‘Crazy days’ and ‘silly season’ in the stock market raise the prospect of a ‘late-stage’ AI bubble finally popping

Markets are pricing an 85%+ probability that the Fed will raise rates by 25bps to 3.75%-4.00% on Wednesday, despite core CPI falling to a post-pandemic low of 2.4% in August. Capital Economics warns the AI trade is a late-stage bubble and forecasts the S&P 500 could ultimately decline at least 30%, citing dot-com-level earnings expectations, hyperscaler capex that could drive combined free cash flow negative by 2027, and rising equity issuance. Morgan Stanley remains constructive, retaining S&P 500 targets of 8,000 by year-end and 8,300 by mid-2027, arguing that forward P/E has already compressed from 22.5x to 19.5x while earnings growth exceeded 30% year over year. A hawkish Fed pivot would increase downside risk for highly valued AI and semiconductor equities, although market participants remain divided on whether policy tightening will materially curb hyperscaler investment.

Analysis

The relevant transmission channel is not a single 25 bp move, but a repricing of terminal rates and the equity risk premium applied to long-duration AI cash flows. MSFT, AMZN and META can fund capex internally today, yet their 2027-28 return-on-invested-capital debate becomes materially harder if incremental debt costs remain elevated while inference monetization lags. The more vulnerable leg is the AI supply chain—NVDA, AVGO, ANET and data-center infrastructure names—where consensus revenue assumes hyperscaler capex remains both large and economically rational.

Cross-sectional dispersion is likely to rise before index-level weakness becomes obvious. A tighter policy regime favors profitable, near-term cash-generative software and financials over pre-profit AI infrastructure, private-market proxies and IPO beneficiaries; it also increases the hurdle rate for venture-backed AI firms, potentially reducing future cloud workload demand. For MS and UBS, elevated volatility and financing activity can support trading and underwriting revenue near term, but a sustained risk-off move would impair wealth-management flows and IPO calendars within 1-3 months.

The contrarian case is that a modest hike is already discounted and could relieve uncertainty if the statement does not validate a multi-hike path. The key falsifier for the bearish AI-duration thesis is stable or rising 2027 hyperscaler capex guidance accompanied by evidence that cloud/AI revenue is accelerating faster than depreciation and financing expense. Conversely, any capex moderation, AI revenue disclosure disappointment, or widening of investment-grade tech credit spreads would turn a valuation concern into an earnings-revision cycle over 6-18 months.

Key Decisions for Investors

• Initiate a 1-3 month pair: long JPM / short MSFT in equal dollar beta. JPM benefits from a higher-for-longer earnings backdrop and shorter-duration cash flows; MSFT is the cleanest mega-cap duration exposure. Exit if the 10-year Treasury yield falls below its pre-meeting level or MSFT raises AI monetization guidance.
• Buy 3-month QQQ put spreads, financed by selling a farther-out-of-the-money put, rather than outright index shorts. Target a 7-10% downside window; risk is limited if the Fed delivers a dovish hike and earnings estimates remain intact.
• Underweight NVDA, AVGO and ANET versus AAPL over the next 1-3 months. This is not a structural AI short; it is a multiple-risk trade against the highest capex-beta beneficiaries. Cover if hyperscalers reaffirm aggregate 2027 capex growth above current consensus or semiconductor order lead times extend.
• Treat any Anthropic-related IPO enthusiasm as a liquidity-risk indicator, not a standalone short catalyst. Raise hedges if IPO pricing is upsized, aftermarket turnover is extreme, and high-grade technology credit spreads widen simultaneously; that combination would indicate speculative equity issuance is coinciding with deteriorating marginal funding conditions.