Washington Is About to Back Open Weights. China Already Set That Price.
Twenty-five US tech firms lobbied against open-model restrictions the same week US developers started reaching for a Chinese model by default. That is defense, not principle, and it tells you where model margins go next.
On July 24, twenty-five of the largest names in American technology sent Washington a letter asking it to stay out of the way. Nvidia, Microsoft, Meta, IBM, Dell, Palantir and Hugging Face all signed, urging the US government not to impose premature restrictions on open-weight AI models. The stated case was competition, security, and American AI leadership.
Two days later, the Associated Press published a feature explaining why they bothered. US developers, including Mozilla's CTO, are adopting Moonshot AI's Kimi K3, a Chinese open-weight model released in mid-July, because it is cheap and it works.
Read those two stories in that order and the letter stops looking like a philosophical position.
## The default was already set, and not in California
The uncomfortable fact for US labs is that the open-weight question is no longer being decided in Washington or in Menlo Park. It is being decided one API key at a time by developers picking the cheapest thing that clears their quality bar.
Kimi K3 is pulling users away from OpenAI and Anthropic in specific workloads, and Alibaba is shipping more capable systems into the same lane. When a developer in Austin defaults to a Chinese open-weight model on a Tuesday afternoon, no export-control memo reaches that decision.
So when twenty-five US firms rally around open-weight policy, they are not opening a door. They are pointing out that the door came off its hinges months ago, and that restricting American open models now would only remove the American option from the menu.
Nvidia's incentive here is worth naming plainly. It sells shovels. It does not care whose model runs, only that many models run. Meta and Hugging Face have their own open-weight commitments. This is a coalition of firms whose economics improve when weights are free, arguing that weights should stay free. That does not make them wrong. It does mean the letter is a business position wearing a policy costume.
## What this does to margins
Here is the part that will show up in earnings before it shows up in legislation.
If a good-enough open model is free, the price of a closed model is no longer set by what it costs to train. It is set by the gap between the closed model and the best free one. That gap is compressing from both directions. Anthropic just released Claude Opus 5 at roughly half the price of Fable 5 for comparable work, which is what price competition looks like at the frontier. Chinese open weights are compressing it from below.
The Chinese side is not frictionless either. DeepSeek put its second fundraising round on hold after founder Liang Wenfeng's remarks on US-China competition went viral, while still emphasizing open-source models and a long-term AGI push. Giving the weights away and raising capital against them remains an unsolved problem on both continents.
Which is why the money keeps moving down-stack. NVIDIA and SK Group are planning a $500 billion-plus partnership for a 2GW AI factory and next-generation HBM memory. If model weights trend toward zero, the durable margin sits in the silicon, the memory, and the power. Every large check being written right now agrees with that reading.
## What to do with this
**Stop pricing model access as a moat and start pricing it as a input cost.** If you are building on a closed API, run the same evaluation against Kimi K3 or an Alibaba open model this quarter. Not to switch, necessarily, but because you should know the size of the premium you are paying and what it buys you. If you cannot articulate that gap in a sentence, you are not negotiating, you are subscribing.
**Watch whether the letter changes anything.** The specific thing to track is not whether Washington issues a supportive statement. It is whether any US lab responds by releasing competitive open weights of its own. That is the only move that would actually contest the default. If the next quarter passes with strong lobbying and no strong American open release, the argument was already lost when the letter was written.