The model company wants a say in the silicon
A lab hiring chip designers and signing compute with a rocket company are the same decision viewed from two angles: the supply chain stopped being something you can simply buy from.
Anthropic is building a team to co-design custom AI chips with its models. Separately it raised usage limits alongside a compute deal with SpaceX.
Co-design is the operative word
This is not an attempt to compete with NVIDIA on general-purpose accelerators. It is a bet that knowing your own architecture lets you specialise silicon around it, and that the efficiency is worth owning rather than renting.
Google's TPU programme is the precedent that worked: years of nothing, then enormous advantage. The counter-precedent is every other company that decided custom silicon was cheaper than buying and learned the schedule is measured in generations.
The counterparty tells you more than the deal
Compute supply used to mean a hyperscaler or a colocation operator. A launch company on that list reflects how far the definition of infrastructure has stretched — power, land, cooling and capital are the inputs, and a lot of firms outside computing have all four.
For customers the effect is unglamorous and welcome: rate limits doubling on the plans that actually run agents. Long-horizon agentic work is capacity-hungry in a way chat never was, and the ceiling has been the real constraint on what teams could deploy — which is exactly what the enterprise packaging is built to sell into.
Read it against the selloff
Six suppliers lost $100 billion each in a week without a bad quarter between them. When the market is that unsure about the demand curve, and forecasts for 2028 nearly triple in a single quarter, a buyer taking control of its own roadmap is not empire-building.
It is a company that has looked at its supply chain and concluded the risk is no longer priced.
TechCrunch — Anthropic is hiring an AI chip design team → · Anthropic — Higher usage limits for Claude and a compute deal with SpaceX → · CNBC — Chip stocks shed more than $1 trillion as selloff hits companies powering AI boom →