ChatGPT Is the Tip of the Iceberg — A Japan-Based Commentator Reads the AI Industry's Tectonic Shift and Anthropic's Reversal

The host, who has lived in Shanghai, New York, and now New Zealand with his family, opens with a paradox: 'I barely open ChatGPT for work anymore' — even though his overall AI usage is far higher than before. That contradiction is his entry point into the episode's core subject.

In Japan, 'AI' still mostly means ChatGPT, and by monthly active users it remains the world's largest. But the industry-wide numbers tell a different story: in April 2026, Anthropic overtook OpenAI in revenue. 'Claude — that's the chat thing, right? Why is it so strong?' is exactly the naive question that leads into the distinction between AI's visible surface and its invisible backend, which is the spine of the whole episode.

What's happening behind the scenes? The symbolic event is Google announcing up to $40 billion in additional investment in Anthropic. That Google — with its own strong AI in Gemini — would still pour that much into Anthropic signals just how seriously it's chasing the enterprise-backend AI market.

Shift perspective to China, and the strategy is a complete opposite: distribution above everything. DeepSeek and Alibaba's Qwen are the flagship names. Price-wise, DeepSeek's latest model claims equivalent performance at roughly a quarter of Claude's top-tier price, and it's distributed as downloadable open weights anyone can run on their own hardware or servers. As of 2025, Chinese open models' global download share overtook the US — nearly 20% of the total worldwide is Chinese-made. To summarize: the US side runs closed, high-quality, high-price enterprise sales; the Chinese side runs open, cheap, mass distribution. Under the same word 'AI,' the business model has split into two completely opposite poles.

From here the host shifts to his own opinion. Financially sensitive analysis or intricate decision-making — work that can't afford to fail — goes to closed, high-quality models. But bulk, low-cognitive-load processing is being taken over fast by cheap Chinese open models, where the price gap can be one or even two orders of magnitude. For simple, high-volume tasks like routine classification, price competition is essentially over.

And information that absolutely can't leave the premises, or processing that needs to run fully local, is now 80-90% Chinese open models. This isn't a deliberate preference for 'Chinese' — it's a process of elimination: that's practically the only option that can be downloaded and run on his own server. Recalling the opening premise — that AI isn't just chat-box Q&A — the volume of AI usage in coding and business automation dwarfs one-off chat interactions by orders of magnitude. Behind that much larger volume, the same US-high-quality-high-price versus China-cheap-broad split is happening naturally even at the individual-user level, and ChatGPT — long the runaway leader — is quietly losing ground. That's the real shape of the tectonic shift happening in AI right now, and he notes this isn't getting much coverage in Japan.