---
title: "Anthropic will deploy up to 2 gigawatts of AMD's new AI chips"
description: "AMD and Anthropic announced a strategic partnership to deploy up to 2 gigawatts of AMD's next-generation Instinct MI450 Series GPUs. It is a large vote of confidence in AMD as a credible alternative to Nvidia for training and running frontier AI, and a marker of just how much power the AI build-out now consumes."
category: "Tech"
category_url: https://boursel.com/category/tech
author: "Hannah Blackwood"
published: 2026-07-22T16:16:00.000Z
updated: 2026-07-22T16:16:00.000Z
canonical: https://boursel.com/article/anthropic-will-deploy-up-to-2-gigawatts-of-amd-s-new-ai-chips
tags: ["amd", "anthropic", "ai-chips", "data-centers"]
---
# Anthropic will deploy up to 2 gigawatts of AMD's new AI chips

AMD and Anthropic announced a strategic partnership to deploy up to 2 gigawatts of AMD's next-generation Instinct MI450 Series GPUs. It is a large vote of confidence in AMD as a credible alternative to Nvidia for training and running frontier AI, and a marker of just how much power the AI build-out now consumes.

The chips that train artificial intelligence are the industry's real bottleneck, and on July 22 one of the leading AI labs placed a large bet on a supplier other than the market leader. AMD and Anthropic announced a strategic partnership to [deploy up to 2 gigawatts of AMD Instinct MI450 Series GPUs](https://ir.amd.com/news-events/press-releases), AMD's next-generation AI accelerators. (A disclosure: Anthropic makes the AI model used in producing this coverage; we report the deal from the companies' own announcement.)

## What was actually announced

The confirmed substance is straightforward and significant. This is a partnership under which Anthropic will deploy AMD's forthcoming MI450 accelerators at very large scale, measured not in a number of chips but in the power they draw: up to 2 gigawatts of computing capacity.

Measuring a chip deal in gigawatts has become the industry's shorthand, and it tells you where the constraint now sits. Two gigawatts is a serious amount of electricity, comparable to the output of a couple of large power stations. When AI infrastructure is sized in gigawatts, the message is that the limiting factor is no longer just the availability of chips but the power and data-center capacity to run them.

## Why it matters for AMD

For AMD, landing a frontier lab like Anthropic is strategically important well beyond this single deal. The market for AI training chips has been dominated by Nvidia, whose GPUs have been the default for building large models. AMD has spent years positioning its Instinct line as a genuine alternative, and a public commitment from a top AI developer to deploy its next-generation MI450 at gigawatt scale is exactly the kind of validation that argues its hardware can do the job.

The competitive stakes are large. Every gigawatt of compute that a major lab commits to AMD is capacity it is not buying from Nvidia, and in a market where demand for AI accelerators has run far ahead of supply, credibility with a marquee customer can reshape how buyers, and investors, view the number-two player.

## Why it matters for Anthropic

For Anthropic, the logic is about securing supply and diversifying it. Training and serving frontier models requires enormous, sustained access to high-end accelerators, and any lab that relies on a single supplier is exposed to that supplier's pricing, roadmap and allocation decisions. Committing to a second major chip source is a way to reduce that dependence and to lock in future capacity in a market where compute is the scarce input.

It also fits a broader pattern across the AI industry, in which the leading labs are signing long-dated agreements for chips, power and data-center space years ahead of need, effectively pre-committing to the infrastructure they expect their models to require.

## The bigger picture

Strip away the specifics and the deal is a data point about the scale and cost of the AI build-out. The competition among AI labs is increasingly a competition for physical resources: advanced chips, the fabs that make them, the data centers that house them, and above all the electricity to power them. A single partnership sized at up to 2 gigawatts underlines how quickly those requirements are growing.

For investors, the read-across is twofold. First, AMD is establishing itself as a real second source in AI accelerators, which matters for the competitive dynamics, and the margins, of the whole chip sector. Second, the relentless framing of these deals in gigawatts is a reminder that the AI story is now inseparable from an energy story: the constraint, and much of the cost, is increasingly power. We have not verified the reported financial terms of the arrangement, and have left any dollar figure out rather than state one we cannot confirm; the deployment scale and the partnership itself come directly from AMD's own announcement.