AI Squeezes Server Parts
- Vendors are diverting power-management and board-control chips from general servers to AI hardware, tightening supply. - Specialist firms like Navitas saw share gains as investors bet on gallium nitride and other power-semicondutor demand. - The shift means shortages and repricing now hit non-GPU parts, forcing data-center teams to rethink ordinary server expansion plans. (theregister.com) (ibtimes.com.au)
The parts crunch in data centers has spread beyond graphics processors: power-management chips and board-control chips for ordinary servers are now getting squeezed as suppliers steer capacity to higher-margin artificial-intelligence systems. (theregister.com) TrendForce said on April 15 that 2026 server shipment growth is now expected to be about 13 percent, down from a forecast near 20 percent, even though demand for general-purpose servers remains steady. The firm said lead times for some components, including printed circuit boards and central processors, are stretching toward a year. (trendforce.com) The tighter list now also includes power management integrated circuits, which regulate electricity inside a server, and baseboard management controller chips, which let operators monitor and control machines remotely. The Register reported those parts are being redirected into artificial-intelligence hardware, where vendors can charge more. (theregister.com) That marks a shift from the shortages of 2024 and 2025, when the bottleneck was mostly memory tied to artificial-intelligence servers, especially high-bandwidth memory and server dynamic random-access memory. TrendForce said the imbalance now reaches “multiple general server components,” not just the parts attached directly to accelerators. (trendforce.com) Artificial-intelligence racks use far more electricity than standard servers, which has turned power delivery into a selling point rather than a background feature. Specialist chipmakers that make faster, more efficient power components are now drawing investor attention alongside the better-known graphics-chip names. (mckinsey.com) Navitas Semiconductor rose 8.81 percent on April 23 ahead of earnings scheduled for May 5, after investors focused on its gallium nitride and silicon carbide products for data-center power systems. International Business Times Australia said mobile now accounts for less than 25 percent of Navitas revenue as the company shifts toward data-center and industrial markets. (ibtimes.com.au) Navitas has been pitching those materials as a way to handle higher voltages with less wasted energy, a useful trait in dense artificial-intelligence clusters. The company said on April 16 that it will report first-quarter 2026 results after the market close on May 5. (nasdaq.com) Not every server buyer is chasing artificial-intelligence capacity, but they are competing in the same supply chain. TrendForce said unmet demand for general-purpose servers is likely to carry into 2027 as capacity ramps and component shortages ease only gradually. (trendforce.com) For data-center operators, that means expansion plans for routine web, storage, and enterprise workloads can now be delayed by the same artificial-intelligence buildout that once mainly distorted graphics-chip and memory markets. The shortage has moved one layer deeper into the server. (theregister.com)
