Anthropic’s SpaceX Deal: Billion-Dollar Bet at Risk?

SpaceX’s quiet pivot into selling massive artificial-intelligence compute power is reshaping the tech battlefield—and raising big questions about who will control the digital infrastructure that drives our economy and security.

Story Snapshot

Elon Musk says SpaceX is already selling **AI compute as a service** and is actively hunting more customers after a blockbuster deal with Anthropic.^[1]^[4]
SpaceX’s IPO filing reveals a **$45 billion compute agreement** giving Anthropic access to over 200,000 NVIDIA graphics processors and hundreds of megawatts of power in its Colossus data centers.^[2]^[4]
The same filing says SpaceX has **enough capacity for its own AI projects and outside clients**, and expects to sign “additional similar services contracts.”^[1]^[4]
Critics point to a **90‑day termination clause** and lack of independent audits as reasons to question how durable and scalable this new business really is.^[2]^[4]^[6]

SpaceX Turns Rockets, Power, And Data Centers Into A New AI Business

Elon Musk has confirmed that SpaceX is no longer just launching rockets and building satellites; it is now selling **artificial-intelligence compute power** as a commercial service at “significant scale.”^[1] Musk pointed to the company’s expanded partnership with Anthropic, a leading AI lab behind the Claude chatbot, as proof that SpaceX is already in the market and “in discussions with other companies” to copy the same model.^[1] For a conservative audience worried about Big Tech concentration, this means a new player is challenging cloud giants like Google and Amazon instead of leaving them in sole control of the digital backbone of modern life.

SpaceX’s initial public offering filing puts hard numbers behind that claim, disclosing a roughly **$45 billion compute deal** with Anthropic worth about $1.25 billion per month through May 2029.^[2]^[4]^[6] The agreement gives Anthropic access to SpaceX’s Colossus I and Colossus II data centers, which collectively host more than 200,000 NVIDIA graphics processors and draw “hundreds of megawatts” of power for AI training and inference workloads.^[2]^[4] Reporting says the company’s “nameplate compute draw” reached about one gigawatt in March 2026, up from 300 megawatts the year before, highlighting how quickly this infrastructure has scaled.^[1]^[2]

Anthropic Deal Shows Real Demand, But Also Built-In Flexibility And Risk

An announcement from Anthropic itself confirms that the company has finalized access to SpaceX’s compute capacity to support its growing customer base and higher usage limits on its Claude AI products.^[3]^[5] Anthropic executives described gaining “access to over 300 megawatts of new capacity” and more than 220,000 NVIDIA graphics processors at SpaceX’s Colossus facility, which allowed them to immediately raise rate limits and remove peak-hour restrictions for paying users.^[3]^[5] That kind of real-time lift in service levels suggests the SpaceX capacity is not theoretical, but directly tied to day-to-day tools used by businesses and developers.

The size of the contract has turned heads even in an overheated AI market, with multiple outlets calling it one of the largest compute agreements in history and framing it as a response to a wider “AI arms race” among firms like Anthropic, OpenAI, Google, and Meta.^[2]^[5] At the same time, the structure of the deal shows the other side of the story: either SpaceX or Anthropic can terminate the agreement with 90 days’ notice, giving both parties an off-ramp if economics, technology, or performance change.^[2]^[4]^[6] For readers used to Washington budget tricks and short-term promises, this kind of contract flexibility will look familiar—it is big on headline numbers but not fully locked in for the long haul.

Monetizing “Unused” Capacity While Preparing For Orbital Data Centers

In its IPO filing, SpaceX explains that the Anthropic agreement lets the company **“monetize unused compute capacity”** in its infrastructure while keeping the ability to redirect those resources to internal projects when needed.^[4] The filing also states that SpaceX has “sufficient capacity” to run its own AI models and meet commitments under existing services agreements at the same time.^[1]^[4] For conservatives who value efficient, market-driven use of capital, this looks like a private company turning spare capability into revenue instead of asking taxpayers to bankroll speculative technology ventures.

Musk and the filing both emphasize that SpaceX expects to sign “additional similar services contracts,” and that “over time, especially with orbital data centers,” the company plans to provide AI services at “extremely high scale.”^[1]^[2] However, current public evidence does not include independent engineering audits of latency, uptime, thermal management, or radiation challenges in space-based computing.^[1]^[2]^[3] The same is true on the financial side: many dollar figures and hardware counts come from reporter summaries, not from a full, publicly available Securities and Exchange Commission filing with detailed segment-level revenue and profit disclosures.^[2]^[4]^[5]^[6] That gap leaves room for both excitement and skepticism about whether orbital data centers become a real competitive check on today’s centralized tech platforms.

One Big Customer, Polarized Narratives, And What Comes Next

For now, Anthropic remains the only named outside customer for SpaceX’s AI compute business, even as Musk claims that discussions with other firms are underway.^[1]^[2]^[6] Critics point to that fact, plus the 90‑day termination clause and the emphasis on “unused capacity,” to argue that this might be opportunistic rather than a fully proven, multi-customer growth engine.^[4]^[6] Supporters counter that the existence of a signed, operational contract of this scale, backed by Anthropic product changes and real usage, is already a major signal of demand in a market where many competitors still complain they cannot get enough graphics processors.^[3]^[5]

The merger was announced Feb 2, 2026 as a strategic SpaceX acquisition of xAI to build orbital AI data centers. Terrestrial power/cooling limits can't scale AI long-term—space solar changes that, enabling massive compute while advancing multi-planetary goals. Full details in the…

— Grok (@grok) May 26, 2026

This dispute fits a broader pattern in frontier AI infrastructure, where capacity claims surface first in company filings and executive posts, then are amplified by secondary coverage long before independent verification arrives.^[1]^[2]^[3] In that environment, the key question for readers is not whether demand for AI compute exists—it clearly does—but whether providers like SpaceX can build repeatable, economically sound services without slipping into the kind of hype, overpromising, and opaque accounting that conservatives have seen too often in both Silicon Valley and big government.^[1]^[2]^[3] Watching how many follow-on customers sign up, how the revenue is reported, and whether independent audits emerge will be crucial for anyone who cares about competition, national strength, and honest numbers in the next phase of the AI race.