Where does consumer AI stand at the end of 2025? - YouTube
ChatGPT dominates with 800M+ weekly users, but only 9% of users even try competitors—yet Gemini is growing 7x faster through viral creative models while the big labs struggle to ship opinionated products beyond chat, creating a massive opening for consumer AI startups in 2026.
Read Original Summary used for search
TLDR
• Consumer AI is trending winner-take-most: only 9% pay for multiple AI products, <10% of ChatGPT users visit other LLM providers
• Gemini growing 155% YoY (vs ChatGPT's 23%) driven by viral creative models like Nano Banana, which can now generate market maps with web research integration
• Social features flopped because AI content lacks "status game"—Sora 2 works as creator tool (like CapCut) but not as social app
• Labs face structural constraints: compute tradeoffs between training/inference, promo committees favoring safe incremental features, single-model lock-in
• Startups have opening in 2026: models now production-ready, power users driving 100%+ revenue retention through usage pricing, multi-modal products require opinionated design labs can't ship
In Detail
The consumer AI market is consolidating faster than expected, with ChatGPT commanding 800-900M weekly active users while competitors like Gemini sit at 35-40% of their scale. The stickiness is extreme: only 9% of consumers pay for multiple AI products, and for most of 2025, less than 10% of ChatGPT users even visited competing platforms. Yet the landscape is shifting rapidly—Gemini is growing desktop users 155% year-over-year (accelerating even at scale) versus ChatGPT's 23%, driven primarily by viral creative models like Nano Banana and Veo rather than text reasoning improvements.
The a16z consumer team identifies a critical pattern: while OpenAI and Google shipped aggressively in 2025, their product execution diverged in revealing ways. OpenAI kept features inside ChatGPT (Pulse, group chats, shopping, research), while Google launched standalone products through AI Studio and Labs. Both struggled with consumer products beyond their core chat interfaces—social features like Sora 2's TikTok-style feed and group chats failed because AI-generated content lacks the "status game" that drives participation on traditional social platforms. Sora 2 succeeded as a creator tool (analogous to CapCut, not TikTok), with content going viral on existing platforms rather than within the app itself. The labs face structural constraints that advantage startups: compute must be allocated between training new models versus inference (and within inference, between entertainment and productivity use cases); promo committees incentivize safe incremental features over opinionated products; and single-model lock-in prevents them from delivering best-in-class experiences that require combining multiple models.
For 2026, the team predicts "more of the same" from labs on core products, but a breakthrough year for consumer builders. Models have reached production quality for real scalable apps, and power users are driving unprecedented monetization with 100%+ revenue retention through usage-based pricing on top of subscriptions—something impossible in pre-AI consumer products. The opportunity lies in opinionated, multi-modal applications (anything-in to anything-out) where startups can combine the best models across modalities without compute constraints or corporate risk aversion. Key areas include app generation (though labs will attempt it), video-first products with templates and styles (not just raw capability), and specialized verticals like Claude's dominance with technical users. The critical insight: when input and output are both text, ChatGPT's frequency advantage is nearly insurmountable, but creative angles on modality, workflow, or vertical focus create defensible positions.