Verdict: The next three years of AI supremacy will be won on the ground, not in orbit. While SpaceX and Blue Origin chase the "infinite energy" of space, Masayoshi Son and OpenAI are committing over $500 billion to terrestrial, gigawatt-scale campuses. For small businesses and builders, this means compute availability is shifting from a global commodity to a regional, sovereign asset that you must secure today.
Last verified: June 29, 2026 · Core Conflict: Terrestrial Scale (SoftBank) vs. Orbital Efficiency (SpaceX) · Volatile Facts: GPU availability and power contracts change monthly.
Why Masayoshi Son calls orbital compute a "distraction"
In June 2026, SoftBank founder Masayoshi Son publicly dismissed the idea of orbital data centers—a concept aggressively promoted by Elon Musk's SpaceX. Son's argument is rooted in cold economics: while space offers constant solar power, electricity accounts for only 7% of data center operating costs, while the silicon (NVIDIA chips) accounts for 93% [1].
In the "cutthroat battle" for AI dominance, Son argues that the next few years are critical. Waiting a decade for orbital infrastructure to scale is a luxury the industry cannot afford. Instead, SoftBank is doubling down on "boring" terrestrial advantages: land, water, and power grids.
Case Study: The Allbirds pivot to "NewBird AI"
The desperation for compute has reached such a pitch that it is reshaping entire industries. In March 2026, the sustainable shoe brand Allbirds—following a 99% stock wipeout—announced it would sell its brand and intellectual property to American Exchange Group for $39 million [2].
The remaining public company has officially pivoted to AI compute infrastructure under the name NewBird AI. When a lifestyle shoe brand re-emerges as a cloud provider, it is a clear signal that the AI infrastructure moat is the only asset that currently matters to the public markets.
SpaceX's Colossus: The terrestrial bridge to the stars
Despite the orbital rhetoric, even SpaceX is winning the compute war on the ground first. In May 2026, Anthropic signed a landmark deal to use the "full capacity" of SpaceX-xAI's Colossus 1 data center in Memphis [3].
Colossus 1 At-a-Glance:
- Scale: 220,000+ NVIDIA GPUs (H100, H200, and GB200).
- Power: 300+ Megawatts.
- Deployment: One of the fastest-deployed superclusters in history.
This move allowed Anthropic to double its session limits for Claude Code and remove peak-hour throttling, proving that owning the context requires immediate, physical hardware on Earth.
The rise of Sovereign AI and the "Neocloud"
As global compute becomes more constrained, we are seeing the end of "plug-and-play" AI access. Countries and companies are building Sovereign Intelligence to ensure they aren't cut off by foreign power shortages or policy shifts.
SoftBank is leading this in Japan with its "AI Data Center GPU Cloud" (scheduled for October 2026), utilizing NVIDIA GB200 NVL72 racks. This "neocloud" strategy focuses on secure, integrated infrastructure that can be utilized entirely within a single borders, bypassing the latency and geopolitical risks of orbital or cross-border compute [4].
What this means for your business
The "Infrastructure War" isn't just for billionaires; it dictates the price and availability of the tools you use every day.
- Hedge your compute: If you are building agentic systems, don't rely on a single provider's API. Use hybrid RAG architectures to reduce token dependency.
- Audit your "Sovereign Risk": If your primary model provider relies on a single geographic cluster (like Memphis or Hokkaido), consider how a local power grid failure or policy shift would affect your operations.
- Watch the "Last Verified" date: In 2026, a "stable" AI tool is only as stable as its underlying power contract.
FAQ
Q: Is space-based compute actually cheaper? A: Theoretically, yes. Solar power is constant and cooling is "free" in the vacuum of space. However, the costs of launch, maintenance, and communication latency (at least 25ms round-trip for Starlink) currently outweigh the savings on terrestrial electricity.
Q: What is the "Stargate" project? A: Stargate is a $500 billion AI infrastructure venture between SoftBank, OpenAI, Oracle, and MGX. It aims to build massive US-based campuses to provide the "rails" for the next generation of LLMs.
Q: Will orbital data centers ever happen? A: SpaceX has already filed with the FCC for a million-satellite megaconstellation. While skeptics like Sam Altman call it "ridiculous" for this decade, Starlink V3 is already being built with 1 Tbps capacity nodes. It is a long-term play for 2030 and beyond.
Discussion
0 comments