The $1.3 Trillion Data Center Boom Has a Power Problem

A market-research firm announced this week that the "Data Center Solutions Market" will surge from $535 billion this year to about $1.3 trillion by 2031, a 20% annual growth rate. The fast read is obvious: the AI buildout is unstoppable, so buy the sector and let the compounding do the work. The slower read is more honest, and it starts with who is doing the talking. This number is not an order book or a sales contract. It is one firm's projection of demand, produced by a company whose business model is selling reports about the future it is paid to predict. MarketsandMarkets counts servers, storage, networking, power, cooling, and the software that runs a data center, then extrapolates that the whole pile doubles and more.

That should be the first thing to distrust, not the last thing to assume. A demand forecast is only an investment thesis if the thing it forecasts is actually the thing that limits growth. Right now, it is not.

Demand isn't the gate; power is

The spending that would make that forecast real is, by most measures, already here. The four biggest hyperscalers — Alphabet, Microsoft, Amazon, and Meta — are on track to spend roughly $760 billion on capital expenditures in 2026, up from about $413 billion last year. That is real money in the ground, the actual fuel for everything the report counts. There is no demand problem.

The binding constraint is the power to run it. Access to grid electricity, not chip supply or construction, has become the thing that decides whether a facility opens this year or in 2029. A large fraction of the U.S. AI data centers planned for 2026 are facing delays or outright cancellation because of power shortages and transformer backlogs, and interconnection queues at many utilities now stretch well beyond five years. U.S. data-center power demand is projected to roughly double to about 66 gigawatts by 2027 — a volume that the grid, as it stands, cannot simply switch on. One utility, AEP Ohio, has already stopped accepting new data-center connections because "insufficient power infrastructure."

So a forecast that assumes the whole $1.3 trillion buildout happens on schedule is really assuming away the single hardest problem in the industry. The demand side is already solved; the deliverable thing is megawatts, and megawatts are scarce.

The market already smells the bill

Here is the tension the headline hides: the exact same spending that feeds the forecast is now the thing investors have started punishing. Late last quarter, Alphabet raised its 2026 capex guidance to about $205 billion. Its shares fell 7% on the announcement, and Amazon, Meta, and Microsoft fell alongside it — a real shift in behavior, because for two years investors had cheered every capex hike as proof of demand. No longer. Alphabet's free cash flow went negative for a quarter, and analysts expect Microsoft's to turn negative for the first time since 2001. Amazon's long-term debt is up sharply, and its free cash flow is negative as well.

This is the mechanics of the whole trade in one move: the industry is spending its cash-flow surplus ahead of any proven return. When the market treated higher capex as evidence of a bigger future, it was implicitly betting the compute monetizes. Now it is doing the accounting it skipped before — and the answer is that depreciation and debt are arriving before the AI revenue that is supposed to cover them. The forecast's 20% growth is real only if that calculus closes; investors are no longer granting it for free.

The same label, very different economics

"Data center solutions" is a genuinely useless bucket for deciding what to own, because the value inside it is distributed wildly unevenly. The largest slice of the report's $1.3 trillion is IT infrastructure — the servers, storage, and networking on the floor. That is also the least profitable part of the stack. The companies that assemble those boxes, like Super Micro, trade at roughly single-digit enterprise-value-to-EBITDA — the market prices them as commodity assemblers, because that is what they are: their revenue balloons with each AI order, but their margins barely move.

The economics concentrate in the constraint. Vertiv, which sells the power and liquid-cooling gear that a power-starved buildout cannot defer, trades at roughly 40 times EV/EBITDA, nearly four times Super Micro's multiple. That gap is the market's own verdict on where pricing power lives: not in the box that holds the chips, but in the electrical and thermal systems that the grid bottleneck makes scarce. And at the very top sits Nvidia, at a roughly $5.5 trillion market cap, where the market has already paid for a future that has to be enormous to justify the entry price.

The lesson is not to prefer any one of these. It is that a single "data center is surging" headline tells you almost nothing about which of them you are buying. The number the report leads with counts the boxes; the economics sit in the bottleneck and the compute.

What the headline is quietly telling you

The most useful thing in this press release is not the projection. It is the timing. A firm whose job is to sell optimism published a 20%-CAGR boom story in the same month that the equity market started marking down the very companies doing the spending — the highest-conviction signal yet that the crowd is asking whether this decade's buildout turns into this decade's profits. When the guy selling the map and the traders walking it disagree this sharply, the map deserves less weight, not more.

That is how a retail investor should consume every TAM headline: as a statement of how big the market could get if everything goes right, never as a statement of who profits if it does. Demand is not the open question. Power is. And the question the market is starting to charge for — whether the trillion dollars converts to earnings — is the one no report publisher is in a hurry to answer, because the answer shows up in somebody's stock price long before it shows up in a CAGR.