律动BlockBeats|11月 16, 2026 13:37
[SemiAnalysis Significantly Lowers AI Chip Demand Forecasts: Nvidia May Bear the Biggest Pressure]
BlockBeats reported on November 16th that AI research firm SemiAnalysis, in its latest report, significantly lowered its forecasts for AI chip and server shipments and related capital expenditures after 2028. The report cites a slowdown in the average revenue per share (ARRR) of new cutting-edge AI labs, a tightening financing environment, and delays in US data center construction as factors that will collectively suppress computing power expansion in the coming years.
According to the firm's new model, global AI accelerator capital expenditures are projected to reach $758 billion in 2028, a 28% decrease from previous forecasts; $796 billion in 2029 and $923 billion in 2030, representing decreases of 41% and 40% respectively. The report attributes the downward revisions primarily to falling chip prices, reduced shipments, and lower assumptions about the power consumption per chip in future products.
Among major suppliers, Nvidia is considered to be the most impacted. SemiAnalysis projects that Nvidia's shipment forecasts, based on compute revenue, will be lowered by 22%, 39%, 56%, and 66% for 2027, 2028, and 2030, respectively. The firm also lowered its average selling price forecasts for Nvidia's future Feynman product lines, with Feynman dropping from approximately $110,000 to approximately $78,000 and Feynman Ultra from approximately $175,000 to approximately $98,000.
The report argues that Nvidia faces pressure not only from slowing overall demand but also from customers re-prioritizing cost, energy efficiency, and financing capabilities. Anthropic is showing a preference for TPUs, Trainium, and AMD GPUs, while OpenAI is expected to prioritize its own ASIC projects. Meanwhile, Nvidia has significant exposure to neocloud customers, who are facing greater financing and execution risks.
In contrast, Broadcom is considered the least affected, with its ASIC revenue forecasts for 2026-2030 only lowered by 11%, primarily due to price adjustments resulting from declining HBM costs, rather than significant reductions in shipments. AMD's forecasts were also lowered, but its customer base of Meta, OpenAI, and Anthropic, along with related stock option arrangements, are believed to provide some support for demand. Amazon Trainium's forecasts for 2027-2028 remained unchanged.
The report's overall assessment is that AI computing power construction will not stop, but future growth will be more constrained by financing capabilities, data center supply, chip prices, and the progress of customers' self-developed chips; in this round of repricing, Nvidia may be the most significantly affected supplier.
Share To
Timeline
HotFlash
APP
X
Telegram
CopyLink