AI Chip Shortage May Last Beyond 2028, Fed’s Williams Warns

New York Federal Reserve President John Williams warned Tuesday that the global supply chain for artificial intelligence hardware will remain strained for more than two years, a timeline that threatens to keep upward pressure on prices for semiconductors and the consumer and enterprise goods that rely on them.

The assessment marks one of the first times a senior U.S. central banker has explicitly linked the physical bottlenecks in AI infrastructure buildouts to broader macroeconomic inflation risks. Speaking with Bloomberg, Williams cautioned that even if demand stabilizes — an outcome he described as a significant assumption — manufacturing capacity for advanced processors, high-bandwidth memory, and specialized networking equipment cannot expand fast enough to close the gap before late 2028 at the earliest.

Fabricating the advanced chips required for AI training and inference depends on a highly concentrated supply chain dominated by TSMC and Samsung, utilizing extreme ultraviolet lithography machines produced almost exclusively by ASML. Expanding this capacity requires multi-billion-dollar fabrication plants that take three to five years to construct and certify. Packaging technologies such as TSMC’s CoWoS, which stacks memory directly next to processors to increase data throughput, have already proven to be a severe chokepoint throughout 2024 and 2025. The shortage extends beyond silicon to include liquid cooling systems, high-capacity transformers, and backup power generation units necessary to operate modern data centers.

Williams indicated that these persistent supply constraints factor into his view that the Federal Reserve may need to raise interest rates once more before the end of 2026 to ensure inflation continues its downward trajectory. He declined to forecast monetary policy moves into 2027, stating he currently sees no reason to believe further tightening would be necessary beyond this year.

Despite the prospect of higher borrowing costs, Williams expects capital deployment into artificial intelligence to remain largely unaffected. Investors backing AI development are operating with long time horizons, betting on transformative productivity gains across the economy that they believe justify current valuations regardless of short-term rate fluctuations. Monetary policy, in his view, will not deeply alter the trajectory of this specific investment cycle.

That price-insensitive capital flow is intensifying competition for finite hardware resources, particularly for companies attempting to pivot existing infrastructure toward AI workloads. Bitcoin miners, which operate massive facilities equipped with industrial-scale power connections and cooling systems, have increasingly sought to repurpose their sites for AI cloud computing. However, the prolonged equipment shortage complicates these transitions, forcing miners to compete directly with hyperscalers like Microsoft, Amazon, and Meta for limited allocations of GPUs and networking gear.

The financial stakes for these pivots have escalated rapidly. Mining firm IREN secured $2.8 billion in new AI cloud contracts in July, while Anthropic recently finalized a $9 billion compute agreement with Riot Platforms. Yet securing the contracts does not guarantee access to the underlying hardware needed to fulfill them. Lenders financing these expansions have grown increasingly selective, scrutinizing whether mining firms can actually procure the necessary chips and power equipment within contractually obligated timelines.

The Federal Open Market Committee convenes next in early November, where policymakers will weigh Williams’ inflation concerns against recent labor market data to determine whether the anticipated final rate hike materializes before year-end.