Microsoft bets on open AI models, and the Azure logic speaks for itself

On 24 July 2026, Microsoft signed an open letter titled Open Weights and American AI Leadership together with Meta, Nvidia and more than twenty other companies. In it, the signatories advocate for broader adoption of open-weight AI models. Alongside the tech giants, investors such as Andreessen Horowitz and Y Combinator also appear under the letter, as do companies including Mistral, Hugging Face, IBM, Palantir and Perplexity.

The framing as a principled stance warrants a caveat: for Microsoft, it is also simply a good deal. More open models running on Azure means fewer purchases from OpenAI and Anthropic, two external providers on which Microsoft is still heavily dependent. That dependency costs money, and the figures make clear how much is at stake.

In total, Microsoft has invested approximately $13 billion in OpenAI since 2019, rising from $1 billion in 2019 through a second round in 2021 to $10 billion in 2023. Following a recapitalisation, Microsoft's stake in OpenAI Group PBC is estimated at around $135 billion on a diluted basis, representing roughly 27 percent. OpenAI, in turn, has contracted for $250 billion worth of Azure services.

In-house models as an alternative to costly procurement

In parallel with the open-weight lobbying effort, Microsoft is building its own model family. On 2 June 2026, the company announced seven models from the so-called MAI family at Build. These are already being deployed in products such as Copilot, where they are gradually replacing external models. The strategic rationale is straightforward: every query handled by a MAI model no longer needs to be paid for to OpenAI or Anthropic.

Independent benchmarks currently place the MAI models considerably below the models they are replacing. That finding, published on 9 June 2026, raises questions about the trade-off between cost savings and quality. Microsoft has not commented publicly on the matter.

The partnership with Mistral adds a third pillar. On 21 July 2026, the two companies announced a significant expansion of their collaboration. Microsoft is reported to have committed several billion dollars to leverage the Paris-based company's GPU infrastructure in Europe. Mistral models are available through Azure, broadening Microsoft's offering without the company bearing all of the development costs itself.

A partnership of growing complexity

The relationship between Microsoft and OpenAI dates back to 2016, when the two began working together. Since then, Azure has been OpenAI's primary cloud provider, and Microsoft's investment grew incrementally. On 28 October 2025, both parties signed a new definitive agreement, formally reaffirming the partnership.

That long-standing relationship has delivered considerable value to Microsoft: early access to GPT models, integration into Office and Azure, and a strong position in the enterprise AI market. But it also creates a structural dependency. As long as the most capable models come from external parties, those parties have a say in the margins and the terms.

The open-weight movement offers a partial way out. If powerful open models become more widely available, both customers and Microsoft itself can choose which model to run, on whatever infrastructure they prefer. For Azure, that is advantageous: the platform benefits from usage regardless of which model is running. Microsoft is therefore signing the open letter from a position in which the outcome the letter advocates directly serves its own cloud strategy.

What the coalition reveals about the broader sector

The list of signatories also says something about the wider balance of power in the AI sector. The fact that Y Combinator and Andreessen Horowitz are supporting the letter shows that influential venture capital firms view the open-weight model as favourable for their portfolios. Startups building on open models pay no licensing fees to model providers and retain greater control over their product development.

Companies such as Replit, Box and ServiceNow also appear under the letter. For them, the reasoning is similar: open models lower the barrier to integrating AI into products and reduce the vendor lock-in that comes with exclusive API dependency.

Moonshot AI, the Chinese developer behind the Kimi K3 and Kimi-K2 Thinking models, is not listed as a signatory, but illustrates how rapidly the open-weight landscape is evolving outside the American context. The letter explicitly focuses on American AI leadership, underscoring the geopolitical dimension of the debate without elaborating on it further.

For European players in the AI ecosystem, including Dutch funds, startups and policymakers, the combination of Microsoft's Mistral deal and the broader open-weight coalition offers a concrete point of reference. European model developers such as Mistral benefit from the capital inflows and distribution power of large cloud platforms, while the open-weight principle simultaneously aligns with European policy objectives around technological sovereignty and the accessibility of AI infrastructure.