AMD Bets $5B on Anthropic and OpenAI Plans $750B: The AI Compute Race Goes Gigawatt-Scale
AMD will invest up to $5B in Anthropic and supply 2GW of MI450 GPUs, while OpenAI raised its infrastructure plan to $750B through 2030 with a $20B Georgia data center. What the deals reveal: AMD challenging Nvidia, labs building instead of renting, and the rise of circular financing.
Table of contents
Two announcements on the same day made the scale of the AI compute race impossible to ignore. AMD said it will invest up to $5 billion in Anthropic, which will deploy up to 2 gigawatts of AMD Instinct MI450 GPUs. And OpenAI lifted its infrastructure spending plan to $750 billion through 2030, starting with a $20-billion data-center campus in Georgia. The headline isn't any single number — it's what these deals reveal about how the AI build-out now works.
AMD's move: cracking the Nvidia grip
For years, "AI compute" has been near-synonymous with Nvidia. The AMD–Anthropic deal is a real attempt to change that. Anthropic — which already runs some AMD Instinct MI355X chips — will scale onto 2GW of MI450-series GPUs, with the first gigawatt coming online in the first half of 2027. In return, AMD invests up to $5 billion in Anthropic, Anthropic uses Claude to optimize AMD's ROCm software stack, and AMD adopts Claude across its own engineering.
That last part is the tell: this isn't just a purchase order, it's a mutual dependency designed to make AMD a credible second source for frontier-scale training and inference. For a market desperate for alternatives in the chip race, a serious AMD challenger is arguably bigger news than the dollar figure.
OpenAI's move: building, not renting
OpenAI raised its planned compute spend to $750 billion through 2030 — up roughly 25% from about $600 billion earlier this year. The first piece, "Project Camellia," is a $20-billion campus on 1,400 acres near Savannah, Georgia, drawing at least 3.2 gigawatts of power, with total costs likely to top $30 billion. Notably, unlike its leased capacity from Oracle and AWS, this is OpenAI's first facility as principal designer and builder — a shift from renting compute to owning it.
The pattern worth noticing: circular deals
Step back and a structure emerges. A chipmaker invests in the AI company that buys its chips; AI labs commit to multi-gigawatt hardware while burning cash today. These "circular" arrangements — capital flowing between vendor and customer — are powering an extraordinary build-out, but they also concentrate risk and make the ecosystem's health hard to read from the outside. It's the supply-side counterpart to the market anxiety we covered when Alphabet and Tesla's stocks fell on soaring AI capex, and the capex squeeze reshaping IT budgets.
Bottom line
The compute arms race is entering a phase measured in gigawatts and hundreds of billions, with two consequential shifts underneath: AMD is finally a real challenger to Nvidia's dominance, and the biggest labs are moving from renting compute to building it. Both make the AI infrastructure story more competitive — and more capital-intensive — than ever. The open question is the same one hanging over the whole sector: whether the returns will eventually match the build.
Reporting: AMD Investor Relations, Tom's Hardware, CNBC, TechCrunch, Quartz. AMD's investment and GPU deployment figures are "up to" commitments; OpenAI's spending plan is a projection through 2030.
