Despite ChatGPT’s commanding lead with nearly a billion weekly users, Google’s Gemini is growing at a staggering 155% year-over-year, largely fueled by viral multimodal models like NanoBanana. This dynamic underscores a broader shift in consumer AI: raw model quality is becoming table stakes, and the battle is now being won or lost on product sensibility and interface details. As the A16Z consumer team analyzes 2025, they observe that while the big labs dominate general-purpose assistants, their forays into standalone social or vertical apps often stumble, leaving ample room for focused startups.

The conversation details how OpenAI and Google took divergent product strategies. OpenAI largely integrated new features like Pulse, group chats, and shopping directly into the ChatGPT interface, reinforcing its position as a daily hub. Google, meanwhile, frequently launched standalone products and websites, like those for its viral VO video and NanoBanana image models, allowing for more tailored interfaces. This approach helped Gemini gain traction, especially when its models went viral for their creative capabilities. However, even with Google’s distribution advantage, ChatGPT remains the “Kleenex” of AI for most users, highlighting the immense power of brand and habit.

Looking ahead to 2026, the analysts predict that multimodal “anything-in, anything-out” models will become the norm, blending text, image, and video reasoning. This will open new frontiers for design and creative tools. Furthermore, they argue that while lab-made models will keep improving, the labs themselves are often structurally ill-suited to build opinionated, standalone consumer applications. This failure to vertically integrate into every use case, combined with inherent compute trade-offs between training and viral inference, creates a durable opening for startups that can leverage these powerful models to build superior, focused products.

Surprising Insights

• Low Multi-App Usage: Less than 10% of consumers pay for more than one major LLM service, and a similarly small fraction of ChatGPT users even visit a competitor, suggesting the market is trending toward “winner-take-most.”
• Interface Over Raw Power: A key differentiator is not just model capability but user onboarding. For example, ChatGPT’s new image pane uses TikTok-like trending templates to guide users, while Gemini presents a blank screen, creating a significant barrier to first-time creation.
• Social AI’s Missing Ingredient: Attempts by labs to build social apps like Sora 2 or ChatGPT group chats have struggled because AI-generated content lacks a real “status game”; when the output isn’t an authentic representation of the user, the core emotional driver of social media diminishes.
• The Power User Economy: For the first time, consumer AI products are achieving greater than 100% revenue retention, as power users exceed subscription quotas and pay for additional usage, transforming the traditional consumer subscription model.

Practical Takeaways

• Experiment Beyond the Giants: For specific workflows, try dedicated products that may outpace lab offerings: Gamma for slide decks, Granola for AI notetaking in meetings, or the Perplexity Comet browser for an AI-native workspace.
• Leverage Templates for Creativity: The viral success of styles like “Ghibli” shows that accessible templates are crucial for mainstream adoption. Use platforms like Kriya or Heygen that popularized this format, or look for these features within lab products to lower the barrier to creation.
• Use AI for Audio Assimilation: Convert reading lists into listening material with tools like 11 Labs Reader to consume written content on the go, effectively adapting to changing media habits.
• Build for Depth, Not Just Breadth: Startups should focus on building deeply opinionated products for power users in specific verticals, as labs are incentivized to improve broad, core metrics and often lack the focus to win in narrow, high-depth domains.