Broader component tightness
- Recent notes flag shortages in DRAM/storage, optical transceivers, power infrastructure, and CPUs. - Analysts point to rising chip costs driven by capex, energy, and labour as the pressure behind these shortages. - That tightness is shifting allocation toward AI and high‑margin customers, reducing flexibility for industrial buyers. ( )
The supply crunch around artificial intelligence is no longer just about graphics chips; it now reaches memory, storage, networking optics, power gear and general-purpose server processors. (trendforce.com) TrendForce said on March 31 that conventional dynamic random-access memory contract prices were set to rise 58% to 63% quarter over quarter in the second quarter of 2026, while NAND flash prices were projected to climb 70% to 75%. The firm said suppliers were shifting capacity toward high-bandwidth memory and server products, and steering more NAND output into enterprise solid-state drives. (trendforce.com) The same pattern is showing up in networking parts that move data between AI servers. TrendForce said on April 20 that the market for AI optical transceivers is expected to jump to $26 billion in 2026 from $16.5 billion in 2025, and called component shortages the main bottleneck to adding capacity. (trendforce.com) Power equipment has become part of the constraint set too. The U.S. Department of Energy said data centers used about 4.4% of U.S. electricity in 2023 and could reach roughly 6.7% to 12% by 2028, while transformer lead times can stretch to as long as three years, according to Utility Dive’s reporting on manufacturers and grid officials. (energy.gov) (utilitydive.com) Electrical suppliers are reporting the same strain in their order books. Eaton said on February 3 that its Electrical sector backlog grew 29% year over year and that orders in Electrical Americas accelerated 16% on a rolling 12-month basis, driven by data-center momentum. (eaton.com) Memory makers have been explicit about where scarce capacity is going. SK hynix said in late 2025 that its 2026 supply of key memory products was already sold out, and Micron said in February 2026 that its entire 2026 high-bandwidth memory capacity had been booked under multiyear agreements. (techspot.com) (r40.io) That is changing who gets priority. TrendForce said suppliers are reallocating output toward server-related applications and enterprise storage, leaving consumer and lower-priority segments to absorb tighter supply and higher prices. (trendforce.com) The cost pressure starts well before the finished server leaves the factory. Vertiv’s 2026 data-center report said AI racks are moving beyond 25 kilowatts and often into triple-digit kilowatt levels, which forces heavier spending on power distribution, cooling and on-site generation alongside the semiconductors themselves. (vertiv.com) For industrial buyers, the result is a market with less slack and fewer substitutes. When memory fabs favor high-margin AI parts, optical suppliers chase 800G and faster links, and electrical vendors work through data-center backlogs, smaller customers face longer waits, steeper quotes and fewer chances to switch suppliers mid-cycle. (trendforce.com 1) (trendforce.com 2)
