Key Takeaways:
The future of AI data centers may be in orbit, with a top tech investor predicting the concept will be proven viable within 24 months, challenging hundreds of billions in terrestrial investment.
A top technology investor predicts that data centers in space will be economically viable within two years, beginning a campaign to capture significant market share from ground-based facilities by 2030. The forecast, from Atreides Management CIO Gavin Baker, suggests the AI infrastructure landscape could be radically reshaped, threatening the terrestrial supply chain that supports the industry’s massive power and cooling needs.
"I think in the next two years, its viability and economics will be proven," Baker said at the 2026 Sohn Investment Conference. "By the end of the decade, it will start to take meaningful market share."
The concept, which relies on near-limitless solar power and the vacuum of space for cooling, is already being explored by major technology firms. Google is reportedly in discussions with SpaceX and other rocket-launch companies for an orbital data center project, according to The Wall Street Journal. Meanwhile, SpaceX has an existing deal with AI lab Anthropic to build "multiple gigawatts of orbital AI compute capacity."
The prediction challenges the current investment paradigm, where the big four AI hyperscalers plan to spend $650 billion on capital expenditures this year alone. Amazon, the biggest spender, has plans for around $200 billion in 2026 to meet AI demand. A potential shift to orbit could redefine the winners and losers in the AI infrastructure build-out, moving the focus from ground-based real estate and utilities to launch providers and specialized satellite manufacturers.
The AI boom has triggered an unprecedented spending spree on data center infrastructure. Nvidia, the primary beneficiary, has seen its revenue climb to $215 billion from $27 billion just three years ago on the back of its powerful GPUs that have become the industry standard. This has fueled a massive build-out on the ground, with hyperscalers like Amazon spending heavily to keep pace. Amazon Web Services (AWS) is spending around $200 billion this year on capital expenditures, largely to satisfy commitments from major clients for AI computing capacity. This entire ecosystem, however, is predicated on solving the increasingly difficult constraints of power, cooling, and land for terrestrial data centers.
While the industry builds out based on Nvidia's chips, Baker argues that the most significant challenge to its dominance is being "severely underestimated." He identified Amazon's custom Trainium AI chip as the "dark horse" in the race. The reason, he explained, is technical: modern AI models increasingly use a "Mixture of Experts" (MoE) architecture, which requires a specific "Switched Scaleup Network" to run efficiently. "Global currently only has two companies with running switched scaleup networks—one is driving Nvidia GPUs, and the other is Amazon's Trainium," Baker said. He believes that after Trainium 3 enters mass production in late 2024, its standing in 2026 will be equivalent to Google's TPU in 2025. This puts Amazon in a unique position to compete directly with Nvidia on performance and cost for the most advanced AI workloads.
The idea of orbital data centers is moving from theory to practice, lending weight to Baker's aggressive timeline. Beyond the high-profile talks involving Google and SpaceX, smaller players are already building the foundational elements. Sidus Space (SIDU), a small-cap space and defense company, is currently building an orbital data storage payload for Lonestar Data Holdings. That payload, named StarVault, is scheduled to launch no earlier than the spring of 2027, representing a concrete, near-term step in creating a data ecosystem in orbit. These early projects provide tangible evidence that the technology and business models for space-based computation are actively being developed.
For investors, Baker's analysis suggests the AI infrastructure play is broadening beyond a single chipmaker. While Nvidia, trading at 28 times forward earnings, remains the dominant force, Amazon's custom Trainium chip presents a formidable challenger that could improve the parent company's margins and competitive position. Furthermore, the long-term viability of orbital compute, evidenced by projects from Sidus Space and interest from Google, introduces a new layer of risk for companies dependent on the current data center construction boom and its supply chain.
This article is for informational purposes only and does not constitute investment advice.
