GE Vernova and the AI-Electricity Basket

Not advice. GE Vernova stays where Friday put it. Talen is coverage only. This note does not hire or fire names. It does not re-argue uranium; Cameco's fuel-plan split from 13 August 2026 stands.

The AI-electricity tape treats gas turbines, contracted nuclear megawatts, and a uranium fuel book as one power-stock basket. When the hyperscaler capex print runs, GE Vernova, Constellation, Vistra, Cameco and Talen move together. When it cools, they cool together. The sorting variable is the megawatt.

The megawatt is the wrong variable for half the list. A hyperscaler that has already signed a 20-year nuclear PPA with Constellation is not shopping the same megawatt next Tuesday. A manufacturer that has already taken a slot reservation for a gas turbine, and a producer that has already contracted the pounds that fuel the plant, are not the same instrument as a name whose offtake is one plant and one customer. Treating those as one trade is the AI-electricity version of sorting uranium by the spot price: a label that is easy to see and does not predict the thing the framework is built to measure.

Five covered names sit inside the sentence "AI electricity." One of them has already sold the kit — turbines, slots, transformers — into a hyperscaler build. Two of them have already sold the megawatt into 15–20 year nuclear PPAs. One of them has already sold the pounds that fuel the plant, which is a fuel-plan fact the Cameco note already sorted and this page will not reopen. One of them is an honest megawatt-beta: one Amazon PPA on one plant. The scorecard below is that split applied to the framework's scores, so the comparison is visible before any prose tries to win it.

Two things fall out immediately. First, the framework does not treat this as one trade: the name that has already sold the kit, the names that have already sold the megawatt, and the name that is still a call on a single plant's offtake do not live in the same quality band, and they should not, because one group is already a line item in a hyperscaler's build and the other is a wager that the megawatt will still be there when the offtake finishes ramping. Second, GE Vernova is not the name the screen is warning you about. The disagreement the AI-electricity spread is built to find — a category label covering very different businesses — is Talen, and only as coverage. GE Vernova is a name Friday already put in the book. The AI-electricity tape does not draw that line.

Read the first column against the last. GE Vernova, Constellation and Vistra have already sold the thing the AI-electricity tape is supposed to pay for — kit, megawatts — into contracts a customer cannot casually unwind. Cameco has already sold the pounds; that split lives in the Cameco note and is not re-tried here. Talen has sold one offtake on one station. The current print does not show five names in one electricity trade. It shows who has already sold the megawatt or the kit into a hyperscaler build, and who is still a call on the megawatt.

The slot book is the load-bearing line. Equipment-plus-slots moved from 100 to 116 GW in the quarter; the company now targets at least 125 GW by year-end. It signed 20 GW in Q2 — 18 of that slots, 2 of it orders — converted 10 GW of existing slots into orders, and shipped 3 GW. That is a manufacturer already inside a hyperscaler's build schedule, not a call on electricity demand. The HA fleet underneath it is a service relationship: 130 units commissioned in 21 countries, 195 under contract, four million commercial operating hours, as the July 22 company news post states it.

Electrification is the second half of the same sale. Orders of $6,347 million against $3,283 million are +93% reported and +66% organic; Prolec's $860 million is the gap. Equipment backlog is $40.6 billion, up 69% year-over-year, including $5 billion from Prolec. Data-center orders are more than $5 billion year-to-date, more than double 2025. Book-to-bill is about 1.7. Power orders of $16.7 billion, +134% organic, with 52 heavy-duty units including 15 HA and 61 aeroderivatives, is the turbine half of that bundle. Company adjusted EBITDA printed $1.25 billion — the $1.2 billion highlight is rounding — and free cash flow was $5.1 billion in the quarter.

SMR is optionality, not a second order book. Darlington BWRX-300, a 14 March Hitachi Southeast Asia MoU, and two technology selects on the Q2 call are real options. They are not a sold megawatt and they are not a sold kit. This 8-K does not convert them into either.

Wind H1 $(657) million has to sit on this page, because it is the honest residual inside a kit name. Full-year Wind EBITDA losses are guided at about $400 million. The first half has already overshot the year. Q2 was $(275) million. Orders of $1.2 billion fell 40% organically; revenue of $2.0 billion fell 10%, 11% organically. A reader who stops at Power orders and Electrification book-to-bill will conclude the equity is a clean AI-electricity compounder. The Wind line is why it is not. The kit still sold. The residual still lost more in six months than the company said it would lose in twelve.

The megawatt side of the basket is not a second GE Vernova. Constellation has already sold 20-year nuclear PPAs to Microsoft, Meta and Alphabet, and in Q2 added 920 MW of 15–20 year nuclear PPAs. Vistra has already sold Amazon 1,200 MW and Meta 2,609 MW, and Helix sits with KKR, NVIDIA and KIA as a powered-land channel. Those are contracted megawatts a hyperscaler cannot casually unwind. They fail on a different clock than a slot reservation: a PPA delay of more than 500 MW is the trip this piece named; a slot that does not convert is GE Vernova's.

Cameco is in this cohort because the tape puts uranium in the same AI-electricity sentence, not because this page has a second uranium essay. The 13 August note already sorted it: more than 28 million pounds a year contracted through 2030; Q2 realized US$67.79 against an $85.18 spot is old ceilings, not a premium. That split stands. This page does not reopen it.

Talen is the control. One Amazon PPA, 1,920 MW through 2042. One plant, about 90% of 2.5 GW Susquehanna. That is real offtake and real concentration. It is coverage only — an honest megawatt-beta, not a hire. Pricing Talen as AI-electricity beta is not a misread of Talen. It is the business. The framework's complaint is not that the tape is wrong about Talen. It is that the tape has bundled GE Vernova, Constellation and Vistra into the same trade.

Transaction embedding is the column that does the work. It is the slot book and the LTSA at GE Vernova, the 15–20 year nuclear PPA at Constellation and Vistra, the multi-year uranium book at Cameco — and it is a single-hyperscaler ramp at Talen, which is the framework saying one offtake on one plant is not in the customer's transaction the way a 20-year PPA book or a converted slot is. Regulatory lock-in does not sort the cohort. It is strong or intact for every name here, because an air permit and an HA-class factory, an NRC licence, and a CNSC mine licence are all real barriers. That is the column the tape is pricing, and it is the column that cannot tell GE Vernova from Talen. Bundling is the secondary tell: GE Vernova can sell generation-to-grid; Constellation and Vistra sell the megawatt; Cameco can sell a Western fuel-cycle package; Talen sells offtake from one station.

• GE Vernova. The 116 GW equipment-and-slot book, moving toward ≥125 GW by year-end, is the switching cost. Electrification orders of $6,347 million against $3,283 million, with Prolec's $860 million in the reported gap and data-center orders above $5 billion year-to-date, is why a hyperscaler that wants the turbine and the transformer is not substituting a single-product competitor mid-build. Wind H1 $(657) million is the residual that equity still has to carry.
• Constellation. The renaissance as cash flow is a 20-year PPA on an existing reactor, not a turbine in a slot. Q2's additional 920 MW, on top of Microsoft, Meta and Alphabet already in the book, is the demand side of the same embedding test. Replacing Constellation requires 24/7 carbon-free baseload that does not exist at scale elsewhere.
• Vistra. Amazon 1,200 MW and Meta 2,609 MW are the contracted megawatts. Helix with KKR, NVIDIA and KIA is a channel to co-develop powered sites, not a second PPA book. The nuclear fleet is smaller than Constellation's and the Texas retail stack is the hedge the pure-play generator does not have.
• Cameco. McArthur River and Cigar Lake are the geology. More than 28 million pounds a year already contracted through 2030 are the switching cost. Q2 realized US$67.79 against $85.18 spot is last cycle's ceilings, not a scarcity premium. The fuel-plan test lives in the Cameco note; this page carries the ticker so the AI-electricity tape cannot smuggle uranium back in as a fourth megawatt.
• Talen. One Amazon PPA, 1,920 MW through 2042, on a plant that is about 90% of 2.5 GW Susquehanna. That is real offtake. It is also one customer and one station. Coverage only. Honest megawatt-beta. Not a hire.

The AI-electricity frame is not confused. It is correct about Talen, it can be correct about GE Vernova on any quarter the slot book stops filling, and it has a serious claim on the whole list that the matrix will not show you: the bid is one bid. Those are different failures, and the article that pretends they are the same is doing the tape's job rather than the framework's.

Talen is the cohort member where the other side is right, and it is right all the way down. One Amazon PPA. One plant, about 90% of 2.5 GW Susquehanna. A 17-year offtake that still has to ramp to full volume no later than 2032. If AI load keeps clearing through PJM at a premium, that empty second customer is the highest-torque way in this coverage set to own the megawatt. If Amazon holds deliveries or a Susquehanna outage takes the station offline, the same book is the whole equity. Pricing Talen as megawatt-beta is not a misread. It is the business. The framework's complaint is not that the tape is wrong about Talen. It is that the tape has bundled GE Vernova into the same trade.

The more dangerous counter is the one that can hide inside GE Vernova's own print. Wind H1 EBITDA of $(657) million has already overshot a full-year guide of about $400 million of losses. Q2 was $(275) million. Orders fell 40% organically; revenue fell 10%, 11% organically. On those facts the AI-electricity frame has a point this article has to take in full: the current half does not show a clean kit compounder, and the residual that can still re-rate the equity sits in the segment the tape is not pricing. If first-half losses are the run-rate rather than the overshoot — if Onshore stays soft and Offshore project costs keep printing — then GE Vernova's slot-and-Electrification story is a Power print sitting on top of a Wind P&L the year has already spent. The kit still sold. The residual can still be the stock.

The third counter is the one-bid argument, and it is the one the two trips are written to catch. Hyperscaler electricity demand is a single capex committee. If that committee slips, a slot reservation and a 20-year PPA both get revisited, and the tape's basket was the right instrument. Siemens Energy and Mitsubishi Power are the substitution the slot book is supposed to have already beaten; a year-end equipment-plus-slots print still at 116 GW against a ≥125 GW target would mean the kit was reserved, not bought. A Constellation or Vistra disclosure that a hyperscaler PPA of more than 500 MW is delayed or cancelled, while GE Vernova slots still fill, would mean the megawatt names were the ones the committee walked away from. Either print would make this page wrong. Neither has printed. Watch the year-end slot book and the next PPA 8-K, not the five equities on a capex-up day.

The useful output of the screen is not a ranking of five power names. It is a rule for the next time the tape puts a sold kit and a one-plant offtake in the same AI-electricity trade. GE Vernova's Q2 did not show a business that is still a call on the megawatt. It showed a business whose gas equipment and slots already sit at 116 GW against a ≥125 GW year-end target, whose Electrification orders are +66% organic and +93% reported with Prolec's $860 million in the gap, whose data-center orders are already more than $5 billion year-to-date, and whose Wind residual has already spent the year at $(657) million in the first half. The valuation pillar on the analysis page is the live read on whether the tape paid you for that distinction. This page is the argument that the distinction is real. Nothing here is financial advice.

Friday already put GE Vernova in IM25. This page does not reopen that book. Talen is coverage only — an honest megawatt-beta, not a hire. Cameco stays where the 13 August note left it.

The lesson is the same one the uranium-beta piece ended on, pointed at a different sorting. When a category-wide narrative re-prices the sold kit and the one-plant offtake identically, the framework's job is to find the names where the narrative does not fit the book. On this cohort the AI-electricity tape was the right read on Talen. It was the wrong read on who has already sold it.