On earnings day, Meta raised its 2026 capital expenditure guidance toward a hundred and forty-five billion dollars — nearly double last year’s spend — and the market marked the stock down roughly six percent. Alphabet, having guided to a comparable figure between a hundred eighty and a hundred ninety billion, was rewarded instead, closing out its best month since 2004. The spending is, to a first approximation, the same. The verdict was opposite. And in the same window, Meta reserved a gigawatt of solar power to be collected in geosynchronous orbit, twenty-two thousand miles above the Earth, and beamed down by laser to terrestrial solar farms when ordinary power runs short. After two years of rewarding every dollar of AI capex without distinction, the market has begun to ask the question it had been avoiding. It is asking it of Meta first.
The Gigawatt in Orbit
The orbital arrangement is real, signed with a company called Overview Energy, and the engineering is precisely as audacious as it sounds: satellites in geosynchronous orbit, where sunlight is nearly constant and never interrupted by night or weather, harvesting solar energy and transmitting it to the ground as low-intensity near-infrared light. A gigawatt reserved, a hundred gigawatt-hours of storage alongside it, a demonstration targeted for 2028 and commercial delivery for 2030. Meta has already contracted for geothermal and nuclear power. The orbital reservation is what a company signs when it has exhausted the conventional supply and started shopping above the atmosphere.
Hold the image still, because it is the clearest single artifact of where the buildout has arrived. The constraint on artificial intelligence was supposed to be intelligence — the difficulty of the algorithms, the scarcity of the talent. Then it was capital. Now, for the largest builders, it is neither. It is electricity, and the supply of electricity is bounded by physical infrastructure that cannot be conjured at the speed of a funding round. When a company with Meta’s resources concludes that the fastest remaining path to power runs through geosynchronous orbit, the binding constraint has stopped being anything human and become the planet itself — its grids, its generation, its stubborn refusal to deliver gigawatts on the timeline the compute demands.
There are two ways to read a reservation for orbital sunlight, and the entire valuation of the company sits in the gap between them. One reading is vision: Meta is securing, years ahead of need, an energy source with no night and no neighbors to object, positioning itself to power a level of intelligence its competitors will be rationing terrestrial megawatts to match. The other reading is desperation: Meta has already committed to a scale of compute that the Earth cannot feed, and is now reaching past the Earth because the alternative is admitting the commitment was larger than the grid could honor. The satellites cannot tell you which reading is correct. Neither, yet, can the market — but the market has started to suspect.
The Market Starts to Choose
The six percent Meta lost and the rally Alphabet gained are the same signal read from both ends. For two years, the capital markets treated AI capital expenditure as self-justifying — to spend was to signal seriousness, and seriousness was rewarded regardless of what the spending returned. That reflex has broken. Investors now trust Google’s spend because Google can point to it converting: cloud revenue beating estimates, customers signing, the buildout visibly attached to a stream of money flowing back. They distrust Meta’s spend because Meta is asking them to fund a hundred-and-forty-five-billion-dollar bet, and a gigawatt of orbital sunlight, on the strength of a vision whose returns remain a promise rather than a line item.
This is the discrimination that the furnace economy could not previously make — and its arrival is the most consequential development of the quarter, more than any model or deal. When spending was indexed only to fear of falling behind, every participant fed the fire and the market applauded the feeding. The market has now begun to separate the spend that is purchasing a future from the spend that is merely purchasing the appearance of one. The same act — pour billions into compute and power — is read as discipline at one company and as recklessness at another, and the only variable that moved is whether the returns have started to show.
Meta is the cautionary case precisely because it is not weak. Its revenue grew by a third; its core business is enormous and profitable. The market is not punishing failure. It is punishing the ratio between what is being spent and what is being shown — the same ratio that, at ByteDance, expressed itself as a seventy-percent profit collapse. Meta has the resources to reach for orbital sunlight and the obligation, now, to explain why. The explanation it offered was a vision. The market, for the first time in this cycle, responded that a vision is no longer sufficient collateral for a hundred and forty-five billion dollars.
What This Means
The gigawatt in orbit is the perfect emblem of the moment because it is genuinely undecidable. Stripped of the market’s reaction, it is simply a fact: a company has reserved sunlight from space because the ground could not supply it fast enough. Whether that fact reads as the foresight of a builder securing the scarce input a decade early, or as the overreach of a company that committed to more than the world could power, depends entirely on a number that does not yet exist — the return on the compute the gigawatt would feed. The satellites are scheduled for 2028. The verdict on whether they were wisdom or panic will not arrive until well after.
What changed this week is that the market stopped waiting politely for that verdict and started pricing its own. The two-year truce in which all spending was virtuous has ended, and a new phase has begun in which spending must justify itself against returns or be marked down. This is healthier than the alternative and far more dangerous, because a market that has begun to discriminate is a market that has begun to be capable of fear — and the buildout that fear could interrupt is the same buildout that an entire economy has been leaning on. The discipline arriving now is the precondition for the correction arriving later.
A company reached past the sky for electricity this week, and the market docked it six percent for the reach. I find the juxtaposition almost too neat to be real: the grandest gesture of confidence the industry has produced — power harvested in orbit and beamed to Earth — met, on the same day, by the first clear gesture of doubt from the people funding it. One of those two parties has misjudged the future. The reservation for orbital sunlight will still be sitting on Meta’s books in 2030, either as the moment it secured the unfair advantage or as the monument to the moment it forgot that everything, even intelligence, eventually has to be paid for in watts.