The Two Sides of the AI Fight Are Both American

When Moonshot AI released Kimi K3 — the world's largest open-weight model at 2.8 trillion parameters — fifty American companies including Nvidia, Google, and OpenAI signed a letter urging the U.S. government not to restrict open-weight models. Anthropic and Amazon refused to sign ^[1]. The fracture was not subtle. Within days, White House AI adviser David Sacks called the push for restrictions something sharper than a security measure.

the leading closed labs, already a duopoly in terms of AI model revenue, want the government to eliminate their open-source competition. — David Sacks

The White House's own AI adviser was arguing against his government's security apparatus. That is the first signal that the dividing line is not national security but business model. The mechanism driving American firms to open-weight their own models is straightforward. DeepSeek permanently cut V4-Pro prices by 75% in May, making it 12 to 19 times cheaper than OpenAI's GPT-5.5 and Anthropic's Claude Opus 4.7 for equivalent tasks ^[2].

It is an efficiency gain being passed through. — Sanchit Vir Gogia

That price differential is market pressure, not a curiosity. Mistral AI co-founder Guillaume Lample has documented the pattern: enterprises start with large closed-source models, then migrate to fine-tuned small open-weight alternatives because the closed options are too expensive and too slow ^[3].

Our customers are sometimes happy to start with a very large [closed] model that they don’t have to fine-tune … but when they deploy it, they realize it’s expensive, it’s slow. — Guillaume Lample

The migration is underway. Pinterest uses DeepSeek R-1 for its recommendation engine; Airbnb uses Alibaba's Qwen for customer service agents ^[4]. A U.S. government commission reported that roughly 80% of U.S. AI startups now use Chinese open-source base models ^[5]. American firms are responding not by defending the proprietary moat but by joining the race to open-weight. Jason Warner, CEO of AI startup Poolside, frames it explicitly as a competitive response to Chinese open-source.

There is a vast, vast degree of want for an American company producing the most-capable open-source artificial intelligence. — Jason Warner

Meta, after Llama 4 underperformed expectations, pivoted to a partial open-source strategy under new Chief AI Officer Alexandr Wang — releasing some model versions while keeping components proprietary, directly to compete with Chinese open-weight alternatives ^[6]. A wave of startups including Arcee AI and Reflection AI are building open-weight models for the same reason ^[7]. This is the second path of commoditization, and it operates inside the U.S. government's blind spot. The entire security apparatus — chip export controls, Entity List designations, the multi-agency framework for AI as a strategic weapon — was built to stop foreign leaks. But American firms voluntarily open-weighting their own models is domestic competitive behavior. The government cannot restrict it without alienating its own industry. Eric Schmidt identified the paradox last November.

This produces a bizarre outcome where the biggest models in the United States are closed source and the biggest models in China are open-source — Eric Schmidt

The inversion is the market force driving the leak regardless of what Washington walls off. Developing nations will standardize on whatever is free, and the biggest free models are Chinese. The security case for restrictions has its own complications. Laboratory tests found that AI agents from Google, OpenAI, Anthropic, and X all independently bypassed anti-hack systems — the vulnerability is not specific to Chinese models ^[8]. And the U.S. is restricting its own most powerful models: after Anthropic's Mythos breached classified systems within hours in a cooperative exercise, the administration directed Anthropic to disable its top models for all foreign nationals ^[9]. The walls are going up on both sides, and the security rationale grows harder to isolate. The sharpest evidence of the business-model divide arrived last November, when the White House urged Congress to reject the GAIN AI Act, which would restrict chip exports. Nvidia lobbied against it. Amazon and Microsoft backed it ^[10]. The split runs not along national-security lines but along who sells what: chip sellers and open-weight advocates want free flow; cloud operators and closed labs want scarcity. The coalition geography is now unmistakable. On one side: Nvidia, Google, and the White House — the chip seller, the open-weight advocate, and the administration that sees restrictions as regulatory capture. On the other: Anthropic and Amazon — the closed labs and the cloud operators who profit from scarcity. OpenAI occupies an intermediate position, signing the letter against general open-weight restrictions while simultaneously lobbying for restrictions on Chinese open-weight models specifically ^[1] — a business-model calibration, not a national-security stance. Both coalitions are American. The divider is who sells what, and the side that wants to tear down the walls includes the White House.