There aren’t enough AI chips to support data center projections, report says

A new analysis by London Economics International (LEI) for the Southern Environmental Law Center (SELC) suggests that U.S. data center growth and associated AI electricity demand forecasts are significantly overstated, primarily due to global semiconductor chip production constraints. The report indicates that meeting projected U.S. data center demand would unrealistically consume 90% of global chip supply through 2030, leading utilities to potentially overinvest in energy resources. This overestimation poses substantial economic risks for ratepayers and underscores the critical need for regulators to enforce more transparent and evidence-based infrastructure planning to mitigate speculative capital deployment.

Analysis

A new analysis by London Economics International (LEI) introduces a significant counter-narrative to the prevailing bullish outlook on U.S. data center growth, arguing that projections are almost certainly overstated due to global semiconductor supply constraints. The report's core finding is that for U.S. grid operators to meet their forecasted 57 GW of data center demand growth by 2030, the U.S. would need to absorb over 90% of the world's new semiconductor chip supply, a scenario LEI deems unrealistic given the U.S. currently accounts for less than 50% of global demand. This analysis provides a quantitative sanity check, noting that even an accelerated 10.7% annual growth in AI chip manufacturing—up from a historical 6.1%—would only supply an incremental 63 GW of demand globally. The direct implication is that utilities, responding to these potentially inflated forecasts, risk over-investing in speculative infrastructure, creating significant economic risks for ratepayers who would bear the financial burden of stranded assets. This skepticism is reinforced by industry observations that data center interconnection requests are frequently duplicative, with utilities seeing five to ten times more requests than projects that materialize, suggesting a substantial "upward bias" in demand signals.

Key Decisions for Investors

• Investors with exposure to utilities and energy infrastructure should scrutinize capital expenditure plans that are heavily predicated on data center demand, as the risk of asset overbuilding and lower-than-expected returns appears elevated.
• The analysis highlights semiconductor manufacturing as a critical bottleneck, suggesting a need to monitor chip capacity growth as a lead indicator for the actual pace of AI infrastructure deployment, which may temper growth expectations for hyperscalers and data center REITs.
• Consider favoring companies across the value chain, from utilities to tech, that demonstrate disciplined capital allocation and can provide evidence of firm, non-speculative offtake agreements for new data center projects.
• This report introduces a medium-term physical constraint on the AI growth narrative, and positions in the semiconductor equipment sector may offer a more direct way to play the build-out theme than those exposed to potentially inflated end-market demand forecasts.