BLOG — July 28, 2026
Behind the AI boom: The electronics supply-side constraints
By Chris Rogers
Discussion of the Artificial Intelligence boom has centered on demand: tokens generated, models trained and use cases emerging. The more consequential issue sits upstream, across power infrastructure, memory components and specialty materials. This is where the tightest bottlenecks are forming, and where pricing and availability pressures are likely to persist over the next 18 months.
Electrical components: The building bottleneck before the silicon bottleneck
Before AI accelerator capacity can be deployed, the underlying data center infrastructure must be built and energized. Transformers needed to step down grid electricity face lead times of several years for the largest units, reflecting tight supply of the specialty electrical steel used in their construction.
Gas turbines, increasingly used to supplement on-site power generation, face similar constraints, with elevated order books pointing to extended lead times.
These constraints are keeping equipment prices high. The supply-chain implication is direct: silicon capacity delivers limited value without the power infrastructure to support it.
Processors and memory: Accelerating growth, concentrated risk
Once power availability is resolved, the constraint shifts to processors and memory, where supply-demand imbalances are most acute. AI accelerators combine graphics processors, conventional processors and high bandwidth memory, and demand across all three continues to outpace capacity additions.
Leading accelerator designers are forecasting annual revenue growth well above 50% through the rest of the decade, but that trajectory depends on the rollout of next-generation chip systems, which bring added manufacturing and assembly complexity.
Memory has become the main constraint. Producers are reallocating capacity from conventional formats toward the high bandwidth memory required by accelerators, a shift reflected in rising export prices from the leading production hub for legacy products. That pricing signal is driving investment: high bandwidth memory output is expected to expand rapidly over the next several years, while conventional memory grows more moderately.
The durability of this technology is not assured. Compression techniques and alternative architectures could reduce reliance on current memory formats, turning today's shortage into a future oversupply risk.
That risk fits the memory sector's boom-bust pattern, in which downturns have historically recurred every five to seven years as capacity investment outpaces demand. Even with AI-driven demand, a comparable correction could emerge before the end of the decade.
Processor pricing is also likely to remain firm. Producer price indices covering processors, accelerators and logic devices are rising across major markets, reflecting persistent capacity constraints. This imbalance is unlikely to fully ease before new fabrication and packaging capacity comes online toward 2028. Until then, costs should remain under upward pressure, moderated only by greater emphasis on performance-per-dollar efficiency rather than continued adoption of the highest-cost processors.
Semiconductor manufacturing equipment: Capital expenditure set to double
Elevated prices and sustained demand are driving a major expansion in fabrication equipment spending. Aggregate capital expenditure across major chipmakers outside mainland China is projected to nearly double from the prior three-year period, funding new facilities across multiple jurisdictions and attracting new entrants.
Next-generation lithography equipment is entering production in parallel, though its high cost may lead some manufacturers to delay adoption rather than absorb the premium.
Export controls continue to shape where the newest equipment can be deployed, keeping a large share of chipmaking equipment trade concentrated in established manufacturing hubs even as investment diversifies.
Electronics components and materials: Shortages driving diversification
Below processors and memory sits a less visible but critical layer of specialty materials and passive components. These inputs are not always expensive individually, but they determine whether advanced servers can be assembled at scale. Concentration risks in critical minerals, industrial gases and electronic-grade chemicals are pushing buyers to diversify supply, including by restarting idled mines for strategic materials.
Helium illustrates the fragility of these inputs. It is used in semiconductor manufacturing, optical fiber production and other electronics processes where stable, inert conditions are required. Disruption to a major export source left spot markets tight despite long-term contracts and inventories. Capacitors used in AI servers face similar pressure as demand strains supplies of the metals, films and ceramic materials required for production.
These pressures converge in printed circuit board assembly. Higher component density, larger memory footprints and more capacitors and connectors per board are lifting costs, while strong AI, data center and defense demand lets suppliers pass those costs through. The result is a broader inflationary effect across server bills of materials, not just the headline accelerators. Board assembly prices rising faster than underlying inputs are a clear signal of where pricing power currently sits.
Yan Hoong and Emiliano Pérez contributed to this article
This article was published by S&P Global Market Intelligence and not by S&P Global Ratings, which is a separately managed division of S&P Global.
Content Type
Location
Products & Offerings