Nvidia, Meta, Microsoft and OpenAI Jointly Urge Washington on Open AI Models

Tech companies that normally compete head to head found themselves on the same side of a Washington dispute this week. Nvidia, Meta, Microsoft, IBM, Dell, Palantir, ServiceNow, CrowdStrike, Hugging Face, Mistral, Perplexity, and investors Andreessen Horowitz and Y Combinator signed a joint open letter urging US policymakers to drop planned restrictions on open artificial intelligence models. OpenAI joined them a day later, leaving Anthropic as the coalition's only significant opponent.

The immediate trigger for the letter was a series of events in mid-July. On July 16, Chinese company Moonshot AI released the Kimi K3 model, which outperformed some American rivals on coding benchmarks. Six days later, White House AI advisor Michael Kratsios accused Chinese firms of industrial-scale distillation, building their own models on data pulled from closed systems at American labs. Washington began considering tougher rules for models released with open weights, arguing that they make such practices easier and erode the US advantage.

Who signed the letter

The list of signatories brings together companies that normally compete at every layer of the technology stack, from chipmakers to cloud providers to application developers. Nvidia and Microsoft compete for AI infrastructure customers, Meta and Mistral rival each other in the open model market, and Palantir and ServiceNow fight over enterprise contracts. Yet all of them found common cause in defending open weights as a foundation of competitiveness.

The letter emphasizes three main arguments. First, open models lower the barrier to entry for startups and public institutions that cannot afford to train their own frontier models or pay for access to closed APIs. Second, they increase competition at every layer of the technology stack, from chips to cloud infrastructure to applications. Third, they let organizations avoid vendor lock-in, since they can control their own data and adapt models to their own needs.

The security argument, reversed

The signatories also flip the usual security argument made against open models. Rather than claiming open weights are a threat, the letter argues that network defenders facing AI-equipped attackers need access to models with comparable capabilities to detect threats effectively. The reasoning echoes the long-running debate around open source software in cybersecurity, where code transparency is often cited as an argument for security rather than against it.

On the question of distillation, the signatories propose a distinction: distillation as a technique for training smaller models on larger ones is a legitimate and widespread research practice that must be separated from the unlawful extraction of value from closed models without their creators' consent. The letter opposes blanket restrictions that would lump both situations under a single ban.

OpenAI's turnaround, Anthropic's isolation

The most notable element of the whole story is OpenAI's stance. The company did not sign the letter on the first day, even though its CEO Sam Altman had previously voiced support for open models. It joined only a day later, along with nine other companies, expanding the coalition to 35 signatories. OpenAI president Greg Brockman explained the move by saying that artificial intelligence is something that truly needs to be democratized, and that having more models is good for everyone.

AI is something that truly needs to be democratized. Having more models is good for everyone - Greg Brockman, President of OpenAI

The community can study behavior, identify vulnerabilities and develop safeguards - excerpt from the 35-company coalition letter

OpenAI's decision leaves Anthropic isolated as the sole opponent in this dispute. The company has consistently warned about the risks of open weights for months. Dario Amodei argued before Congress that scaling them further is heading in a very dangerous direction, and he called the security arguments raised by open-model advocates a red herring, a distraction from the real problem.

What it means for Polish companies

For Polish companies and institutions, this debate has direct practical significance. Open models such as Google DeepMind's Gemma family or Mistral's models are currently one of the few paths to deploying artificial intelligence without depending on foreign API providers or sending data outside their own infrastructure. If the US administration actually restricts the export or development of such models, European companies could lose access to technology they already build their deployments on, and get pushed toward more expensive, closed alternatives.

The episode also shows how quickly technology geopolitics translates into real business decisions. The release of a single Chinese model, along with the accompanying accusations of data theft, was enough to mobilize thirty-five companies, including direct competitors, into a joint intervention in Washington within days. The outcome of this regulatory dispute will now be watched not just by the US market, but by everyone building products on open models.