US AI labs have a simple plan: raise the monthly rates for using their models and shift most customers to token pricing.
China stands against that plan.
A seven-month lag for one-twentieth of the price — for most tasks in the world, the choice is pretty clear.
The Game Is Last Man Standing
Chris Daniel once broke down this strategic move on the Wardley Maps forum, and Amazon plays it masterfully.
Here's how it works:
First, you keep prices high and make money. But you don't pay the profits out as dividends. You increase efficiency. From the outside, you look like a company with normal margins. Competitors take your price as a benchmark and plan a ‘reasonable’ business around it.
Then you cut prices. Demand spikes and competitors face two problems at once: margins are gone, and they lack capacity for the new customers. Before they recover, you cut prices again.
You repeat, until you're the last one standing.
Amazon has been doing this for years. It cuts prices regularly, and with each cut, its services make sense for more customers who previously found them too expensive. The market grows, and Amazon grows with it. A competitor who priced their business against Amazon's old numbers has to run the math again.
China Plays Efficiency
The US bets on performance: the best model, a premium price. China doesn't have that choice — the lack of top-tier chips pushes it toward efficiency. Cheap training, cheap deployment, cheap everything.
That's exactly what Last Man Standing needs. And Chinese companies add one more strategic move: open-weight models. Open solutions accelerate the commoditization of the market.
The Chase for Performance Will End
New generations of models still jump noticeably up. At some point, you'll stop caring.
A model that reliably sorts your inquiries or drafts first versions of contracts — you don't need to swap it for a better one. You need to make it cheaper to run.
And then Last Man Standing kicks in.
Countermoves
What can US labs do?
Spread fear — Chinese models are a security threat, your data ends up in Beijing. Expect louder arguments that open-source AI is dangerous.
Pull the government in — bans on Chinese models in government agencies and sensitive sectors. And for all (sub)contractors serving the state.
Export controls — choke China's access to top-tier chips so it can't play on performance. Except that's exactly the pressure that taught China to play on efficiency.
Open approaches — release their own open-source or open-weight models. OpenAI released gpt-oss back in August 2025; this year, Google and Nvidia added open models too. All of them still behind the Chinese open-source frontier... because otherwise the labs would cannibalize their own profits. So this move is toothless for now.
Ecosystem instead of the model — sell the environment around the model: agents, tools like Claude Code, integrations. Watch what customers build with them and pull the best into the product. The model is replaceable. The ecosystem is not.
A new layer above models — Japan's Sakana AI showed this with Fugu: it assembled performance comparable to Fable 5 from publicly available models and sells a single API on top of them.
What to Do About It
Stop worrying about expensive models. That's PR fearmongering from US labs. Long-term, prices will fall.
Ask yourself what you'll do when intelligence is nearly free.
Not just Wardley's strategic moves — I break them down plainly and clearly in the book MAKE CUTS. Pre-orders are open.
"Your margin is my opportunity."
– Jeff Bezos
Weekend Reading
When Americans choose Chinese AI
US developers reach for Chinese models and tell no one about it.
