Procurement costs for server CPUs have suddenly become unpredictable. The AI boom's component shortages were first discussed in terms of GPU scarcity, then spread to reports of tight supply for HBM (high-bandwidth memory) and networking equipment. Server CPUs, however, had long been regarded as components that customers "had no trouble procuring." That assumption is now crumbling, according to a Reuters report from July 23, 2026. Intel and AMD are reportedly shifting Chinese server customers toward long-term contracts that lock in purchase volumes while leaving prices unfixed. In China, some server CPU products have already seen prices rise more than 40% since the start of the year, with some products climbing at a pace exceeding 10% per month. Volume is promised, but price is not—let's unpack what this contract structure means by looking at the broader supply chain picture.
A Contract Design That Locks In Volume But Not Price
According to Reuters' report detailing the contracts, most agreements run for about a year, with some negotiations reportedly underway for supply arrangements spanning two years or more. The sources who spoke on the matter—two people described as "not authorized to speak to media"—declined to be named, and neither Intel nor AMD responded to requests for comment.
What Reuters confirmed is the basic structure: purchase volumes are fixed while prices are not. Details such as how far requested volumes are actually guaranteed, or whether any price ceiling exists, remain unclear. Even so, a structure that fixes only volume while letting prices float works to shield the supplier—Intel and AMD—from demand-fluctuation risk during a period of rising prices, while shifting price-fluctuation risk onto the Chinese customers.
There are also points the report leaves unaddressed. The names of the specific companies involved in the contract negotiations, as well as the actual price levels behind the reported increases, have not been disclosed. The report also does not touch on how the ongoing US export restrictions on China might factor into transactions involving general-purpose server CPUs (Xeon and EPYC). The fact that only the existence of the contracts has been reported, with specific counterparties and price levels withheld, may reflect the suppliers' reluctance to reveal their hand to other customers still in negotiations.
The Spread of Demand Pressure: From GPUs to Memory to CPUs
The semiconductor supply crunch began with the AI boom starting in 2023, first manifesting as a shortage of NVIDIA-family GPUs. Reports of tight supply for HBM (high-bandwidth memory) and networking equipment followed, and by 2026 the supply constraints have spread even to server CPUs, which had previously been relatively easy to procure. Intel CEO Lip-Bu Tan said at the April earnings call that demand "continues to run ahead of supply," specifically citing tight supply of Xeon server CPUs.
AMD has projected that its server CPU revenue for the second quarter of 2026 will grow more than 70% year-over-year, with AMD's CFO (Chief Financial Officer) attributing the growth to expanding demand for agentic AI. The procurement difficulties that started with GPUs have now spread to the general-purpose CPU segment that underpins AI workloads. This growth rate is also, in its own way, evidence that customers have not stopped buying even as prices keep climbing.
Server CPUs are no different from GPUs in that they use near-leading-edge manufacturing processes—AMD EPYC at the 3-4nm class, Intel Xeon 6 on Intel 3—but until now they were seen as less caught up in the scramble for manufacturing capacity than AI accelerators, and easier to address through increased production. What these contracts reveal is a shift in that very premise: production headroom for CPUs, too, has begun to feel the effects of AI-related demand.
AMD's Market Opportunity Forecast Doubled From $60 Billion to Over $120 Billion—Who Bears the Risk?
At its Financial Analyst Day investor presentation in November 2025, AMD projected a $60 billion TAM (total addressable market) for the data center CPU market in 2030. By the time of its Q1 2026 earnings announcement, that figure had been raised more than twofold, to over $120 billion. At an exchange rate of 1 USD = 163.06 JPY (as of July 23, 2026), $120 billion converts to roughly ¥19.6 trillion or more. This figure is not a third-party market research estimate but AMD's own calculation of the total market size (TAM), and it is not AMD's own revenue forecast. Still, the fact that the company more than doubled the figure in under half a year shows just how strongly it now expects AI demand to expand.
If a company expects the market to keep expanding over the coming years, locking in prices early would mean giving up future gains from price increases. Contracts that fix only volume, and not price, directly reflect the supplier's expectation of growing demand. The fact that AMD sharply raised its TAM forecast at its Q1 2026 earnings in May 2026 aligns with the demand outlook underlying this contract design.
If the pace of price increases exceeding 10% per month were to continue for a full year, compounding would put the annual rate of increase above 200%. Given that most contract terms run for about a year, if such a pace of price increases continues, the burden on customers who have committed to volumes could end up far exceeding their initial expectations. There is no guarantee that actual price increases will continue at the same pace going forward, but this calculation illustrates just how much cost-fluctuation risk a volume-only commitment contract can carry.
The parties who stand to benefit from this contract structure are Intel and AMD. Locking in volume makes revenue forecasting easier, while leaving price unfixed preserves room to pass on further price hikes. Meanwhile, the parties bearing the cost are Chinese cloud providers and internet companies, with smaller server makers—those with weaker price-negotiating power—particularly exposed. A similar pattern was seen during the 2023-2024 NVIDIA H100 GPU shortage, when major cloud providers who had secured multi-year advance purchase contracts early on obtained supply first, while later buyers without such contracts were left to absorb both price increases and delivery delays. If contract prices end up exceeding market rates once supply normalizes, Chinese customers could once again be left holding the risk of having bought at inflated prices.
Favorable Treatment for Google, Volume Restrictions for China: A Contrast in Allocation
On April 9, 2026, Intel officially announced a multi-year partnership with Google for AI and cloud infrastructure, combining Xeon CPUs with custom IPUs (Infrastructure Processing Units—dedicated chips handling data center infrastructure processing). This partnership involves technical collaboration with application-specific design work, extending beyond Xeon supply into co-designing custom IPUs. Lip-Bu Tan cited this partnership as an example of addressing the situation through long-term contracts during the same April earnings call. The contrast is striking: major Western customers are offered multi-year contracts involving technical collaboration, while the China-related contracts reported months later reportedly involve only volume commitments.
Despite China accounting for more than 20% of Intel's revenue (according to reports as of February 2026), the price-related risk remains with the customer. This structure echoes the supply gap that emerged between early contractors and later buyers during the 2023-2024 NVIDIA H100 shortage. The differing content of these contracts can also be seen as a manifestation of decisions about which customers receive priority allocation of components in a supply-constrained environment.
While the Google partnership was announced jointly, with some technical specifications made public, the China-related contracts have only been reported through anonymous sources, and even the identity of the customer companies remains undisclosed. This gap in how much of the contract details are made public alone hints at how the supplier positions each camp differently. There is also a clear disparity in the degree of information disclosure between the two sides.
Ripple Effects Reaching Japan's Supply Chain, and What to Watch Next
The reasons the CPU shortage has surfaced this late apparently differ between Intel and AMD. On AMD's side, the cause is said to be that TSMC, which handles leading-edge logic manufacturing, is prioritizing production of AI-related chips, squeezing the manufacturing capacity allocated to server CPUs. On Intel's side, the issue is believed to stem from challenges with in-house manufacturing yield. Though the causes differ, both point to the same underlying trend: shrinking production headroom for general-purpose server CPUs. As of February 2026, reports indicated that lead times for some Intel server CPU products had stretched to as long as six months, while some AMD products faced lead times of 8 to 10 weeks. Nearly half a year later, rather than improving, the situation has only taken on a new form—surfacing now as price increases and long-term contracts—a development that should not be overlooked.
Volume-committed long-term contracts allow suppliers to lock in gains from price increases while avoiding risk, but for buyers, they are also one of the few means available to secure priority supply amid the scramble for production capacity. Even if buyers choose these contracts out of self-preservation, the fact remains that as long as prices are not fixed, it is ultimately the Chinese customers who bear the resulting cost increases.
This supply chain is not confined to the Chinese market. In Japan too, cloud providers such as Sakura Internet, along with server makers like Fujitsu, NEC, and Hitachi, procure components from the same global supply chains operated by Intel and AMD, meaning that if the current tightness persists, it could ripple through to delivery times and prices in Japan as well. During periods of supply constraint, customers with existing trading relationships tend to be prioritized, and buyers with weaker price-negotiating power are more likely to be forced into accepting unfavorable terms. The degree to which this kind of contract structure affects Japanese companies is likely to depend on the nature of the trading relationships they have built with Intel and AMD.
As long as this contract structure—locking in volume while leaving price unfixed—persists, it is ultimately Chinese customers who will bear the burden of price increases. Intel and AMD, for their part, avoid bearing demand-fluctuation risk, and compared to the multi-year contracts involving technical collaboration offered to major Western customers like Google, the contracts for China stand out precisely for their focus on volume restrictions alone. If this asymmetry is to be resolved, it would likely require one of the following: TSMC easing its production capacity allocation on AMD's side, Intel improving its manufacturing yields, a resulting recovery in supply headroom, or a shift in negotiations toward contracts that include price-linked clauses.