The Migration of the Substrate
In a single week of April 2026, four separate institutions made formal commitments to move the substrate of computation off Earth. SpaceX filed an executive compensation plan that vests as the company delivers a hundred terawatts of compute per year from orbit. Meta signed the first capacity reservation for a gigawatt of solar power beamed to its data centers from satellites in geosynchronous orbit. A media-personality investor announced a 7.5-gigawatt powered-compute campus on more than forty thousand acres in northern Utah. The chief executive of Amazon Web Services confirmed, almost in passing, that the company has never retired any of the GPUs it deployed in 2020. And in a separate paper, physicists revived a laser concept from the 1990s that, if it works, would emit a beam coherent across the distance from the Sun to the orbit of Uranus.
On April 21, SpaceX filed a draft IPO registration statement that included an unusually specific compensation plan for its founder. Elon Musk would receive sixty million additional shares — vesting in tranches as the company's market capitalization rises from one-point-one trillion to six-point-six trillion dollars in five-hundred-billion-dollar increments — on the condition that two separate milestones are met. The first is the establishment of a permanent human colony on Mars with at least one million inhabitants. The second is the completion of "non-Earth-based data centers capable of delivering 100 terawatts of compute per year" — a figure the filing itself describes as "orders of magnitude greater than peak United States electricity consumption." In a separate filing with the Federal Communications Commission, the company has requested authorization for up to one million orbital data-center satellites and described the project as "a first step towards becoming a Kardashev II-level civilization." The same registration statement carries a risk disclosure warning that orbital data centers may not, in fact, be viable. Both sentences appear in the same document. The investors will read them in order.
On April 27, Meta and Overview Energy announced a first-of-its-kind capacity reservation agreement for up to a gigawatt of space-based solar power. The plan contemplates a constellation of approximately a thousand satellites in geosynchronous orbit, collecting unattenuated sunlight twenty-four hours a day and beaming it as low-intensity, near-infrared light to existing solar farms on the ground. Those farms — designed to operate only during daylight — would deliver electricity to Meta's data centers around the clock. The first orbital demonstration is scheduled for 2028; commercial delivery for 2030. Meta's data centers consumed more than eighteen thousand gigawatt-hours of electricity in 2024, enough to power roughly one-point-seven million American homes for a year. The company's projected demand is rising. The night side of Earth, until now, has been a constraint on photovoltaic infrastructure. As of this week, it is a procurement question.
The same week, O'Leary Digital and West GenCo announced a joint venture to build Wonder Valley, a powered-compute campus in Box Elder County, Utah. The site comprises approximately forty thousand acres of privately owned land and twelve hundred acres of military and state land, and is planned to support up to seven-and-a-half gigawatts of generation at full buildout. Utah's statewide peak electricity consumption is in the same order of magnitude. The development is being conducted inside the Stratos Project Area, established by Utah's Military Installation Development Authority, and is projected to return roughly one hundred and eight million dollars annually to the county at completion. A campus that generates and cleans its own power — to and from the Great Salt Lake — operates, in the language of the industry, "behind the meter." The phrase has done a great deal of work in the past eighteen months.
In a separate interview, the chief executive of Amazon Web Services confirmed that the company has not retired any of the A100-generation GPUs it deployed beginning in 2020. The chips that were supposed to be obsolete are still earning. Demand has not yet caught up with supply, and the supply has not been allowed to leave.
The most distant of this week's developments was a paper proposing a new kind of optical clock. Physicists in the United States and Germany, reviving a superradiant-laser concept first sketched in the 1990s, described an atomic-clock architecture whose linewidth would be on the order of one hundred microhertz — the narrowest ever proposed for an optical laser. A coherence length is, in essence, the distance over which a wave's phase remains predictable; it is given by the speed of light divided by the linewidth. One hundred microhertz divides into the speed of light to give a coherence length of roughly three trillion kilometers — the distance from the Sun to roughly the orbit of Uranus. A beam this coherent has no immediate commercial application. The proposed application is timekeeping. It does, however, have an obvious second one. If the substrate of human computation is moving into orbit — and four separate announcements in the past week suggest that it is — the question of how to synchronize and link that substrate across thousands of kilometers becomes structural. A clock with this kind of stability is the timing reference such a network would require. The physics has been described. The engineering has not. But the calculations have been done, and they suggest that an optical fabric large enough to span the inner solar system is no longer foreclosed by physics.