The US built the worlds best AI and ranks 20th in who actually uses it. Meanwhile, ChatGPT and Claude are splitting into two completely different businesses, and that divergence is the real signal for anyone building in this space.
Singapore, Hong Kong, the United Arab Emirates, and South Korea lead per capita AI adoption globally, according to the sixth edition of Andreessen Horowitz's Top 100 Gen AI Consumer Apps ranking, released March 10, 2026. The U.S. sits at number 20 — despite being home to OpenAI, Anthropic, Google DeepMind, and every other major AI lab. a16z Top 100 Gen AI Consumer Apps, 6th edition
The divergence between who builds AI and who adopts it is one of two striking patterns in the report. The other is a clear strategic split between the major platforms themselves. ChatGPT is 30 times larger than Claude on the web and 80 times larger on mobile, according to the a16z podcast featuring a16z General Partner Anish Acharya and partner Olivia Moore. But the platforms are now building toward different destinations — and anyone launching a product in this space needs to understand which bet they're making.
The trust gap is part of the U.S. story. Edelman data cited in the report shows just 32 percent of Americans trust AI, compared to 72 to 87 percent across surveys in China. Edelman Trust Barometer 2025 The cultural anxiety around AI — job displacement, creative harm, existential risk — appears to be suppressing adoption even where access is universal. About a third of Americans are monthly active AI users. Some smaller countries are past 50 percent. Pew Research data released in February 2026 shows 57 percent of U.S. teens using AI for information search and 54 percent using it for schoolwork, Pew Research Center, February 2026 which suggests the next generation will narrow that gap — but the institutional skepticism is real and durable.
ChatGPT's scale remains in a category by itself. It reached 900 million weekly active users in February 2026, TechCrunch, February 2026 putting it within striking distance of a number no consumer AI product has touched. On paid subscribers, ChatGPT is 8 times larger than Claude and 4 times larger than Gemini. But the growth rates tell a different story: Claude added paid subscribers at over 200 percent year-over-year, while Gemini grew at 258 percent. a16z Top 100 Gen AI Consumer Apps, 6th edition
The platform app ecosystems are starting to reflect divergent bets. ChatGPT and Claude each now list more than 200 apps in their directories — but they share just 41 apps in common, roughly 11 percent of the combined catalog. a16z Top 100 Gen AI Consumer Apps, 6th edition ChatGPT's additions skew toward consumer marketplaces, travel, nutrition, and transaction-oriented tools. Claude's are research databases, financial data, and science tools. "ChatGPT is going more of the Google approach," Acharya said on the podcast. "They'll monetize through ads and probably transactions — if they're the gateway to booking a trip, they can take a cut of that traffic." Claude, by contrast, is explicitly a subscription business: "it won't be everyone," Acharya noted.
Voice is the next consumer wave and may arrive faster than adoption curves suggest. "I think that is going to spread to the mainstream consumer in the next six to nine months," Acharya said. a16z podcast Currently, AI usage at scale is still concentrated in information retrieval and homework — the same tasks Google handled before AI existed. The real expansion will come when AI can execute tasks across systems, not just respond to prompts.
We are still at 10 percent weekly global penetration. The technology, Acharya argued, has outrun the culture.
The agent wave is real but uneven. Manus — which lets users delegate tasks across email, web browsing, spreadsheets, and slides — was arguably the first consumer-grade agent that worked reliably. Meta acquired Manus for over $2 billion in December 2025, TechCrunch, December 2025 a price that reflects both the capability and the strategic threat a horizontal agent poses to platforms with distribution. Acharya's read: "Once everyone has that agentic capability, if you're such a horizontal product, you may be better off with the distribution forces of Meta or Google." The question for the next wave of agent startups is whether they can build something the platforms can't replicate — or whether they should aim to be acquired before the window closes.
OpenClaw, the Austrian open-source AI assistant project, would have ranked #30 on the a16z web list if the data window extended to February 2026, a16z Top 100 Gen AI Consumer Apps, 6th edition when it crossed 68,000 GitHub stars to become the most-starred project on the platform. But new user signups have plateaued since early February even as GitHub stars continue climbing — a pattern that suggests it's become essential infrastructure for technical users without yet escaping containment to the mainstream. Peter Steinberger, the creator, TechCrunch, February 2026 joined OpenAI on February 15, 2026. The open question — whether OpenClaw narrows to OpenAI-only models or stays multi-model — cuts to the heart of what gives it value. Multi-model flexibility is what makes it useful to developers building across the frontier. Lock-in to a single provider may be what makes it valuable to OpenAI.
On memory, OpenAI is building what amounts to an authentication layer — a way for users to carry their context and usage history across apps that build on ChatGPT. a16z podcast The logic mirrors Google's early search lock-in: if your preferences and history live in one system, you'd rather stay there. "Essentially you'd be able to log in with your ChatGPT account and take your memory and your tokens with you," Acharya said. "And then that other product would be able to borrow those things to be even more powerful." The user wins. The developer wins by not paying for inference. ChatGPT wins by being the layer everything else runs on.
The platforms are making their bets. The data shows who's winning today. What it can't show is which architecture wins in three years — only that the divergence is real, and building in this space means choosing a side.
