Meet Kevin posted on November 16 that SemiAnalysis had sharply lowered its 2030 AI spend outlook, and that timing points to the firm’s November 13 update to its Accelerator and HBM Model. Phemex News and Marked Chart both report the 2030 estimate moving from $1.531 trillion to $923 billion, with 456X Research noting the update arrived just 11 days after the previous release. The post therefore captures both the size of the cut and the unusually short interval between forecasts.
The revision was not limited to 2030. Phemex News, 456X Research and OfficeChai report cuts of 28% for 2028, 41% for 2029 and 40% for 2030, with the cumulative reduction across 2027 to 2030 reaching roughly $1.6 trillion. Much of the dollar decline reflects lower chip prices rather than fewer chips: by 2030, accelerator spending falls 40% while logic silicon area falls only 15%, according to 456X Research.
SemiAnalysis attributed the downgrade mainly to slower revenue growth at frontier AI labs in the third quarter of 2026 and tighter financing for data centers, especially projects without hyperscaler offtake agreements, OfficeChai and 456X Research report. A secondary factor was election-related delays affecting about 5.8 gigawatts of US data center capacity due in late 2027 and early 2028. NVIDIA absorbs most of the dollar reduction, while demand from Broadcom and Amazon’s internal chips proves more resilient.
The downgrade does not erase expected growth. SemiAnalysis left 2026 essentially unchanged at $392 billion and still projected spending to rise, albeit much more slowly, through 2030. OfficeChai notes the model is sold to institutional subscribers and the figures circulated through summaries shared on X, but Phemex News, 456X Research, OfficeChai and Marked Chart all report the same revised 2030 number and the same pattern of cuts, reinforcing the post’s account.