AI Bubble: It's Not One Bubble, But Three Ready To Burst

The author argues that the AI market faces overlapping bubbles in earnings expectations, interconnected financial risks and concentration of capital. Goldman Sachs forecasts trillions of dollars in AI infrastructure spending, while Gartner estimates cumulative revenue may fall short of what is needed to justify a 12% return.

Analysis

The investable risk is a mismatch between long-lived infrastructure commitments and uncertain, shorter-cycle monetization—not simply whether AI adoption continues. If customer revenue fails to support current build plans, hyperscalers may slow orders before end demand disappears, transmitting pressure from accelerators and networking into data-center construction, power equipment, and financing. The reverse feedback matters: weaker utilization can reduce the return on new capacity and make the next capex decision harder, while concentrated ownership may amplify a repricing across index-heavy technology shares.

Near term, the key test is whether upcoming cloud-provider guidance and supplier order commentary confirm conversion of spending into paid workloads, rather than pilot activity or backlog. Over 6–18 months, utilization, inference economics, and depreciation burdens should distinguish durable infrastructure demand from overbuild. The article alone does not establish that spending is excessive or that a downturn is imminent; it provides no company-level exposure, valuation, or positioning data. A sustained rise in commercial AI revenue and capacity utilization would falsify the bearish case.

Key Decisions for Investors

• No broad short solely on this signal. Treat it as a reason to review concentration and avoid adding to highly correlated AI-infrastructure exposures without evidence of revenue conversion.
• If valuations and crowded positioning independently confirm vulnerability, consider a modest, staged relative-value underweight in semiconductor and data-center infrastructure exposure versus diversified, cash-generative technology. Initiate on rallies rather than after a sharp drawdown; the risk is that capex guidance and monetization continue to surprise higher.
• Over the next 1–3 months, monitor cloud-provider capex plans, supplier order/backlog conversion, and reported AI workload revenue. A capex cut without corresponding revenue growth would strengthen the downside case; sustained utilization and monetization would argue against it.
• Reassess over 6–18 months using capacity utilization, returns on deployed infrastructure, and depreciation trends. Without those data, avoid assigning a specific downside target or treating aggregate spending estimates as proof of impaired returns.