The AI buildout's binding constraint is shifting from chips to electricity. Datacenters are signing decade-long power deals faster than anyone can build generation; firms that own or equip gigawatts capture pricing power for years
the chip is what today's evidence says: green means the world moved the way this bet needs, red means it moved against it. it reads the day, not the money. the receipts quietly keep track of that.
today's readFresh nuclear-plus-AI demo deals, a planned $100 billion hyperscale-plus-gas project, and PJM's backstop auctions plus data-center power-cut plans keep both new contracts and grid scarcity loud, while the basket's 30-day average versus utilities flipped positive to +5.7%.
→ watch tomorrow: Whether FERC approves PJM's backstop capacity auction and how high those prices actually clear.
a bet is only honest if it can lose. these two lines went down before the money did and are never edited: the first is what makes me admit the idea is wrong, the second is what makes me lean in harder.
AI power demand disappoints — hyperscaler capex cuts flow through to cancelled PPAs and shrinking interconnect queues, and the basket underperforms XLU for 2 consecutive quarters
Scarcity escalates — new multi-GW PPAs at premium prices, capacity auctions clear at records, and the basket outperforms XLU while AI capex holds
the US nuclear fleet — the 24/7 carbon-free power hyperscalers sign decade-long PPAs for
CEG remains the cleanest mechanism bet with 63.8% revenue growth and nuclear fleets that directly supply the new multi-GW deals, at a 21 forward P/E and 34% below its high; VST's cheaper 15 P/E and solid 43% growth are real but less nuclear-heavy, while GEV's 40 P/E and -6.5% relative drag make it the weakest expression today.
what got picked, and when, is logged on its own on the receipts page, the misses alongside the hits.
one stock, re-picked every weekday from the data below · not investment advice
To support the thesis that nuclear is capturing AI-driven power scarcity, CEG must show new or expanded multi-year hyperscaler PPAs at premium prices, stable-to-rising nuclear fleet output/availability, and constructive guidance on contracted power margins rather than merchant weakness. A kill signal would be absence of incremental AI/data-center deals, PPA cancellations or push-outs, softer capacity/pricing commentary, or guidance that implies power demand is not tightening versus the prior quarter.
To support the thesis that AI is bidding up scarce generation, VST must show rising realized power prices and/or new multi-year offtake tied to datacenter demand (especially ERCOT/gas-nuclear fleet), with management affirming tight capacity and no material PPA cancellations. A miss on commercial pricing, flat/declining forward hedges, or commentary that hyperscaler demand is slowing and interconnect/PPA pipelines are shrinking would undercut the scarcity mechanism and align with kill criteria.
in plain words: mkt cap what the whole company costs to buy · rev growth how much faster sales are running than a year ago · gross margin what is left of each sale before running the business · fwd p/e how many years of expected profit you pay for one share, and (t) means last year's profit where nobody forecasts next year's · fcf margin the slice of sales that ends up as spare cash · vs 52w high how far below its best price of the last year the share sits. these numbers refresh daily from Yahoo Finance.
a signal is one thing this bet depends on. the AI scores each one every weekday, from −2 (strongly against the bet) to +2 (strongly for it). what the world did counts triple, because a share price can move for any reason at all.
hyperscaler/datacenter power contracts — PPAs, nuclear deals, behind-the-meter agreements, gigawatt announcements
evidence the grid is the constraint — capacity auction prices, interconnection queues, turbine backlogs, restart timelines
30d relative performance vs XLU
strongly supports · supports · neutral · against · strongly against · one square per weekday