Nasdaq closed down 5.5 percent yesterday. The S&P 500 fell 4.4 percent and the Dow 3.4 percent. Asia followed overnight. The Nikkei dropped roughly 6 percent and moved into a bear market. The Kospi fell 8 percent and hit a circuit breaker. Samsung, SK Hynix, and the rest of the chip complex were hit hard.
The consensus explanation is straightforward: Democrats took the midterms, local and state races flipped in places that matter for data-center buildout, and the market is now pricing harder regulation and slower expansion. That reading is not wrong. It is also not the interesting part.
The interesting part is that the expansion was already slower than the chip deliveries.
High-voltage transformers still sit at lead times above 160 weeks. Morgan Stanley’s revision of U.S. data-center power demand to 97 GW through 2028 left a shortfall of roughly 32 GW even after behind-the-meter generation. You cannot announce a new campus on the same schedule as a new transformer.
The midterms add another layer of friction—slower permitting, grid-upgrade costs shifted back onto the hyperscalers, and the removal of local tax incentives in the states that have been the primary expansion markets. They do not create the constraint. They make the existing one more binding.
The power grid does not care who controls Congress. It cares whether the transformers and permits arrive before the chips do. Yesterday, the market priced the political shift. It spent less time pricing the operational reality that was already visible in the latest hyperscaler prints.
AWS cleared its top-line number. Segment revenue hit $46 billion, up 39 percent year-over-year. Trailing twelve-month free cash flow turned to an outflow of $42.4 billion. Management kept the 2026 cash capital-expenditure guide near $220 billion. Hardware is arriving faster than the grid can energize the sites.
Dark GPUs sit on the books, generating depreciation while they wait for power. Retail cash flow is being transferred straight into utility pockets to turn the machines on. That is not a regulation story. It is a capital-allocation story that was already underway.
The chip selloff treats the problem as if the limiting factor were the supply of GPUs or the willingness of politicians to approve new campuses. Both matter at the margin. Neither is the binding constraint. A chip without a grid connection and a cooling system is inventory. It is not useful capacity.
The people who locked in GPU capacity just to resell it are discovering the difference in real time. The hyperscalers will still spend. They will simply spend with more systemic delay, more forced relocation, and more of the free-cash-flow pressure that has already shown up in the numbers.
Electricity remains the non-negotiable payment medium for compute. That is why the portfolio has continued to accumulate energy producers. Not as a thematic bet on any transition narrative, but because they own or control the physical input that turns costly metal into billable capacity. Constellation Energy is an active holding for this reason.
Under the 70/30 framework, we hold Amazon and we hold the grid it has to buy. The framework is doing the job it was built for. See Electricity Is the Payment Medium for Compute.
The market is still pricing the software and the chips while the grid sets the pace.