Nadella's Anthropic Attack: The Open-Source Shield for a Closed-Source Empire
The code does not lie, but the market does. Satya Nadella's recent characterization of Anthropic's model restrictions as 'illogical' is not a technical critique—it is a strategic maneuver by the CEO of a company that has spent $13 billion to secure exclusive access to the world's most popular AI model. The attack, reported by Crypto Briefing, a publication that usually tracks blockchain assets, reveals a deeper fault line in the AI industry: the conflict between closed-source safety and closed-source monopoly, dressed in the clothing of competition. As someone who has spent a decade auditing smart contracts and tokenomics, I recognize the pattern—the same 'openness' rhetoric used by centralized exchanges to defend their order books, or by Layer-2 networks that promise decentralization while maintaining admin keys. The code does not lie, but the founders do.
On February 26, 2025, during a panel at the AI Action Summit in Paris, Microsoft's CEO declared that Anthropic's decision to restrict how its Claude models can be used is 'illogical' and 'anti-competitive.' He argued that such limitations reduce diversity in the AI ecosystem and ultimately concentrate power. The statement was widely covered by tech and financial media, but Crypto Briefing's readership, accustomed to parsing trust assumptions in decentralized protocols, may see the cracks. Anthropic's restrictions, detailed in its Commercial Terms of Service, prohibit using Claude to compete with Anthropic, limit deployment in high-risk scenarios without additional compliance, and forbid training competing models on Claude's outputs. These terms are similar to those used by OpenAI—except OpenAI's terms are enforced primarily on API users, while Anthropic also restricts certain on-premise deployments. Nadella's criticism conveniently ignores that Microsoft's own agreement with OpenAI grants it exclusive rights to run OpenAI's models on Azure, effectively making Microsoft the only cloud provider for GPT-4 Turbo and future models.
Let's engage in a systematic teardown. First, the technical reality: model restrictions are not inherently anti-competitive—they are a safety control mechanism. Anthropic was founded by former OpenAI researchers with a mandate to build 'constitutional AI' systems that minimize harm. Its restrictions are part of its safety stack: by controlling how the model is accessed and fine-tuned, Anthropic can perform red-teaming, monitor for abuse vectors, and enforce usage policies. In my audit experience, when a protocol imposes strict access controls on its core contracts (e.g., only whitelisted addresses can call a mint function), we do not call that anti-competitive—we call it security. Nadella's framing conflates safety with monopoly. Second, the incentive alignment: Microsoft's business model depends on Azure's lock-in. By criticizing Anthropic, Nadella encourages enterprises to choose a 'diverse' mix of models—but he expects them to run those models on Azure through AI Studio. The 'openness' he advocates for is a gateway product to cloud compute. The gas fees don't lie: every inference call on Azure either goes to Microsoft-owned compute or a partner they control. The real competition isn't between models—it's between clouds.
Now, the contrarian angle. What did the bulls get right? Nadella has a valid point: excessive restrictions can stifle innovation. If Anthropic's terms prevent a startup from fine-tuning Claude for a medical diagnosis tool because of the 'high-risk' clause, that startup may simply abandon the idea, reducing societal benefit. Moreover, model diversity is critical infrastructure—relying on a single provider (OpenAI or Anthropic) creates single points of failure, as we saw when OpenAI suffered outages in late 2024, halting thousands of applications. Nadella's call for diversity is a genuine market need. But his solution—more openness—is hypocritical when his own platform enforces vendor lock-in through proprietary APIs and exclusive inference rights. The contrarian insight is this: both sides are right and wrong. Anthropic's restrictions may be too broad, but Microsoft's 'openness' is a Trojan horse for Azure dominance. The real winner is the open-source community, led by the Llama and Mistral models, which offer both freedom and control. In my analysis of tokenomics, the most sustainable systems are those where value accrues to users, not intermediaries. Open-weight models with permissive licenses (Apache 2.0, MIT) offer that—but they lack the safety guardrails that enterprise clients require.
The takeaway is an accountability call. The rug was pulled before the mint even finished—in this case, the 'mint' is the AI market, and the 'rug' is the illusion of choice. Nadella's statement is a deliberate distraction from the cozy duopoly that Microsoft and OpenAI have built. Meanwhile, Anthropic's restrictions are a liability that will limit its market share. The industry needs a third path: models that are inspectable, immutable, and verifiable on-chain, where license terms are enforced by smart contracts rather than legal agreements. Until then, the debate between 'open' and 'closed' is a luxury that only the incumbents can afford. For the rest of us, the choice is not between freedom and safety—it is between two forms of captivity. The code does not lie; only the founders do.