SemiAnalysis Cuts AI Spending Forecasts by Up to 41%: Bubble or Reality Check?

SemiAnalysis reportedly cut its 2029 global AI accelerator spending forecast by 41%, while its 2030 estimate fell by $608 billion to $923 billion. The reset raises a question for Nvidia shareholders and AI infrastructure investors: is demand collapsing, or were long-term expectations too aggressive? The revised model still projects annual spending to more than double from $392 billion in 2026, pointing to a reality check rather than an industry collapse.

The AI Spending Forecast Cuts, Year by Year

The reported November 13 revisions lower AI accelerator spending estimates most sharply in later years, when chip prices, financing and data center completion are harder to predict.

Year Previous forecast Revised forecast Reduction
2026 $392B $392B Approximately unchanged
2027 $770B $627B $143B (18.6%)
2028 $1.054T $758B $296B (28.1%)
2029 $1.352T $796B $556B (41.1%)
2030 $1.531T $923B $608B (39.7%)

The four revisions total $1.603 trillion in lower projected future spending for 2027–2030, not canceled orders, losses or debt. The percentages compare forecasts, not actual spending.

Yet $923 billion in 2030 would be 135.5% above the 2026 projection, or approximately 23.9% annualized growth. The revised outlook still implies rapid expansion.

Why the Outlook Changed

The reported update points to three interacting pressures: slower growth in revenue at frontier AI laboratories, tighter funding for projects without strong cloud-customer commitments, and less aggressive assumptions about future chip and memory prices. A separate account of the November 13 update also describes a reassessment of how much compute demand can support commercially.

Lower chip prices reduce dollar spending even when unit volumes hold up. Financing gaps and delayed power connections can postpone installations, while weaker AI service revenue makes new commitments harder to justify.

For chip valuations, unit shipments, selling prices and margins must be examined separately.

Nvidia Faces the Largest Reported Shipment Revision

For Nvidia (NASDAQ: NVDA), the reported SemiAnalysis model is particularly consequential. Nvidia accounts for 66% of the reduction in cumulative accelerator logic-silicon shipment area over 2026–2030, while the overall market’s projected shipment area is down 17%, according to the November 16 report. The 66% figure describes Nvidia’s share of the total forecast reduction. It does not mean Nvidia’s own shipments were cut by 66% across the five-year period.

Nvidia’s fiscal Q2 2027 results showed $96.2 billion in revenue, including $89.0 billion from Data Center, and a 75.0% GAAP gross margin. Its fiscal Q3 guidance called for $108.0 billion in revenue, plus or minus 2%, and a 74.0% GAAP gross margin, plus or minus 0.5 percentage points.

ALSO  RKLB vs. ASTS vs. ACHR: Which Frontier Stock Has the Strongest Business Model?

The question is whether margins and cash flow hold up. Nvidia disclosed $279 billion in supply and capacity commitments as of July 26, versus $119 billion one quarter earlier, plus $36 billion in AI-cloud agreements and a conditional Ohio project guarantee capped at $105 billion. These are different obligations, not immediately due debt.

For the shareholder exposure, see DualFinances’ analysis of Nvidia’s AI commitments.

HBM and Electricity Forecasts Also Move Lower

The adjustments extend beyond GPUs. The reported 2028 high-bandwidth memory (HBM) shipment forecast fell from 9.0 to 6.4 exabytes, a 28.9% cut. Forecast 2030 AI accelerator power demand dropped from 122.8 to 80.7 gigawatts, down 34.3%, according to the same industry report.

Lower HBM demand or pricing could pressure memory suppliers while reducing chip production costs. Lower anticipated power demand also changes expectations for future data center construction, cooling and electricity supply.

The impact will differ across equipment makers, as shown by Applied Materials’ results, and networking suppliers, including Cisco.

Bubble Warning or Reality Check for Investors?

Earlier expectations may have assumed too much growth, pricing power or easy financing. Yet the revised baseline still calls for substantial growth through 2030, not an industry contraction.

The shareholder question is who earns a return. Advanced Micro Devices (NASDAQ: AMD) warned in its November 4 Form 10-Q that power, construction and financing constraints can delay customer purchases.

Investors should track customer cash revenue, funded data center completions and chipmaker gross margins. The financing pressures at CoreWeave show why customer balance sheets matter.

Nvidia’s fiscal Q3 2027 earnings call on November 17 at 5 p.m. ET will offer another test of revenue, margins and financing discipline.

Bottom Line

Bottom line: SemiAnalysis’s reported revisions challenge the assumption that the AI buildout can keep accelerating at any price. They do not predict an end to AI spending growth: $923 billion in projected annual accelerator purchases by 2030 remains a substantial market. For shareholders, the key test is whether chip revenue, margins and customer cash generation justify the capital committed to building that market.

Data as of 1:00 p.m. ET, November 16, 2026. For informational purposes only; not investment advice.