Listen to the audio version of this article (generated by AI).
Most investors think the AI infrastructure arms race is about chips. And chips matter enormously – Nvidia’s (NVDA) Blackwell GPUs, Broadcom’s (AVGO) custom AI accelerators, Marvell’s (MRVL) networking silicon. These are real, important, and worth owning.
But the constraint that is becoming the ceiling on AI buildout isn’t silicon… it’s watts.
A single next-generation AI training cluster (the kind that trains frontier models like GPT-5 or Claude 4) can consume 1 to 2 gigawatts of electricity. That is enough power to run a mid-sized American city. The hyperscalers, which include Meta (META), Microsoft (MSFT), Google (GOOGL), Amazon (AMZN), are each trying to build dozens of these clusters simultaneously around the globe.
The U.S. electrical grid was not designed for this. Utility interconnection queues – the waiting list to plug a new facility into the grid – now stretch 5 to 10 years in many regions. Environmental permitting adds more time. Transmission infrastructure is decades behind where it needs to be. And the laws of physics don’t care how much money you have: you cannot run a 2-gigawatt data center if there are only 500 megawatts available on the local grid.
This is what we call the resource-bound problem. And the companies that figure out how to route around this constraint will win the AI infrastructure race. Meta just showed you one of the most audacious routes imaginable.
[The resource-bound problem isn’t unique to energy, by the way. It’s showing up in payments, too – where legacy banking infrastructure is the grid, and a handful of tech companies are racing to route around it. Elon Musk’s X Money is one of the most ambitious attempts yet. We broke down the full investment story here.]
One company isn’t waiting for Washington to fix the grid, or hoping some utility builds a new transmission line in time. It went over all of it… literally.
That company is Meta, which earlier this week, announced something that would have sounded like science fiction five years ago.
The company struck a commercial agreement with a startup called Overview Energy to power its artificial intelligence data centers using solar energy collected in space and beamed back to Earth. Overview plans an initial orbital demonstration in 2028, with commercial power delivery beginning in 2030.
The world’s fifth-largest company by market cap is now contractually committed to receiving electricity from orbit.
Let that land for a second.
This is no simple corporate press release designed to make the ESG crowd feel warm and fuzzy but rather a commercial deal with a delivery schedule signed by the same company that runs Facebook, Instagram, and WhatsApp, and that is currently spending $60 billion to $65 billion on AI capital expenditure in 2025 alone. Meta does not sign orbital energy contracts for fun. Meta signed this contract because its engineers ran the numbers, looked at the global power grid, and concluded that Earth – the entire planet – cannot supply enough electricity to support their AI ambitions.
That is an extraordinary thing to admit. And it has profound investment implications.
Enter Orbital Compute: The Bull Thesis from the Ground Up
The core insight behind what we call the Orbital Compute thesis is elegantly simple:
Terrestrial infrastructure is resource-bound. Orbital infrastructure is technology-bound.
Resource-bound systems hit ceilings that don’t scale with innovation. You cannot Moore’s Law your way past a decade-long utility interconnection queue. But technology-bound systems – where cost and capability are driven by engineering progress – fall in price and improve in performance over time, just like semiconductors did.
Space infrastructure is technology-bound. Launch costs have fallen approximately 90% over the past decade, almost entirely due to SpaceX’s development of reusable rockets. Starship, SpaceX’s next-generation launch vehicle currently in testing, promises to drop costs another order of magnitude. As launch becomes cheaper and more reliable, putting infrastructure in orbit stops being a moonshot and starts being a business decision.
Space is actually a remarkable place to generate power. Solar panels in orbit face the sun 24 hours a day with no clouds, no night, no weather; producing roughly 8 to 10 times more electricity per panel than the same hardware on the ground. Cooling is easier too: heat simply radiates into the cold of space, no giant air conditioning units required. And for the heavy-duty AI jobs that don’t need to happen in real time – training a model, crunching a massive dataset – it doesn’t matter where the computer physically sits. Why not run those jobs in orbit, where the power is essentially free and unlimited?
Meta’s Overview Energy deal validates Phase One of this thesis: orbital power generation beamed back to terrestrial data centers. Phase Two – actual compute infrastructure in orbit – is the logical next step once the orbital energy and launch infrastructure matures. We are somewhere in the early innings of a multi-decade infrastructure buildout that eventually moves meaningful AI compute off-planet entirely.
The 5 Orbital Pure-Play Stocks to Buy
NASA ETF (NASA): The most direct way to own SpaceX in the public markets. The NASA ETF holds a basket of space-economy companies with meaningful SpaceX exposure embedded in the portfolio. SpaceX is the keystone of the entire orbital economy – it dominates launch, is developing Starship to push costs lower, and operates Starlink, the world’s most commercially successful satellite constellation. Nothing in the orbital compute thesis works without cheap, reliable access to orbit, and SpaceX owns that chokepoint more completely than any company in history. The NASA ETF is the cleanest public-market proxy for that monopoly.
Rocket Lab (RKLB): If SpaceX is the Boeing of orbital infrastructure, Rocket Lab is the Airbus – a credible, well-capitalized second player in the launch market with a track record of successful missions and an expanding product line. Rocket Lab’s Electron rocket handles small-payload launches reliably and affordably, and its larger Neutron rocket is in development to compete for medium-payload contracts. Beyond launch, the company is building out a full-stack space systems business – designing and manufacturing satellites, not just launching them. As orbital infrastructure demand scales, Rocket Lab is positioned to capture a disproportionate share of the second-tier launch market while building a recurring revenue base in satellite manufacturing.
Planet Labs (PL): Planet operates the world’s largest constellation of Earth-observation satellites, capturing daily imagery of the entire planet’s landmass. That sounds niche until you realize what it enables: real-time monitoring of supply chains, agricultural yields, infrastructure construction, military positioning, and environmental change at a scale no terrestrial system can match. Planet is essentially building a persistent, AI-queryable visual feed of Earth from orbit. As AI models get better at extracting signal from imagery, Planet’s data becomes dramatically more valuable. Think of it as a picks-and-shovels play on the intersection of orbital infrastructure and AI data.
AST SpaceMobile (ASTS): One of the more speculative but potentially transformative names in the space, AST SpaceMobile is building a constellation of large, phased-array satellites designed to deliver broadband connectivity directly to standard mobile phones – no specialized hardware required. The commercial implications are enormous: billions of people in underserved markets gaining reliable internet access, with telco partners including AT&T and Verizon already signed. Early satellites are in orbit and operational. This is a high-risk, high-reward name for investors who want direct exposure to the next phase of satellite infrastructure buildout.
Redwire Space (RDW): A less well-known name but an important one. Redwire manufactures specialized hardware for space environments – solar arrays, deployable structures, in-space manufacturing systems. As orbital infrastructure scales from a handful of satellites to a genuine off-planet industrial economy, companies that build the physical components of that infrastructure become critical suppliers. Redwire is the kind of name that looks small and obscure today and looks indispensable in 2035.
The 7 Overlapping AI Infrastructure Stocks to Buy
Here’s the important thing to understand about the orbital compute thesis: it doesn’t replace the terrestrial AI infrastructure trade. It extends it.
The buildout happening on Earth right now – hundreds of billions of dollars in data center construction, chip fabrication, power infrastructure, and networking – is the bridge that carries us to the orbital era. While Overview Energy prepares for its 2028 orbital demo, the hyperscalers are still building as fast as humanly possible on Earth. Meta’s deal confirms the long-term direction, but the near-term dollars are still flowing into terrestrial infrastructure.
That means the classic AI infrastructure basket remains as relevant as ever.
Nvidia (NVDA): Needs no introduction. Nvidia supplies the GPUs that power virtually every major AI training and inference workload on the planet. The Blackwell architecture extends the company’s dominance into the next hardware generation, and the insatiable demand from hyperscalers pouring capital into AI infrastructure means the order book stays full. Nvidia is the one stock in this basket where the bull case requires the least imagination.
Broadcom (AVGO): The sleeper hit of the AI infrastructure trade. Broadcom designs custom AI accelerators – XPUs – for hyperscaler customers including Google, Meta, and ByteDance. These custom chips are purpose-built for specific AI workloads and offer dramatically better efficiency than general-purpose GPUs for those tasks. Broadcom’s AI chip revenue has been growing explosively, and the company has guided for over $100 billion in AI-related revenue in fiscal 2027. It also dominates the networking silicon that connects AI clusters together, making it a dual beneficiary of compute and interconnect spending.
Marvell Technology (MRVL): Marvell’s story is similar to Broadcom’s but earlier in the earnings ramp. The company designs custom AI accelerators and high-speed networking chips for Amazon, Google, and Microsoft, among others. Its fiscal 2028 AI revenue target of $15 billion implies roughly 5x growth from current levels. Marvell is the higher-beta AI infrastructure play for investors who want more upside leverage on the custom silicon buildout.
Eaton Corporation (ETN): Every data center – terrestrial or, eventually, orbital – requires power management and distribution equipment. Eaton is the dominant supplier of that equipment globally. It makes the switchgear, busways, uninterruptible power supplies, and electrical distribution systems that sit between the grid and the servers. As AI data centers get bigger and more power-hungry, Eaton’s addressable market expands in direct proportion. It is the least glamorous name in this basket and among the most reliable.
Coherent (COHR) and Lumentum (LITE): As AI clusters grow to consume multiple gigawatts and process exabytes of data, moving information between chips and between servers at the speed of light stops being optional and becomes the central engineering challenge. Coherent and Lumentum manufacture the optical transceivers and components that make high-speed data transmission inside data centers possible. The shift from copper to optical interconnects at shorter and shorter distances – driven by AI’s insatiable bandwidth demands – is a multi-year secular tailwind for both companies.
Astera Labs (ALAB): The newest name on this list and one of the most compelling. Astera Labs designs semiconductor connectivity solutions – CXL memory expansion, PCIe retimers, and optical interconnect controllers – that solve the data bottleneck between processors and memory inside AI servers. As AI models get larger and memory bandwidth becomes the binding constraint at the chip level, Astera’s products become increasingly mission-critical. It’s a pure-play on the AI interconnect buildout with a differentiated product portfolio and hyperscaler customers already in production.
The Bottom Line
Meta’s Overview Energy deal is a data point of extraordinary significance, and investors who dismiss it as a curiosity are missing the forest for the trees.
This is not a company experimenting with speculative technology to look innovative. This is a company that has already committed over $60 billion to AI infrastructure spending this year alone, looked at its power options, and concluded that going to space was the rational choice. When the hyperscalers start sourcing electricity from orbit, the old investment frameworks for AI infrastructure need to expand.
The orbital compute bull thesis rests on a simple, durable premise: Earth is running out of room to build AI, and the companies solving that problem from orbit will be among the most important infrastructure investments of the next two decades. Meta just handed us the clearest confirmation signal yet that the thesis is not only correct, it’s already being acted on at the highest levels of corporate capital allocation.
The question, as always, is whether you’re positioned before the rest of the market notices. The companies solving that problem from orbit are still flying under most investors’ radar, but there’s another infrastructure revolution unfolding right now… one that could be just as transformative, and just as underfollowed.
I just released a full presentation on X Money — Elon Musk’s long-awaited attempt to turn X into America’s first super-app bank. The same way SpaceX cracked open the orbital economy, X Money could crack open the payments economy. And just like orbital compute, the real opportunity isn’t in the headline name — it’s in the public companies powering it behind the scenes.
