Microsoft (MSFT) CEO Satya Nadella launched a sharp internal critique of artificial intelligence startup Anthropic this week, telling engineers that the company’s high-end Fable model imposes unreasonable restrictions on user requests and that the current industry structure where a handful of players control AI computing capacity makes no economic sense.
Speaking to employees working on Microsoft’s Copilot AI software on Wednesday, Nadella questioned the logic behind Fable’s frequent refusals to answer seemingly harmless queries. “If you use Fable, when it refuses for any random thing, it just is like, when was the last time you had a creation tool that was so editorially controlled?” Nadella said, according to a copy of his remarks obtained by CNBC. “It doesn’t make sense.”
The comments mark a notable moment of tension between Microsoft and one of its most important partners. In November, Microsoft committed a $5 billion investment to Anthropic, while the startup agreed to spend $30 billion on Microsoft’s Azure cloud platform. Earlier this year, Microsoft also launched Copilot Cowork, a business productivity assistant built around Anthropic’s models. Microsoft declined to comment on Nadella’s remarks. Anthropic did not immediately respond to a request for comment.
Anthropic’s Fable model has faced growing scrutiny from users on social media who complain that the system rejects too many normal requests. According to Anthropic’s own support documentation, when end users ask Fable about certain topics—including aspects of building large-scale AI models, cybersecurity, biology, chemistry, or model distillation—the platform may automatically route those queries to an older version of the model, specifically Claude Opus 4.8, rather than providing a response from the current version.
The restrictive safeguards stem in part from a turbulent launch. Anthropic introduced Fable 5 in early June and initially claimed it was reducing false positives on blocked requests. However, just three days after the launch, the company was forced to suspend access to comply with a U.S. government export control directive. When Anthropic restored the model on July 1, the company acknowledged that “the new safeguards will flag a slightly higher fraction of harmless requests than the previous Fable safeguards.”
Nadella’s criticism extended beyond Fable’s content moderation to the broader economics of the AI industry. “It can’t be that there are only two companies in the world with token capital, and everybody else is renting it,” he told the engineers. “It makes no economic sense.” Tokens are the units used to measure computing usage of AI models.
The Microsoft chief has been building this argument publicly for weeks. In a blog post published Sunday, he invoked Palantir (PLTR) CEO Alex Karp, who previously told CNBC that technical organizations “want to know they own the means of production.” Nadella has argued that enterprises should be able to cost-efficiently develop custom models using their own data without handing valuable proprietary information to foundation model developers.
Microsoft’s own positioning in the AI landscape has grown increasingly complex. The company remains deeply tied to OpenAI through a series of investments, but the relationship has frayed since the abrupt 2023 ousting and reinstatement of OpenAI CEO Sam Altman with little notice to Nadella. OpenAI said in April it would bring its models to Amazon Web Services (AWS), expanding beyond Azure. Microsoft’s stake in OpenAI’s for-profit business was valued at $135 billion as of October.
Meanwhile, Microsoft has been expanding its own AI portfolio. The company now offers more than 11,000 models through its Foundry service, including offerings from both Anthropic and OpenAI, and announced in June that it was building its own in-house coding model. Nadella also recently said it was good that Microsoft is merging products for consumer and corporate workers, a unification he said “we should have done maybe day one.”
The competitive landscape is intensifying. Chinese startup Moonshot AI announced an open-source model on Thursday that it claimed outperforms recent releases from both Anthropic and OpenAI. Anthropic’s Claude Code software development tool has meanwhile become popular among programmers, helping the company’s valuation surpass that of OpenAI. Anthropic reported its annualized revenue climbed from roughly $9 billion at the end of 2025 to more than $30 billion, exceeding OpenAI’s annualized revenue run rate of about $24 billion.
Investor concerns about Microsoft’s position in the AI race have weighed on the stock. Microsoft shares have fallen 17% year-to-date, a stark contrast to the Nasdaq Composite’s 11% advance over the same period. Investors are pricing in the risk that Microsoft’s Azure cloud economics could erode if enterprise customers find cheaper alternatives for AI workloads, even as the company allocates tens of billions of dollars per quarter to data center expansion.
Anthropic has faced its own challenges beyond user complaints. The company was designated a supply-chain risk by the Pentagon after it refused to allow its models to be used for autonomous weapons or domestic surveillance—a label Anthropic has called legally unsound and challenged in court. The company is also reportedly moving forward with plans for an initial public offering, with Bloomberg reporting that Anthropic will meet with investors and banks in the coming weeks and is looking to list its shares as soon as October.
Nadella’s internal remarks underscore a fundamental tension in the AI industry: as frontier model developers impose increasingly strict safeguards to address safety concerns and regulatory requirements, enterprise customers and partners are pushing back against what they see as overreach that limits practical utility. “When was the last time you had a creation tool that was so editorially controlled?” Nadella asked. The question, directed at his own engineering team, signals that Microsoft intends to position itself as a champion of more open and customizable AI access—even if that means publicly criticizing a company in which it has invested billions.