JPMorgan: AI Capex Cycle to Continue, Semiconductor Equipment to Be Key Bottleneck

In an Asia technology strategy report released on September 16, 2026, JPMorgan analysts led by Gokul Hariharan addressed three major market concerns about AI infrastructure investment, concluding that the fundamentals of compute demand remain intact and the capex cycle will continue. They predicted that semiconductor equipment would become the most critical supply chain bottleneck around 2027.

Regarding worries that open-source models could lower token prices and erode profitability, the analysts argued that open-source models have historically driven token costs down by 80% to 90%, accelerating AI adoption across industries rather than weakening overall compute demand. They described the token price rebound in the first half of 2026 as an anomaly caused by severe compute shortages and the rise of agentic AI.

On AI safety regulation concerns, the report said frontier model training is unlikely to slow down significantly, and competition from open-source models will push labs to maintain pace. The analysts viewed AI safety debates as a regulatory response to rapid technological progress, not a signal of weakening demand.

The report also dismissed fears about negative free cash flow at hyperscalers, noting that net debt-to-equity ratios remain low at around 13% and that public cloud revenue growth provides support. JPMorgan estimated combined capex of Amazon, Microsoft, Google, Meta, Oracle, CoreWeave and SpaceX would rise from $443 billion in 2025 to $933 billion in 2026, then to $1.577 trillion in 2027.

On the supply chain, the analysts said semiconductor equipment would become the core constraint, driven by advanced process expansion at TSMC, a new investment cycle in mature fabs, and memory capacity buildouts. IC substrates and advanced packaging remain key bottlenecks, and near-package optics could be widely adopted within two years.