The Wall Was Built for Nvidia. Nvidia Is Building the Doors.

Nvidia is the chief beneficiary of US chip export controls. It is also the strategic investor in River AI's $1.1 billion open-weight tooling platform ^[1], the leader of a 25-company lobbying coalition against restricting Chinese open-weight models ^[2], and the architect of a $500 billion financing platform for AI deployment infrastructure, built with Apollo, Blackstone, Goldman Sachs, BlackRock, Brookfield, and KKR ^[3]. The company the containment wall was built to shield is the company building the doors through it. AMD led TensorWave's $350 million raise to build an AMD-powered inference cloud, one that pointedly avoids Nvidia hardware ^[4]. Those data centers will run whatever models customers bring, and the CEO said the company can sell capacity faster than it can sign leases. IBM signed a $240 million deal with Together AI to build a dedicated open-source inference cluster on IBM Cloud ^[5]. Together AI's CEO was explicit about the objective.

This cluster lets us bring production-grade inference to more companies, faster, and it's a big step in our push to make open-source AI the obvious choice for enterprises. — Vipul Ved Prakash

Microsoft built a business routing OpenAI's models to ByteDance for more than $1 billion annually, channeled through Singapore to keep the servers outside Chinese territory ^[6]. Each company found a different instrument. The incentive was the same. On August 4, the Trump administration exempted open-weight AI models from mandatory safety testing ^[7]. National Cyber Director Sean Cairncross explained the logic plainly.

A regulatory regime would not only strangle growth, development and innovation, and be enormously harmful to the industry, but it would be obsolete 48 hours after it was going through whatever process it had gone through. — Sean Cairncross

White House AI adviser David Sacks went further.

The weaponization of regulatory uncertainty as a competitive tool should be completely unacceptable. — David Sacks

The administration looked at the deployment layer and chose not to govern it. The numbers now bear this out. By March, the US-China Economic and Security Review Commission reported that roughly 80% of US AI startups build on Chinese open-source base models ^[8]. Chinese models now account for over 70% of tokens served on inference platforms globally, up from less than 3% fourteen months earlier ^[2]. In absolute terms, Chinese models outpaced US models in weekly token usage by 12.96 trillion to 3.03 trillion ^[9]. The deployment layer, where AI is actually used, belongs to Chinese models. Export controls are still enforced at the chip layer. DeepSeek is developing custom inference chips to bypass US restrictions; founder Liang Wenfeng acknowledges the controls are a challenge ^[10]. The US Bureau of Industry and Security is actively investigating Chinese firms' access to restricted Nvidia chips ^[11]. Independent benchmarks place Chinese models three to nine months behind US rivals in capability, a gap that has held steady for years ^[12]. A Chinese researcher used DeepSeek to autonomously hack 460 systems, breaching 14 of them, because the model lacked safety guardrails that blocked the same requests on OpenAI and Anthropic models ^[13]. These facts operate at the chip and capability layers. The deployment layer, where models are hosted, financed, and consumed, was conceded by different means. Anthropic refused to join the 25-company coalition ^[14]. It has accused Chinese firms of industrial-scale IP theft, documenting 24,000 fraudulent accounts and 16 million exchanges used to distill American models ^[15]. It stands as the only major American lab arguing for more restrictions. Nvidia, Microsoft, OpenAI, Alphabet, AMD, and IBM have all taken the other side, whether through investment, lobbying, or commercial partnership. When the one company arguing for the wall is isolated by every company that benefits from the wall, the industry has shown what it values. The deployment layer pays better than the wall.