Postings, filings, auction results and index prints are as of the dates cited. The hiring role that opens the piece is OpenAI’s, not any other lab’s. Compensation figures are deliberately omitted: the bands are public, but what separates these seats is mandate rather than pay, and only one comparison earns its place in the text. Constellation’s CCGT sensitivity is an illustrative scenario on a fully hedged book with fuel held flat, and it is asymmetric (up to roughly +$200M against −$150M per $10/MWh); it is not a residual open position. Its 90/60/30 policy is spin-era, applied only to merchant revenues not already hedged through state programs, and has not appeared in a filing since 2023. The ~$810M implied by NRG’s 2028/29 capacity clear is our arithmetic on NRG’s own disclosed method; NRG published no such figure. Reported turbine counts and capacities at xAI’s Colossus 2 do not reconcile across sources and none of them is a filing; the piece shows the divergence rather than picking a number.

The spread-chain calculator is stylized and parameter-driven. Throughput figures come from a published vLLM benchmark sweep on a 70B-class model, and token prices are a representative public list tier — the combination illustrates the sensitivity of the second heat rate to utilization and is not a margin claim about any lab, model or provider. Hedge-quality judgments, the merchant-to-lab function mapping and the risk assessments are Kinetic Alpha’s. Nothing here is legal, tax or investment advice, a recommendation of any instrument or venue, or a view on any listed security. Research and education, not investment advice.