The recent decline in chip stocks reflects market uncertainty regarding the cost and pace of data center construction, rather than a sudden deterioration in the outlook. It also appears more like a risk-mitigation event than a substantial change in fundamentals. Increased congressional oversight may indeed intensify scrutiny of data center construction and AI regulation. However, unless widespread AI safety issues arise, we believe broad federal restrictions are likely to face legislative and presidential constraints.
While state and local political resistance may delay, shift, or increase the cost of data center projects, this is unlikely to alter underlying demand. The proposed approximately 400 GW of capacity in the US already far exceeds our construction assumption of approximately 90 GW by 2030. Therefore, we believe regulation is more likely to reallocate and extend investment cycles than derail them, with power access, licensing approvals, financing, and enforcement capabilities increasingly becoming the dividing line between leaders and laggards.
Underlying AI-related demand also remains strong. Recent earnings reports show accelerating cloud revenue growth and increasing orders for computing power. We forecast total AI capital expenditures to be $1 trillion in 2026, $1.4 trillion in 2027, and $1.5 trillion in 2028.
For this sell-off to signal a sustained downturn in the AI investment cycle, we believe we need to see evidence of: slowing growth in AI-related revenue, weakening cloud utilization, or a failure of declining AI costs to generate sufficient new usage to absorb expanding capacity. Furthermore, we will reassess the outlook if improvements in algorithmic efficiency significantly reduce computing demands, larger investments fail to translate into corresponding increases in AI capabilities, or tighter financial conditions limit spending by large technology platforms and cutting-edge labs.
