Wall Street's Capital Economics Just Called the AI Boom 'Nearing an End'—Here's Why This Bear Note Sticks

The Capital Economics note, dated September 10, 2026, carries the headline 'Most indicators suggest the AI equity boom is nearing an end.' Higgins keeps his year-end 2026 S&P 500 target at 8,250—about 8% above the September 9 close of 7,636.36—because he still expects a 'blow-off' melt-up before the unwind. The sharper call sits12 months out: a forecast of 6,500 by the end of 2027, a 21% peak-to-trough drop anchored on downward revisions to AI-company profit forecasts as competitive pressure from Chinese memory and model labs starts to bite.

Concentration: the Magnificent Seven now make up 33.9% of the S&P 500 as of September 2026, with NVIDIA alone at 7.9% and a market cap of $5.4T. The top-10 weighting has reached levels that match the March 2000 dot-com peak when ten stocks controlled 27% of the index; today those ten names approach 41% of index weight, per Apollo's September 2025 tracker.

Hyperscaler capex intensity: the four largest U.S. hyperscalers—Microsoft, Alphabet, Meta, and Amazon—are tracking roughly $733B of aggregate capital expenditure in 2026 after raising guidance through the second quarter. Alphabet lifted the top end of its 2026 capex range to $205B; Amazon guided to roughly $220B; Microsoft called for ~$190B; Meta set a $125–$145B range. According to J.P. Morgan Asset Management, AI capex has jumped from 33% of hyperscaler cash flow from operations in 2023 to an estimated 93% in 2026—capital intensity the S&P 500 has not seen outside the 1999–2000 build-out.

Memory-chip tell: SK Hynix is the cleanest single read on the AI trade's health. Its stock lost roughly $2.5T of combined KOSPI market value in a single trading week in late July 2026 after the company said it would slow HBM capacity expansion in favor of conventional DRAM. KOSPI fell 10.84% in one session—its fifth-worst plunge on record—and SK Hynix shares fell about 15%. Capital Economics' volatility criterion treats only the Asian financial crisis, the dot-com bust, and the 2008–09 GFC as historical analogs for that kind of session.

Leverage at the edge: the AI infrastructure build-out has migrated from investment-grade balance sheets to GPU-collateralized debt at the neocloud layer. CoreWeave disclosed $35.6B of total indebtedness on its June 30, 2026 Form 10-Q, against roughly $6.2B of trailing-twelve-month revenue and $31–$35B of 2026 capex guidance. Oracle has telegraphed plans to lift its debt load to $290B for cloud expansion. Capital Economics frames this leverage as the kind of 'circular financing' the BIS warned about in its June 2026 Annual Economic Report—circular flows that can unwind abruptly when one counterparty rerates.

Macro stack: the September timing is not coincidence. The 10-year Treasury yield ended September 10, 2026 at 4.87%, near multi-month highs, with JPMorgan noting that AI capex plus delayed tariff pass-through are now contributing similarly to elevated core inflation. Federal Reserve Chair Jerome Powell acknowledged in March 2026 that the AI-fueled data center boom is contributing to inflation, and the early-2026 closure of the Strait of Hormuz pushed Brent up 67% to an intraday peak of $120 a barrel, per the BIS report. Higher rates and higher energy prices are the macro squeeze that makes Capital Economics' bear case mechanically operable.

If Capital Economics is right, the transmission moves along the AI supply chain rather than across the broad market. Upstream chipmakers carry the most direct exposure: NVIDIA just printed $96.2B of Q2 FY2027 revenue (+106% YoY) and $89.0B of data-center revenue (+117%), but consensus already discounts a deceleration trajectory. TSMC reported a record $40.2B Q2 2026 revenue, lifted 2026 guidance, and is now committing roughly70–80% of an expanded capex envelope to advanced packaging (CoWoS)—a bet that AI volume continues. ASML raised its 2026 outlook to €36–40B with plans for 30% capacity expansion in each of the next two years; if orders roll over, the 2027 cliff is real.

Memory is the second-order warning. SK Hynix's Q2 2026 print was record-breaking on every line—₩79.3T revenue (+257% YoY), ₩60.5T operating profit (+557% YoY), 76% operating margin—yet the stock fell 10% after the report because conventional DRAM missed consensus and management guided a slower HBM ramp. Micron reported fiscal Q3 2026 revenue of $41.46B, GAAP net income of $28.24B, and a non-GAAP gross margin of 84.9%, with Q4 guidance near $50B. The implication is that memory is the first AI leg to absorb any demand digestion.

Downstream demand hinges on three power buyers: Constellation Energy signed920 MW of new contracts starting2029–2031 and lifted2026 EPS guidance to $11.50–$12.50; Vistra signed20-year PPAs with Meta for 2,609 MW of nuclear power and an additional1,200 MW deal with AWS, driving Q2 2026 adjusted EBITDA up 30%; GE Vernova has become the critical-power-supplier thesis with $2.7B of recent data-center order flow. These names are the cleanest hedges against a Capital Economics-style AI unwind, since their contracts lock in multi-decade cash flows that do not require AI capex to keep growing.

The structural damage in the bear case lands on the neocloud complex. CoreWeave carries $35.6B of debt on $6.2B of trailing revenue with $31–$35B of capex guidance—a 5x revenue-to-capex ratio that breaks if hyperscaler customers renegotiate. Oracle is funding $50B of fiscal2026 capex with bond issuance and has telegraphed a $290B debt load for cloud expansion. These leveraged structures are exactly the 'outsourced data centers leased back under long-dated contracts with poorly disclosed terms' that the BIS flagged as a potential cascade risk.

Short-term (days to quarters): the next session-level signals are NVIDIA order-book commentary, any hyperscaler capex guide-down during the upcoming Q3 2026 earnings cycle in October, and the 10-year Treasury's reaction to the September CPI print on September 11, 2026. The Bank of America Bull & Bear Indicator at 9.7 is now the highest sell signal since 2021, and Hartnett's call to 'rotate into defensives'—staples, REITs, small caps, biotech, the U.S. dollar—points to where the marginal dollar goes if the Capital Economics framework gains traction. Watch the S&P 500's September 30 close against Capital Economics' 8,250 year-end target for a clean validation or refutation.

Medium-term (1–2 quarters): the most important data points are (1) whether Microsoft, Alphabet, Amazon, and Meta reaffirm full-year capex envelopes when they report Q3 2026 in October; (2) any2027 capex pre-announcement, which would be the first concrete signal of digestion; (3) TSMC monthly revenue prints showing whether advanced packaging utilization stays above90%; and (4) the trajectory of memory pricing—Micron and SK Hynix are the cleanest demand thermometers.

Long-term (1–3 years): the structural test is whether AI revenue scales to absorb the capex stack. OpenAI is reportedly running a $24B annualized rate as of April 2026 against a 2025 net loss near $14B; Anthropic reached $47B annualized in May 2026 and posted its first quarterly profit; xAI loses roughly $26 for every $1 of revenue. The BIS June 2026 report explicitly warned that AI capex outpaces hyperscaler earnings and free cash flow, forcing debt issuance—and that a confidence shock could turn the boom into a 'protracted investment bust.' Capital Economics is betting that 2027 is when that confidence shock arrives; the question is whether the data confirms it before the S&P 500 reaches 8,250.

Capital Economics is not the first sell-side voice to flag AI bubble risk—the BIS, Bank of America's Hartnett, Goldman Sachs's Kostin, and Morgan Stanley's Wilson have all warned, in various registers, since late 2025. What makes this note different is the combination of a precise end-date (end-2027), a precise drawdown (21%), and a precise mechanism (downward AI profit revisions triggered by Chinese competition and demand digestion). That specificity converts a bubble warning into a tradable thesis—and is the reason this is the first call that deserves a portfolio response rather than a shrug.

The investment implication is not 'sell everything'—Capital Economics still has the S&P 500 at 8,250 by year-end 2026, implying roughly 8% more upside from current levels. The implication is rotation: reduce exposure to the longest-duration AI beneficiaries (CoreWeave, pure-play memory, leveraged neoclouds) and add exposure to AI infrastructure plays whose cash flows are locked in regardless of capex cycles (Constellation Energy, Vistra, GE Vernova, Equinix). Within the hyperscalers themselves, the framework favors Alphabet and Amazon over Microsoft and Meta on a relative basis, because the first two carry more diversified revenue while the latter two have the highest AI capex-to-cash-flow ratios.