“Will Data Center Investment Stall?” AI Boom Extends Semiconductor Shortage to CPUs as Power Grid Constraints and Policy Risks Converge

Intel, AMD Server CPU Supply Disruptions Drive Price Hikes and Delivery Delays
‘Post-Settlement’ Clauses Emerge in Prolonged Memory Shortage
Global Data Center Expansion Expected to Slow Amid Component and Power Grid Constraints

Intel and AMD, which together dominate the global central processing unit (CPU) market, are facing supply disruptions in server-grade CPUs. The semiconductor shortage triggered by the surge in artificial intelligence (AI) infrastructure investment is now spreading from high-performance graphics processing units (GPUs) and memory to CPUs. Market observers warn that the convergence of semiconductor supply constraints, unstable power availability, and mounting policy uncertainty could soon put the brakes on the ongoing expansion of data centers.

CPU Shortages Spread Across the Semiconductor Market

According to overseas media reports compiled on the 9th, Intel and AMD have recently notified major customers in China—including large cloud and platform companies such as Alibaba and Tencent—of CPU supply shortages. Reuters reported that prices of Intel’s server products in China have risen by more than 10%, while order backlogs have worsened, pushing delivery lead times out to as long as six months. AMD has also seen delivery cycles for some products extended to eight to ten weeks.

The primary driver of the shortage is the global boom in AI infrastructure investment. During its earnings call last month, Intel stated that “the rapid proliferation of AI has sharply increased demand for traditional compute CPUs,” adding that while inventories hit a record low in the first quarter of last year, the company expects supply conditions to improve from the second quarter of last year through this year as it responds aggressively. Intel thus expressed optimism that the CPU supply crunch, while real, would not become a prolonged bottleneck.

Industry sentiment, however, remains skeptical that the shortage will ease quickly. Intel is struggling to ramp up production due to manufacturing yield issues, while AMD faces constraints as Taiwan Semiconductor Manufacturing Company (TSMC)—which produces its EPYC processors—prioritizes AI chips in production line allocation. At the same time, the proliferation of high-performance AI agent systems is driving up CPU processing demand, further intensifying supply pressures.

Memory Market Structure Flips

The memory semiconductor market, which has suffered from supply shortages since the second half of last year, has effectively transitioned into a full-fledged seller’s market. This means that major global memory suppliers, including Samsung Electronics and SK Hynix, now hold pricing power. In practice, Samsung Electronics and SK Hynix are reported to have raised prices of server DRAM sold to cloud companies such as Microsoft, Amazon Web Services (AWS), and Google by as much as 70% compared with the fourth quarter of last year.

In addition, memory manufacturers have begun introducing “post-settlement” clauses into contracts. Under this arrangement, even if products are supplied at a fixed price for one year, any increase in market prices at the end of the contract period allows suppliers to charge customers the difference retroactively. There are reportedly no refund provisions if prices fall. A market source explained that “traditionally, annual contracts were signed with quarterly price adjustments within a certain range,” but “as AI-driven demand surged and price volatility widened, memory manufacturers appear to have changed contract terms to maximize profitability.”

This type of transaction structure has been extremely rare in the memory industry, which has historically operated as a commodity market. Commodity transactions typically follow a ‘push’ model, in which manufacturers mass-produce standardized products and supply them to the market in advance. However, as AI-driven growth has tightened memory supply, the market has flipped to a ‘pull’ model, with customers forced to wait for allocation from manufacturers. Some big tech companies are not only accepting post-settlement clauses but are also prepaying contract values in full in cash to secure priority allocation, intensifying competition for advance memory purchases.

Will the Data Center Investment Boom Cool?

Market participants increasingly warn that the pace of data center investment expansion may soon slow. With shortages emerging across key components, the unresolved challenge of securing sufficient power supply has become an additional constraint. According to a study released by Lawrence Berkeley National Laboratory last December, data center electricity consumption is projected to surge from 176 terawatt-hours (TWh) in 2023 to between 325 and 580 TWh by 2028. Yet power grids in major economies lack sufficient transmission capacity to meet this demand, with roughly 2,300 gigawatts (GW) of power capacity in the United States alone currently waiting for grid connection.

Cases of data center facilities failing to come online due to power supply difficulties have become increasingly common. A notable example is the data center project being pursued by Digital Realty and Stack Infrastructure in Santa Clara, California. Digital Realty applied to build a 48-megawatt (MW) data center in 2019 but has yet to complete power interconnection. Marsden Hanna, Google’s global head of sustainability and climate policy, stated at an American Enterprise Institute (AEI) event last month that “utilities in many regions are telling us it will take four to ten years to connect new data centers,” adding that “some have said it would take 12 years just to review grid connection requests.”

Industry stakeholders argue that expanding transmission capacity will require broader policy intervention. The challenge is that political momentum in several countries has recently shifted toward restricting rather than supporting data center investment. In the United States, both Democrats and Republicans have expressed concerns over the negative externalities of data centers. Senator Bernie Sanders has called for a nationwide halt to data center construction, while Florida Governor Ron DeSantis has warned that data centers could drive up energy prices. State governments are also growing reluctant, with Georgia, Vermont, and Virginia considering moratoriums on new data center construction. More recently, lawmakers in New York State have introduced legislation proposing a minimum three-year suspension on new permits for data center construction and operation.