Anthropic’s Microsoft Deal Reflects A Bigger AI Infrastructure Shift

Source Domain: www.forbes.com

Digital background depicting innovative technologies in (AI) artificial systems, neural interfaces and internet machine learning technologies

getty

One of the biggest headlines right now in AI has to do with a company that has scaled impressively over the last couple of years, becoming a household name for its agentic models, and even going toe-to-toe with one of its biggest U.S. government clients over AI safety concerns.

From its humble beginnings as the product of a couple of twins, formerly employed by OpenAI, Anthropic has mushroomed into a market force to be recognized. Two years ago, the company was valued between $4 billion and $15 billion: now, it’s estimated to be worth over $350 billion, and everyone knows who Claude is.

So it’s making news that Anthropic is making a big play in using Microsoft compute for its operations, to the tune of $30 billion in Azure infrastructure services.

“Anthropic is in talks to rent servers powered by Microsoft-designed AI server chips, a coup for the software giant,” report Aaron Holmes and Qianer Lu for The Information.

Why Use Microsoft?

One question you might ask is, why wouldn’t Anthropic, in its expansion push, just use Nvidia hardware in-house like so many other companies?

I asked GPT, and this is what I came up with (I’m paraphrasing);

cost control: NVIDIA GPUs are not only expensive, they’re also supply-constrained.
diversification: AI labs do not want total dependence on one vendor or supply chain.
cloud integration: Microsoft’s cloud ecosystem might offer specialized features and support custom services when it comes to large-scale AI serving workloads.
negotiating leverage: Using alternatives gives companies leverage in pricing and partnerships with NVIDIA.
vertical optimization: Anthropic may tune models around particular hardware architectures for performance-per-watt advantages.

Number four is interesting, in that I wonder whether that was part of the strategy at all. Elsewhere on the web, Ashley Capoot…

Source