From the archive · Wednesday, October 7, 2026
Google's nuclear deal lifted the companies that sell power at market prices, not the buyer and not the regulated utilities. The risk to that trade is political.
The market paid the owner of the scarce thing.
Scarcity priced at the meter
**The scarce megawatt.** On Tuesday Google agreed to buy more power from Constellation Energy's existing nuclear plants under a 20-year contract, the largest deal of its kind. The S&P 500 and the Nasdaq closed at records, and the ten-year eased from its highest level since 2002.
The useful observation is who gained. Constellation rose 12.2%, and the other companies that sell electricity into wholesale markets rose almost as much. Companies whose returns are set by commissions rose a fraction of that. Alphabet, the customer, barely moved. The market was not paying for nuclear technology or for Google's plans; it was paying for owning spare power at a time when there is not much of it.
Our positioning data had been saying the same thing for months. Speculative traders built length in power futures on the largest US grid through the summer while cutting their natural-gas length, which is a bet on running short of capacity rather than on the cost of fuel.
That shifts the risk. A shortage priced at market rates is paid by households and businesses, and energy bills are already political: Washington has just allowed cheaper diesel on highways ahead of the midterms, and Virginia's lieutenant governor opposes NextEra's purchase of Dominion. The price that rose fastest is the one most exposed to a cap.
Elsewhere, two of the three Gulf tanker days in our 4 October test are below average, Houthi forces struck Saudi airports, and the Fed's minutes and India's rate decision arrive today.
*Methodology.* Prices are Tuesday 6 October settlements and closes. Positioning is managed-money net length from the latest CFTC report (29 September), taking the contract with the largest open interest for each commodity. Gulf tanker counts are the engine's daily series against the average of covered days; a day with no coverage is excluded rather than read as zero.