TSMC's 3-Year AI Runway Quietly Rescues US Grid Equipment Suppliers

If hyperscalers aggressively shift to modular nuclear (SMRs) or co-location with existing utility plants at a pace that bypasses traditional grid interconnection, the transformer/switchgear bottleneck partially resolves without benefiting the named industrials. The observable signal would be a string of announced SMR power purchase agreements by Microsoft, Amazon, or Google that specify on-site generation, which would reduce grid interconnection hardware demand. Alternatively, a sharp reversal in AI capex guidance from a major hyperscaler (Meta, Microsoft Q2 earnings) would break the demand thesis at its foundation.

The strongest counter-argument is that the power infrastructure bottleneck is already well-known and has been discussed in industrial earnings calls for 6+ quarters — meaning it may already be priced into names like ETN and GEV, which have re-rated significantly over the past 18 months. The specific assumption most likely to be wrong is that transformer lead times remain structurally elevated; if the Biden/Trump administration IRA reshoring incentives accelerate domestic transformer manufacturing capacity faster than expected, the supply constraint eases and pricing power erodes. The data point that would flip conviction: a sharp sequential decline in new orders reported by Eaton or GE Vernova in their next earnings, signaling that the hyperscaler capex wave is pulling forward rather than compounding demand.

The internet buildout of 1998–2001 offers the clearest analog. Cisco and Sun Microsystems (the 'obvious' plays) saw massive multiple expansion while companies like Emerson Electric — supplying power conditioning and UPS systems to data centers — quietly compounded at 15–20% annually through 2003, even as the Nasdaq crashed 78%. The pattern: transformative technology cycles create derivative infrastructure bottlenecks that the market misprices because they're categorically boring. A second analog: the shale boom of 2010–2014 drove oil services stocks (HAL, SLB) to outperform E&P names in the first two years because the bottleneck was drilling capacity, not oil in the ground. Here, the bottleneck is not compute but the power to run it.