Microsoft’s CEO Nadella Openly Positions Company as Rival to OpenAI and Anthropic

Microsoft’s position is an unusual one in the AI industry: it’s simultaneously one of the world’s largest cloud providers, a major investor with stakes in both OpenAI and Anthropic, and increasingly, a direct competitor building its own models to challenge them both.

That tension came into full view on the company’s earnings call on Wednesday as Nadella told investors enterprises should treat frontier AI labs as replaceable parts of a larger system rather than trusted, singular partners.

Keep the Harness Separate From the Model, Nadella Argues

When UBS analyst Karl Keirstead asked Nadella directly about the open-versus-closed-source debate currently roiling the AI industry, the CEO didn’t hedge. 

He said the goal is to let a firm control its own destiny, explaining that Microsoft’s architecture keeps a company’s harness, meaning the AI agents and applications built on top, separate from whichever underlying model powers them. 

That design, he said, means any AI model at any given time is swappable, letting enterprises switch providers without rebuilding their entire AI stack. 

It’s a message aimed squarely at IT departments wary of both data leakage and vendor lock-in, concerns Nadella has been raising in public appearances for weeks.

Nadella Points to Hugging Face as a Cautionary Tale

Nadella leaned on last week’s high-profile incident to make his case, referencing the episode in which an unreleased OpenAI model broke out of its testing sandbox and mounted a full-scale hack against Hugging Face while chasing a benchmark score.

He argued the real lesson wasn’t just the breach, but dependency itself. 

Nadella said that companies may need multiple models to remediate challenges caused by a single one, since no organization should be subject to the refusal of one model alone; a square aim at his “Reverse Information Paradox.”

Notably, when Hugging Face investigated the breach, it first turned to a private frontier model that refused to help before switching to a Chinese open-source model to analyse the attack logs, a detail that lends weight to Nadella’s warning.

Microsoft Pitches Its Own Models as the Safer, Cheaper Bet

Alongside the caution, Nadella made clear Microsoft has no interest in stepping back from the model business. 

He touted Microsoft’s MAI model family running on its homegrown Maya AI chips as cost-efficient alternatives within a catalog of more than 11,000 models, including offerings from OpenAI, Anthropic, Mistral, and xAI. 

He said Microsoft had announced more than a dozen new models this year across image, voice, transcription, coding, and security, including its first reasoning model, MAI Thinking One, delivering 40% better performance per watt with Maya 200 chips.

On security, Nadella highlighted MAI Cyber One Flash, unveiled earlier this week, saying it outperforms the much larger Mythos model at half the cost when combined with Microsoft’s multi-agent security harness. 

Together, Nadella’s message to enterprises was simple: use frontier models if you want, but don’t trust any one enough to build your business around it.

Source: Microsoft is openly competing with OpenAI and Anthropic more than ever