ben and Marc about what’s changed since, and how software is programming the world… in everything from simulations to distributed systems to other key computing shifts.
You can find other episodes in this series at a16z.com/10.