Speed to power: the AI trade's real bottleneck is a 2028 grid queue
Power queues, contracted nuclear and grid equipment are becoming AI's decisive bottleneck. This research maps speed-to-power beneficiaries and access routes for investors.
Home / Research
Speed to power: the AI trade's real bottleneck is a 2028 grid queue
The Texas pause, a 410-gigawatt queue, and why contracted nuclear became the safest AI trade that nobody calls an AI trade.
Key takeaway:Speed to power is the ability to secure contracted, dispatchable electricity before a new grid connection arrives. For AI data centres, that shifts value from chips toward nuclear generation, grid equipment and transmission services, because projects that get power online sooner can monetize scarce capacity years ahead of rivals.
The consensus
The first-order AI trade says the constraint is chips. Get the GPUs and the rest follows. That was true in 2024. It is not the binding constraint in 2026.
The argument
Electricity is. In the major US markets, connecting a new data centre to the grid takes three to seven years, and in some markets seven to ten. The queue is not a queue of real projects, either: in August, Texas paused new data-centre development pending an audit of the ERCOT interconnection queue, which held roughly 474 gigawatts of requests. ERCOT's own figure is around 410 gigawatts queued, 87 percent of it data centres, against a realistic summer peak for the whole state of 90 to 98 gigawatts. Most of that queue will be delayed or cancelled. Nearly half of the data centres planned for 2026 already have been, on transformer shortages and grid bottlenecks.
Meanwhile the real demand keeps coming. Goldman Sachs projects US data-centre power demand doubling from 31 gigawatts in 2025 to 66 gigawatts by 2027. Around 30 percent of planned US capacity is now shifting toward bring-your-own-power rather than waiting for the grid.
Here is the number that makes the trade. An AI data centre generates something like $10 to $12 million of revenue per megawatt per year. Getting a campus live two years ahead of a grid connection is therefore worth tens of billions of dollars to a hyperscaler, which is why they will pay roughly twice the wholesale price of power, around $110 per megawatt-hour, for electricity that is contracted, dispatchable and available now. "Speed to power" has become the single most important variable in whether a project happens, and the sellers of speed are not technology companies.
Who is exposed
Vistra (VST, NYSE). The clearest expression of the thesis. A 20-year agreement to deliver 1,200 megawatts from the Comanche Peak nuclear plant to Amazon Web Services from 2027, and a 20-year agreement with Meta for 2,609 megawatts across the Perry, Davis-Besse and Beaver Valley nuclear plants, with deliveries starting late 2026 and additional capacity phasing in through 2034, including 433 megawatts of corporate-backed uprates, the largest in US history. Alongside the nuclear, a $4.7 billion acquisition of Cogentrix added 5.5 gigawatts of dispatchable gas in the constrained PJM and New England markets. Vistra's executives said on their August earnings call that they support the Texas pause and do not expect it to touch the Amazon deal. Amazon is building a $5 billion data-centre project beside the plant.
Constellation Energy (CEG, Nasdaq). The largest nuclear fleet in the country, a raised full-year forecast after $7.5 billion of second-quarter sales, and the Crane restart. Constellation is the scale version of the same trade.
The picks-and-shovels of the picks-and-shovels. GE Vernova (GEV) for gas turbines and grid equipment, Eaton (ETN) for electrical distribution, Quanta Services (PWR) for the engineering and construction that turns a contract into a substation. These are the names that win regardless of which power producer signs the deal.
The second-order point
A 20-year power purchase agreement converts a merchant generator's most volatile cash flow into a contracted one, at a premium, for two decades. That is not an AI trade in the sense the market uses the phrase. It is a duration trade, and the counterparties are the most creditworthy companies on earth. The downside case for the nuclear names is muted growth, not decline: the contracted revenue is locked regardless of what happens to the AI cycle, and the PJM capacity revenue is locked with it. That asymmetry is what makes this the safest second-order trade in the stack.
How an international investor gets access
All of the names above trade on the NYSE or Nasdaq and are available through Interactive Brokers, Moomoo, Tiger and the other majors. For a diversified route, a US utilities ETF holds Vistra and Constellation but dilutes the thesis with regulated utilities that do not have the pricing power; a uranium ETF is a different trade entirely and should not be confused with this one.
What would change my mind
Four things. First, the queue really is speculative: if AI load growth stalls rather than accelerates, ERCOT scarcity events fade and the unhedged gas fleet earns less. Second, regulatory interference in the PJM capacity market or in the rules for colocating generation with large loads, which Vistra and Constellation are actively fighting at FERC. Third, valuation: the independent power producers have re-rated hard, and much of the easy multiple expansion is behind them. Fourth, politics: opposition to data centres has become a bipartisan midterm issue on the back of rising electricity bills, and a policy response that shifts costs onto generators would hurt.
Bottleneck status for this layer: Tightening, and it is the tightest link in the stack.
Not investment advice. Do your own research.
Written by Second Order by TKN for international investors. Educational research only — not investment advice. Do your own research.
Home / Research / Speed to power: the AI trade's real bottleneck is a 2028 grid queue
Read The Stack · Related layer: AI infrastructure layers
FAQ
What is speed to power? It is the ability to secure contracted, dispatchable electricity before a new grid connection is available.
Which companies benefit from the power bottleneck? Vistra and Constellation express the contracted nuclear trade; GE Vernova, Eaton and Quanta Services provide grid and generation equipment.
