Server processors were meant to be the boring corner of the AI boom. The exotic money went to graphics accelerators, while the central processing units sitting beside them were treated as a solved problem, cheap, plentiful, and interchangeable. That assumption is coming apart in China.
Intel and AMD have signed longer-term supply commitments with Chinese server customers as data centre CPU prices climb, according to Reuters, which cited two people familiar with the talks.
The contracts lock in purchase volumes rather than fixed prices, a structure that leaves buyers exposed if the shortage drags on.
China is a particularly tight corner of that market, partly because it has spent two years living inside Washington’s export controls.
The 💜 of EU tech
The latest rumblings from the EU tech scene, a story from our wise ol’ founder Boris, and some questionable AI art. It’s free, every week, in your inbox. Sign up now!
With access to the most powerful Western accelerators restricted, Chinese operators have leaned harder on the components they can still buy freely, and general-purpose server CPUs from Intel and AMD remain among them.
The scarcity has already produced some remarkable local pricing. Nvidia’s B300 servers have changed hands in China for close to $1 million, roughly double the going US rate.
The same curbs have pushed the country’s chip designers toward custom ASICs, though those do not remove the need for a conventional processor to run the system.
Chinese customers would typically commit to about a year of supply. This time, both chipmakers have discussed commitments of two years or longer with some buyers, the people said. Neither Intel nor AMD responded to requests for comment, and the customers were not named.
The particulars that would matter most are the ones that stayed private. Reuters’ sources did not disclose the volumes involved, the pricing formulas the contracts leave open, or how many buyers have signed.
What is clear is the direction of travel, and that it applies to Intel’s Xeon line and AMD’s EPYC parts alike.
The pricing moves are steep. Some server CPU products in China have posted month-on-month increases of more than 10%, and certain lines are up more than 40% since the start of the year, the sources told Reuters. It is the kind of curve normally associated with memory chips, not processors.
The cause is the same one reshaping the rest of the hardware market. The build-out of AI data centres has pushed demand well beyond Nvidia’s accelerators and into the surrounding components, memory, networking gear, and the server CPUs that orchestrate a rack.
Intel’s lead times for some Xeon parts have stretched to around six months.
The surge is not confined to China, and it is not confined to CPUs. A broader memory crunch has driven server component lead times out to 40 weeks and beyond as DRAM and high-bandwidth memory are diverted to AI systems, with contract manufacturers warning that the squeeze has now reached fully built servers.
Processors are simply the latest link in the same supply chain to seize up, and the volume deals in China read as an attempt to get ahead of it.
For Intel and AMD, the demand is a welcome problem. “Demand continues to run ahead of supply,” Intel chief executive Lip-Bu Tan told analysts in April, and the comment has aged into an understatement. AMD, for its part, expects the server CPU market to exceed $120 billion by 2030.
The volume-not-price structure of the new deals tells its own story. Buyers are willing to guarantee two years of orders without knowing what they will pay, a bet that supply, rather than cost, is the thing worth securing. Chipmakers, meanwhile, keep the freedom to reprice if the crunch holds.
It is a familiar posture for anyone who has watched Intel’s capacity expansion or the memory shortage of the past year. What is new is where it has spread.
When the least glamorous chip in the rack starts commanding two-year contracts, the shortage has stopped being about AI accelerators and started being about everything they need to run.
