Is this the beginning of the end of the AI stock market boom?

The Nasdaq 100 fell 5.5% last week amid concerns that roughly $US1tn of annual AI investment may not earn adequate returns, though tech stocks rebounded in Monday’s Wall Street trade and the index remains nearly three times its early-2023 level. Hyperscalers spent $US425bn on AI-related capex in 2025 and are on track to invest $US800bn this year and more than $US1.1tn next; their corporate bonds now make up nearly 10% of global investment-grade credit, up from 0.5% at end-2024. Strategists characterize the pullback as a valuation reset rather than a rejection of AI, while warning about overinvestment, debt, circular financing and potential opposition to datacentres and AI’s effects on workers.

Analysis

The key distinction is between AI adoption and returns on AI capital: usage can grow while infrastructure owners and chip suppliers earn below-market returns if capacity is overbuilt or model costs collapse. An open-source efficiency shock could therefore benefit enterprise adopters while pressuring the scarcity premium in compute; any demand rebound is a possible offset, not a sufficient hedge against lower pricing. Circular commitments amplify the downside because weaker project economics can hit both data-centre build-outs and chip orders.

Near term, earnings are a catalyst for repricing, not proof of a cycle peak. Track incremental cloud revenue and margins against capex, free-cash-flow conversion, utilization, and the quality of customer commitments—not headline AI demand. Over 6–18 months, depreciation and underused capacity may expose returns that current spending narratives obscure. Conversely, sustained utilization and monetization would falsify the under-earning thesis. The article’s debt and circularity claims warrant verification issuer by issuer; do not infer equivalent balance-sheet risk across the named companies.

The contrarian opportunity is to separate a potentially durable technology shift from a potentially poor infrastructure investment. A broad AI bust is not required for crowded, capex-sensitive exposures to underperform; nor does one failed Australian listing establish a global funding freeze. Australia’s data-centre exposure appears less systemically consequential than US hyperscaler and credit-market linkages.

Key Decisions for Investors

• Use a modest QQQ put-spread or reduce Nasdaq-100 beta into the US reporting window rather than make an outright top call. Define the hedge against the portfolio’s AI concentration; unwind if hyperscalers show improving cloud economics alongside capex discipline.
• Consider a relative-value underweight in capex-sensitive AI infrastructure exposure, especially NVDA, against a broader or equal-weight equity benchmark. Keep sizing modest: falling model costs could increase total compute demand, and earnings may extend the rally before returns deteriorate.
• For the next 1–3 months, score AMZN, MSFT and GOOG on incremental cloud growth versus capex, cloud margins and free cash flow. Escalate hedges only if capex guidance rises while monetization or margins weaken; this is the clearest near-term thesis confirmation.
• Watch for verified project cancellations, utilization shortfalls, weaker chip commitments, or widening credit spreads over 6–18 months. Those would strengthen the overbuild thesis; sustained utilization, credible customer-backed demand and improving returns on invested capital would falsify it.