Anthropic Researcher Calls Out NVIDIA, Microsoft For Signing Open AI Letter, Asks Them To Open-Source CUDA And Microsoft Office

Anthropic researcher Julian Schrittwieser mocked NVIDIA and Microsoft for signing an open AI letter, calling on them to open-source CUDA and Microsoft Office. Schrittwieser highlighted the irony of historically anti-open-source companies now advocating for openness, while Anthropic itself has lobbied for restrictions on Chinese open-weight models.

Several tech firms have publicly stated https://officechai.com/ai/nvidia-ceo-jensen-huang-backs-open-models-in-first-post-on-x/ that they want open source AI to flourish, but Anthropic — the chief non-signatory of the letter — is calling out what could been seen as their hypocrisy. Julian Schrittwieser, a member of technical staff at Anthropic, posted a string of replies mocking the celebratory tone around the “Open Weights and American AI Leadership” letter. “I’m so excited that Jensen Huang is a believer in open source now, looking forward to the CUDA and GPU driver open source release ” he wrote, followed by a jab at Microsoft: “And satyanadella coming out for open source too, can’t wait for the open sourcing of Windows and MS Office ” He then addressed the reaction his posts were getting directly: “funny to see people jump to the conclusion I must want to ban open weight models.. I actually think open models can be very useful But it’s interesting how some historically extremely anti-open source companies are suddenly all in favor of openness.” It’s a sharp point, and it lands because it’s rooted in something real. NVIDIA’s CUDA platform is arguably the single biggest moat in the entire AI hardware stack, a proprietary software layer that locks developers into NVIDIA chips and that the company has guarded for close to two decades despite constant pressure to open it up. Microsoft’s flagship products, Windows and Office, remain some of the most closed, licensed, revenue-generating software on the planet, built by a company whose founder Bill Gates once wrote an open letter accusing early software hobbyists of piracy for sharing code freely. Both companies now stand as prominent, public signatories on a letter arguing that openness is essential to American technology leadership. Schrittwieser’s sarcasm writes itself. Where The Argument Holds Up Taken purely as a point about consistency, Schrittwieser is right that NVIDIA and Microsoft don’t practice what this letter preaches in their own core businesses. Neither company has ever shown appetite for open-sourcing the products that generate the bulk of their revenue, and there’s an obvious irony in watching a company built on one of the industry’s tightest proprietary moats sign a letter celebrating the virtues of openness elsewhere in the stack. If open weights are genuinely good for security, innovation, and diffusion, as the letter argues, the same logic should in principle apply to a driver stack or an operating system just as much as it applies to a language model. NVIDIA and Microsoft aren’t volunteering to test that logic on their own turf. The framing also fits a pattern: the signatories overwhelmingly skew https://officechai.com/ai/nvidia-ceo-jensen-huang-backs-open-models-in-first-post-on-x/ toward companies that don’t own the frontier model layer themselves, or that stand to gain from commoditizing the layer someone else currently dominates. NVIDIA sells chips regardless of whether the workload running on them is open or closed, and a world with more open-weight deployments is a world with more distinct buyers of GPUs rather than a handful of labs negotiating bulk compute deals. Microsoft, despite its OpenAI stake, has been governing its own AI strategy around reducing dependency on any single model provider. Signing a letter about open models costs neither company anything in their actual core business, which is precisely Schrittwieser’s point about the ease of this kind of advocacy. Where It Falls Apart The argument gets shakier once you look at what Anthropic itself has actually been asking for, which Schrittwieser’s framing leaves out entirely. Anthropic hasn’t just declined to champion open weights — it has actively lobbied for restrictions on the open-weight models coming out of China. CEO Dario Amodei has repeatedly pushed https://officechai.com/ai/chinese-ai-models-are-optimized-for-benchmarks-instead-of-real-world-use-anthropic-ceo-dario-amodei/ for tighter chip export controls to slow Chinese AI development, and Anthropic has been vocal in Washington about restricting the sale of advanced chips to China https://officechai.com/ai/anthropic-talks-about-regulation-and-ai-safety-far-more-than-openai-shows-data/ specifically to keep American compute dominance intact. More recently, Anthropic has pushed the US government https://officechai.com/ai/us-threatens-sanctions-against-chinese-models-that-are-allegedly-distilling-us-models/ toward sanctioning Chinese labs like Moonshot AI over allegations of distilling Anthropic’s own models to build competing open-weight systems. That’s a meaningfully different position than simply not open-sourcing your own product. NVIDIA never openly asked regulators to ban Linux, restrict access to GCC, or sanction companies for building on open-source infrastructure that competed with CUDA. Microsoft, for all its historic hostility https://officechai.com/ai/amid-debates-about-open-source-ai-heres-how-steve-ballmer-had-once-called-open-source-a-cancer-in-2001/ toward open source in the Ballmer era, never lobbied Washington to legislate against the Linux Foundation or Apache. Both companies kept their own products closed while leaving the broader open ecosystem around them alone to develop as it wished, including tools that directly undercut their own businesses. Anthropic’s position is closer to a company that keeps its own product closed while also trying to get its open competitors slowed down or shut out through export controls and sanctions. That’s a materially different thing from just declining to open-source your own work, and it’s the part of Schrittwieser’s own company’s record that his posts don’t address. There’s also a layer of self-interest baked into Anthropic’s posture that’s worth being upfront about. Anthropic’s entire valuation case https://officechai.com/ai/why-anthropic-could-find-it-hard-to-maintain-its-965-billion-valuation/ rests on enterprises continuing to pay a premium for API access to its models rather than switching to cheaper open alternatives once the capability gap narrows. A world where open weights, Chinese or otherwise, close that gap threatens the pricing power that justifies a near-trillion-dollar valuation. Wanting China restricted from catching up isn’t purely a national security position for Anthropic, even if it’s a defensible one — it’s also directly protective of the premium it charges. NVIDIA and Microsoft signing a letter that costs them nothing is one kind of inconsistency. Anthropic actively working to restrict the open competitors closest to catching it, while sitting out a letter that merely asks for open weights not to be prematurely restricted, is a different, and arguably more consequential, one.