Why Every Megawatt

Compute is the product. Power is the constraint.

The build-out of AI is the largest infrastructure contest of the era, and it is being fought on three levels at once.

Nation-states are treating compute the way they once treated oil, steel, and enriched uranium — as strategic infrastructure. Sovereign-AI programs, export controls, and energy policy are converging on a single recognition: the capacity to train and run frontier models is national power, and it is rate-limited by the capacity to generate electricity.

The largest companies in the history of the world are committing capital to this race at a scale corporate balance sheets have never carried — and they are doing it because of the game theory, not in spite of it. If you believe the frontier compounds — that the leading model pulls away while the runner-up's lead decays — then there can only be one, and no single actor can afford to be the one who under-built. That belief turns restraint into a losing move. Everyone races. Everyone over-procures. Demand for power runs structurally larger than any one company's needs would justify, because each is buying insurance against the others.

On the ground, the same race is fought house by house. Localities fight the substations, the transmission lines, the water draw, and the hit to ratepayers. Interconnection queues stretch for years. Moratoria appear. The megawatt that is strategic in a capital is contested at a zoning hearing.

One thing is constant

Through every level — geopolitical, corporate, local — one variable does not move. Power is the bottleneck. Chips get unstuck as fabs come online. Buildings, cooling, and GPUs can be financed in months. Electricity cannot. It is finite, slow to build, and physically tied to a place. Whoever controls firm, energized power near fiber controls the build-out — regardless of who trains the model.

There is more than one way to win it

How a company secures power is a fault line running through the entire sector, and it is where much of the real risk lives.

The defining fork is grid-secured power versus on-prem generation. Wait in the utility interconnection queue for grid power — cheaper, cleaner narrative, but slow and at the mercy of a queue you don't control. Or bring your own generation behind the meter — gas turbines today, nuclear and SMRs on a longer horizon — fast and controllable, but capital-heavy and carrying fuel and emissions exposure. Speed versus dependence. It is the defining bet many of these companies are placing right now.

Other fault lines run alongside it: brownfield sites that can be energized fast — an ex-miner's existing interconnect — versus greenfield builds that start at the back of the queue; owned power versus leased megawatts rented from a landlord; merchant versus regulated markets. Each is a different answer to the same question: how do you get firm power, and how exposed are you to someone else for it.

What it all resolves to

Strip away the geopolitics, the capex, and the zoning fights, and every one of these contests reduces to a single number — megawatts controlled, and how firm they are. That is the asset.

It sits in an unlikely cast of hands: ex-Bitcoin miners, land developers, and energy operators who happened to hold grid connections when the demand arrived. The market is repricing them in real time, and the most valuable power positions are not held by the most famous names in AI. Nobody covers this the way an oil analyst covers reserves — by what a company controls, not what it produced last quarter. That is the gap. That is what we built.

What we measure

One headline number per company: gigawatts owned and contracted. Against it, one price: market cap per gigawatt — Val/GW, our price-to-reserves.

Val/GW will not tell you who has the best software, the cleanest balance sheet, or the highest utilization. It tells you who the market thinks owns the future of the grid, and what it is paying for the privilege. Cheap reserves and expensive reserves both tell you something. The ledger is built to let you see which is which.

The discipline that makes the number honest

A reserve figure is only worth anything if it is honest about what counts. So we split every megawatt into three tiers and never blur them:

• Operational — energized, earning today.
• Contracted — signed, not yet drawing load.
• Announced — pipeline, stated intent.

Most coverage of this sector runs on press releases, which collapse all three into one inflated headline — the announced gigawatt dressed up as a delivered one. We don't. A gigawatt announced is not a gigawatt delivered, and the distance between a press release and an energized substation is where most of the risk in this trade lives. The leaderboard shows operational reserves and total reserves side by side, on purpose.

How we source it

Primary and reputable sources only — SEC filings (10-K / 10-Q / 8-K for domestic issuers; 20-F / 6-K for foreign private issuers), company IR pages, monthly operational updates, and named-outlet reporting. Every figure carries an "as of" date and a source. Where a number cannot be backed, we say so and leave it flagged rather than invent support for it.

What this is not

This is not investment advice, and it is not a stock tip. It is reserve accounting for the power behind AI — analysis, framed as analysis, with the uncertainty stated plainly.

Every megawatt will be monetized. Our job is to tell you what it is worth before they do.