SemiAnalysis Cuts AI Accelerator Capex Forecast by 40% for 2030
The outlook now stands at $758 billion for 2028 and $923 billion for 2030. NVIDIA faces the largest shipment and power-demand revisions.

SemiAnalysis has cut its global AI accelerator capital spending forecast, reducing the 2028 estimate by 28% and the 2030 estimate by 40% as weaker financing and capacity delays reshape expected deployments.
The revised projections are $758 billion for 2028 and $923 billion for 2030. The outlook also includes reductions to accelerator shipment expectations and power demand.
NVIDIA faces the steepest reductions among the suppliers covered. Its projected accelerator shipments, measured by computing revenue, were cut 22% for 2027, 39% for 2028, 56% for 2029 and 66% for 2030. The forecast for total power draw was reduced 14%, 35%, 50% and 59% in those years, respectively.
Broadcom’s ASIC revenue outlook for 2026 through 2030 fell 11%, while projected shipment volumes were unchanged. Lower high-bandwidth memory (HBM) costs are expected to reduce selling prices rather than shipments. Amazon’s Trainium forecast was also unchanged for 2027 and 2028.
Net-new annual recurring revenue growth slowed in the third quarter at frontier model labs. Debt and equity financing for AI infrastructure tightened, particularly for projects without hyperscaler backing. Those financing limits can constrain development of sites, power access and building shells, reducing the scale of systems that can be deployed. Forecast revenue for the labs remained above market consensus, while infrastructure financing depends on capital-market participants’ average expectations.
The forecast also factors in 5.8 gigawatts of U.S. data-center capacity originally scheduled for deployment from the second half of 2027 through the first half of 2028 that was delayed. The 5.8 GW delay followed U.S. midterm-election results, and regulatory delays also reduced expected 2028 capacity. Revenue expectations and financing pressures were more significant drivers of the revision.
Financing terms differ across suppliers. Broadcom can provide creditors with guarantees on customers’ behalf covering land, power, building shells and systems. NVIDIA has greater exposure to smaller, long-tail neocloud operators, whose financing and execution risks have increased.