Two Ways to Solve the Same Problem

Every AI investment thesis eventually collides with the same wall: the power bill. One data center can draw as much electricity as 840,000 homes, the grid can’t keep up, and communities from Pennsylvania to California are pushing back. The money currently flowing to solve this is enormous — but it’s splitting into two very different bets.

Bet one: solve the power problem on Earth. That’s where most of the capital is going. Bet two: route around the problem entirely by putting data centers in orbit, powered by unlimited solar energy. That’s James Altucher’s Musk Millionaire thesis, and it’s the more interesting side of the trade to understand.

The Terrestrial Playbook

The presentation lays out what the tech giants are actually doing, and it reads like a panic:

  • Amazon is putting $20 billion into Pennsylvania for AI, including nuclear
  • Microsoft is restarting old nuclear plants — including, the presentation notes drily, their “old disastrous nuclear program”
  • Meta is funding seven gas plants to power its biggest data center

Add it up and you get the mainstream AI power trade: nuclear developers, uranium miners, gas producers, turbine and grid equipment makers, and independent power producers. It’s a picks-and-shovels story with real policy tailwinds and real cash flows behind it. We’ve covered this angle before in our small modular reactor stocks article and our new power grid analysis, and it remains one of the most crowded themes in energy investing.

The catch, per the presentation: 30 states have filed over 300 bills to slow data center builds, $98 billion in projects have been killed by local opposition, and electricity costs in some affected areas have tripled. Terrestrial power for AI is a fight, every single time.

The Orbital Alternative

Altucher’s counter-case, built around an FCC filing through SpaceX, is space-based solar powering AI compute directly in orbit:

  • Solar panels in space work “up to eight times better than on Earth” — no clouds, no weather, no night, uninterrupted sunlight more than 99% of the time
  • Cooling is free: space sits at 250 degrees below zero
  • No grid connection, no permitting fights, no zoning boards, “no town that can tell him no”
  • The plan scales to one million solar-powered AI satellites, with the 230-foot-wingspan AI1 already built

The economics, per SpaceX’s own quoted words in the filing: “Within a few years the lowest cost to generate AI [power] will be in space.” If that’s even directionally right, the entire terrestrial scramble — the nuclear restarts, the gas plants, the grid upgrades — is fighting for the smaller share of AI power demand that stays on the ground.

The investment expression is different too. The terrestrial trade is mostly large, cash-flowing utilities and industrials. The orbital trade, in Altucher’s telling, runs through SpaceX’s supplier chain: the confirmed Terafab partnership with Intel (up over 100% since the April announcement), the unnamed sub-$100 “AI MasterKey” transmission supplier already producing 5 million chips per day, and the “Elon Eleven” list of suspected and confirmed suppliers across Tesla, SpaceX, xAI, Starlink, Neuralink, and the rest of the empire. See our Musk Millionaire review for the full breakdown.

The Obvious Objection

Space-based solar power is not a new idea, and it has a long history of not penciling out. Launch costs were always the killer. The bull case now rests on reusable rockets collapsing that cost — and, notably, on the only company with a fully reusable heavy launch system also being the company planning the satellites. The vertical integration is the whole argument: SpaceX doesn’t need space-based power to be cheap for everyone, just for itself.

The rivals’ timid progress is the presentation’s best evidence that the barrier is real: NVIDIA-backed Starcloud has launched a single refrigerator-sized satellite, Google’s Project Suncatcher is two test satellites that won’t fly until 2027, and Blue Origin’s Project Sunrise is grounded with Bezos admitting a twenty-year timeline. Meanwhile the presentation counts roughly 100,000 available low Earth orbit slots, of which Elon is racing to claim as many as possible.

Our Read

These are not mutually exclusive trades, and that’s probably the right frame. Even under the most aggressive orbital scenario, Earth-based AI power demand keeps growing for years before any meaningful compute migrates skyward. The nuclear and gas names monetize the near term; the space-supplier names are a longer, more speculative option on the migration actually happening.

What the Musk Millionaire presentation contributes, regardless of whether you buy the stock picks, is a sharp question worth asking about any AI power investment: does this thesis assume the data center stays on the ground? Because the highest-conviction bet in the presentation is that, over the next decade, some of them won’t.

For the satellite plan itself, see our FCC filing breakdown; for the energy math, our AI energy bottleneck article; and for the “SpaceX isn’t a rocket company” framing that ties it all together, our trojan horse thesis article. Jason Bodner’s photonics thesis is another worthwhile contrast — a different analyst solving the same AI bottleneck with light instead of orbit.

NewsletterVetter is an independent publication. We receive compensation from some of the services we review through affiliate links. Nothing on this site is investment advice. Always do your own research.