When we think about the big players in artificial intelligence, we tend to draw pretty clear lines between competitors and allies. Anthropic y Google They usually appear on the same board, yes, but as direct rivals that develop their own models and compete for the same ground. Therefore, the fact that they now appear linked in the same agreement draws attention from the first moment. The firm led by Dario Amodei has closed an alliance with Google and Broadcom to ensure next-generation computing capacity, and that movement, beyond the technical, leaves a message that does not go unnoticed.
If we go to the details of the announcement, what is relevant is not only who participates, but the scale of what has been signed. Anthropic speaks of multiple gigawatts of next-generation TPU capacity that it expects to come online from 2027, an infrastructure designed to support its famous Claude models. In its statement it insists that demand from its clients has accelerated this year, and presents this movement as a direct response to that pressure. In fact, it describes it as its biggest bet in computing so far, although Amazon remains its main cloud provider.
The unexpected partner in the battle for computing
The agreement makes a lot of sense if we look at the figures that the company has shared. In 2024, it registered annualized revenues above $30 billion and more than 1,000 business clients exceeding one million annual spending, when in February there were more than 500. So this undoubtedly translates into a greater load on your infrastructure. And that’s where this movement fits in, not so much as an isolated strategic coup, but as a response to that growth.
And, as we can see, this agreement has two different pieces. On the one hand there is Broadcom, a semiconductor company that has benefited greatly from the rise of AI. On the other hand, the Mountain View giant appears, which in addition to providing infrastructure, driven by its focus on TPU, also competes directly in the development of models. And that is where the agreement gains interest, because it mixes technical collaboration with a competitive relationship that already existed.
It is also worth stopping at where Anthropic is, because it helps to understand why it can close such a deal. The company has been building its position by moving away from the race for the flashiest features and focusing on business environmentwhere security, control and reliability outweigh the initial impact. This approach has allowed him to excel in tasks such as programming, with Claude Code, and security with the new Mythos. And, little by little, it has been gaining something that is not achieved overnight: the trust of large companies.
But there is more. Anthropic makes it clear that Claude runs on Trainium from AWS, TPU from Google and GPU from NVIDIA, and adds that this variety allows it to improve performance and resilience. That gives us a pretty clear clue about what he’s doing now. Rather than betting everything on a single supplier or a single family of chips, it is consolidating a more flexible base to sustain its growth. And in an industry so stressed by hardware demand, that decision makes a lot of sense.
Images | Anthropic
In WorldOfSoftware | The “token economy” is broken: flat AI programming fees are mathematically unsustainable
