Microsoft and Nvidia invest in Anthropic

Background

Training and running large AI models takes very large amounts of computing: specialised chips packed into data centres, together with the electricity and cooling needed to keep them running. For a company that builds models, how much of that capacity it can secure shapes what it can train and how many people it can serve. Arrangements between chip suppliers, cloud providers and model developers have therefore become a routine part of how the industry is financed and supplied.

What was announced

On 18 November 2025, Nvidia and Microsoft announced that they would invest in Anthropic as part of a new AI partnership.^[2] The two companies were expected to put in up to $15 billion in total.^[1][2] Nvidia's share was up to $10 billion and Microsoft's up to $5 billion.^[2]

On the other side of the table sat a purchase commitment. Anthropic agreed to buy $30 billion of compute from Microsoft Azure running on Nvidia systems.^[1][2]

The details

Anthropic's initial compute commitment includes up to 1 gigawatt of capacity built on Nvidia's Grace Blackwell and Vera Rubin systems.^[2] A gigawatt is a measure of power rather than of chips: data-centre capacity is often described in watts, because power supply and cooling set the limit on how much equipment a site can hold. Stating a commitment in gigawatts therefore says something about the size of the facility footprint behind it, not just the number of processors.

Beyond buying capacity, the two companies said they would work together on design and engineering, with the aim of optimising Anthropic's models and future Nvidia architectures.^[3] That is a two-way arrangement: the model developer tunes its software for the hardware, and the hardware designer learns in advance what the models of the next generation will demand of it.

One boundary was stated explicitly. Anthropic said Amazon remains its primary cloud provider and training partner.^[3] The Microsoft and Nvidia relationship therefore sits alongside an existing arrangement with a different cloud company rather than replacing it.

Why it matters

Three roles that are usually separate — chip supplier, cloud provider, model developer — overlap in this deal. Nvidia, which sells the hardware, is putting money into a customer that is committing to buy systems. Microsoft, which sells cloud capacity, is investing in a company that will spend on Azure, while that company keeps another cloud as its primary provider and training partner.^[1][2][3] The result is a set of commitments that tie the three together without making the relationship exclusive.

For Anthropic, the deal adds a large cloud supplier alongside its existing primary provider, and a direct line to the chip designer's engineering teams, at a stated scale of up to 1 gigawatt of capacity and a $30 billion purchase commitment.^[1][2][3] For the two investors, it attaches financing to demand: the money going in is connected to hardware and cloud capacity being bought on the other side.