SemiAnalysis Sharply Cuts AI Computing Power Forecast, with NVIDIA Hit Hardest
13 hour ago / Read about 0 minute
Author:小编   

Several institutions have significantly revised down their forecasts for AI accelerator shipments, capital expenditure, HBM, and power demand over the next few years—a signal that investment in AI infrastructure is cooling. The 2030 forecast for AI accelerator capital expenditure has been slashed from $1.531 trillion to $923 billion, while the cumulative forecast for accelerator logic wafer shipment area from 2026 to 2030 has been lowered by 17%. NVIDIA has taken the biggest hit, with both shipment forecasts and average selling prices for its latest-generation products revised sharply downward. Broadcom remains relatively stable, AMD has been less affected than NVIDIA, MediaTek's shipment adjustments are notable, and the near-term forecast for Amazon Trainium has not been cut. This adjustment amounts to a recalibration of high-growth expectations for AI, and industry competition is set to shift toward a comprehensive contest of performance-to-cost ratio, energy efficiency, and financing capability.