The AI-Capex Bet in Reverse: What Microsoft, Meta and Alphabet's Prices Already Assume

The big hyperscalers are spending like never before. Microsoft, Alphabet, Amazon and Meta have together guided to roughly $725 billion of capital spending across 2026, most of it on AI data centers, chips and power, up about 77% on the prior year.

We have now put three of them, Microsoft, Meta and Alphabet, under the same capex lens, one post each. On our standard forward DCF, as of the 9 July 2026 model run, they land in three different places:

• Microsoft (full post) looks undervalued: model value about $483 against a price near $383.
• Meta (full post) looks modestly cheap: about $637 against $603.
• Alphabet (full post) looks expensive: about $252 against $362, roughly 30% overvalued.

The fair objection

Readers raised a good point about those posts. We focused on the cost of the capex, the cash going out the door, without giving much credit to the return it will eventually earn. If the AI build-out generates a decent return, revenue and profit rise, and these companies turn out cheaper than a cost-focused model makes them look. Rather than argue about that in the abstract, we decided to put a number on it.

Our model already prices in some return

First, one thing worth being clear about: the DCF is not blind to the payoff. It projects revenue off analyst estimates, and analysts have already built their view of the AI return into those numbers. For all three companies, consensus has revenue compounding at roughly 17 to 18% a year through 2030, an aggressive, AI-fueled forecast. So the return is already in the model, second-hand, through consensus growth.

The sharper question is this: at today's price, how much does each company actually have to deliver to be fairly valued, no more and no less? We ran the model backwards, holding the price fixed and solving for the assumption that puts our value exactly on the price. We did it two ways, one lever at a time: what if revenue grows faster or slower, and what if the margin changes. These are alternative routes to the same price, not a forecast; reality would be some blend.

What the prices require

Reverse DCF, 9 July 2026 model run. Consensus revenue growth is about 17 to 18% a year for all three.

Company	Model value	Price	Revenue growth to be fairly valued	vs consensus	Or EBITDA margin	vs today
Alphabet (GOOGL)	$252	$362	25.7%	+8.3 pts	52.1%	+14.2 pts
Microsoft (MSFT)	$483	$383	12.6%	-5.0 pts	46.0%	-8.7 pts
Meta (META)	$637	$603	16.6%	-1.2 pts	45.9%	-1.8 pts

The solve flexes one lever and holds the rest of the model at its standard settings, none of which we test here: a market-derived discount rate (roughly 9 to 10% WACC), 3.5% terminal growth, the default capex path (which glides today's elevated spending down toward maintenance, the "generous" case the capex posts walk through), and the usual tax, working-capital and share-count assumptions. Each moves the bar in a predictable direction: a lower discount rate, faster-fading capex, or higher terminal growth lowers the growth or margin a price requires, and the reverse raises it. We hold them fixed because the capex posts already cover the one that matters most.

Alphabet has to do better than the Street. To justify $362, revenue has to compound about 25.7% a year, roughly eight points above the 17.4% consensus. That is revenue reaching about $1.27 trillion by 2030, not the $900 billion analysts already project. Or the EBITDA margin has to climb from about 38% to about 52%. Either way, the price already assumes the AI return beats what analysts have penciled in.

Microsoft is the opposite: it has room to spare. It is already undervalued at consensus, so it can do worse and still be fairly valued. Revenue can grow about five points slower than consensus (12.6% instead of 17.6%, closer to $509 billion of 2030 revenue than the Street's $632 billion), or the margin can fall about nine points (to 46% from 54.6%), and the model still lands on today's price. That is the cushion the sell-off has built in.

Meta sits just on the line. It has a small cushion: revenue about one point slower, or margin about two points lower, and it is fairly valued. The price is close to what consensus already assumes.

The takeaway

Run this way, the return question stops being vague. The market is not treating the three the same. Alphabet's price only works if the AI spend pays off above consensus. Microsoft's price bakes in a discount to consensus, a built-in margin of safety on the return. Meta is priced for roughly what analysts already expect. Whether each bar is one you would clear is the research; these are candidates for it, not recommendations.

You can run the live model and set your own assumptions on each valuation page: MSFT | META | GOOGL