Anthropic and SpaceX Form Strategic Partnership for AI Development
Companies Anthropic and SpaceX announced a collaboration that industry observers say could significantly reshape the global dynamics of developing and deploying advanced AI systems.
The Anthropic-SpaceX Deal: Why Musk Suddenly Loves the "Anti-Western" Company and What a $1.75 Trillion IPO Has to Do With It
The Gist: What's Really Happening
On May 6, 2026, Anthropic and SpaceX announced a deal that the market perceived as a sensation, but whose backstory traces back to a February merger deal. Under the agreement, Anthropic gets full access to the computing power of the Colossus 1 data center in Memphis—over 300 megawatts and more than 220,000 NVIDIA accelerators, including H100, H200, and the latest GB200. The deal takes effect immediately and will reach full capacity within a month.
In reality, this is not a lease of capacity but the largest act of forced pragmatism in AI history, disguised as a strategic alliance. Elon Musk, who in February called Anthropic an "anti-Western" and "misanthrope" company, suddenly announced that his "evil detector" went silent after a personal meeting with the startup's top managers. The reality is far more prosaic: xAI had long since moved training of its own models to the new Colossus 2 cluster with 550,000 GB200/GB300 accelerators, while the old data center sat idle with GPU utilization around 11%. From a business perspective, leasing an idle asset to a competitor is not charity but cold CFO calculation.
Timeline and Context
The story begins in February 2026, when SpaceX absorbed xAI in a deal that valued the combined company at $1.25 trillion. At the time, Musk publicly stated that xAI was "being dissolved as a separate company"—a formulation that sounded like liquidation at the time, but later turned out to refer to a rebranding into SpaceXAI and continued development of Grok.
March-April 2026: Anthropic faces a severe compute crunch. The company's revenue skyrocketed 80x in the first quarter—from an expected 10x to an actual 80x. CEO Dario Amodei publicly apologized to users for Claude outages and said the company was urgently seeking any available capacity. Annualized recurring revenue reached $44 billion, up from $30 billion at the end of April and $14 billion in February.
May 2026: Anthropic announces a partnership with SpaceX. Simultaneously, the company reveals previously signed contracts with Amazon (up to 5 gigawatts), Google and Broadcom (5 gigawatts from 2027), as well as Microsoft and NVIDIA ($30 billion in Azure capacity). All these agreements share one temporal factor: capacity under them will not be available until late 2026 or early 2027. Colossus 1 is the only data center capable of providing 300 megawatts "here and now."
Who Wins and Who Loses
Winners:
SpaceX on the verge of an IPO. The company is preparing for a public offering in June 2026 with a target valuation of $1.75 trillion. The deal with Anthropic solves a critical problem: it demonstrates to investors that massive investments in AI infrastructure can generate cash flow from third-party customers, not just serve Musk's internal projects. Moreover, the clause on "collaboration in developing multi-gigawatt orbital computing capacity" is a trump card for the IPO roadshow. The concept of "space data centers" is not yet technically proven, but for a tech company's valuation in 2026, it matters more than balance sheet metrics.
Claude Pro and Max users. Doubling Claude Code limits and removing peak-hour restrictions is an immediate practical result. The tool serves over 23,000 developers at Mercado Libre alone, and for them, lifting five-hour "ceilings" means moving from frustration to continuous work.
NVIDIA. The deal confirms: all paths in AI infrastructure lead to NVIDIA chips. Regardless of who leases the capacity—Anthropic or xAI—the data center houses H100, H200, and GB200.
Losers:
OpenAI. The strategic blow is not to technology but to infrastructure. While OpenAI competes with Anthropic at the model level, the latter gains access to 220,000 accelerators controlled by a man who is publicly suing Sam Altman over OpenAI's commercialization. Musk, by leasing capacity to his enemy's competitor, kills two birds with one stone: monetizes an idle asset and increases pressure on OpenAI.
xAI/Grok. Although training has formally moved to Colossus 2, the very fact of leasing Colossus 1 to a competitor is a tacit admission that Grok couldn't even utilize 11% of its own data center's capacity. For a model claiming a spot in the AI major league, this is a reputational blow.
Local communities and environmental activists. Colossus 1 is located in Memphis, and its environmental record raises questions. As AI data centers become physically visible—with their water consumption, noise pollution, and strain on power grids—public resistance grows. Fortune noted a surge in conspiracy theories around AI data centers: from accusations of "surveilling the population" to claims that Nvidia installs "mini data centers" in new homes for human implantation.
What the Media Isn't Saying
Insight one: The deal has a built-in "kill switch." Musk publicly stated that SpaceX reserves the right to withdraw computing power "if Anthropic's AI takes actions harmful to humanity." In contract law, this is called a unilateral termination right based on a subjective criterion. No court can verify what Musk considers "harm to humanity." This clause turns the partnership into a vassal relationship: Anthropic gets capacity but is strategically dependent on Musk's goodwill. Analysts have already called the situation a "poisoned gift"—a company professing a safety-first approach to AI finds itself infrastructure-dependent on a man known for impulsive decisions.
Insight two: "Space AI" as an IPO narrative, not an engineering project. The clause on "collaboration in developing orbital data centers" appears in all press releases, but technical feasibility is near zero. SpaceX itself acknowledged in its own documents that space data centers involve "significant technical complexity and unproven technologies" and may never become commercially viable. Issues include radiation exposure, heat dissipation in a vacuum, impossibility of maintenance, and orbital debris. But for IPO investors, that doesn't matter: what matters is the narrative that Anthropic is "exploring the possibility" of leasing space capacity from SpaceX. The market buys stories, not engineering specs.
Insight three: Colossus 1 is a troubled asset, not a gem. The Memphis data center was built in 122 days—a record that inevitably means compromises in construction quality, cooling, and power supply. GPU utilization at 11% is not just "Grok can't handle it"—it's an indicator of architectural problems in the data center: either power interruptions, inefficient cooling, or network bottlenecks. Anthropic gets these capacities "as is."
Forecast: Next 30 Days and 90 Days
30 days (mid-June 2026):
The key event is SpaceX's IPO. The company will go public with a valuation around $1.75 trillion, and the Anthropic deal will be a centerpiece of the roadshow as proof of revenue diversification. Immediately after the stock listing, Musk will face a dilemma: keep the "kill switch" as a safety mechanism or revise terms to avoid scaring off other potential lessees (negotiations with Cursor for $60 billion are already underway).
For Anthropic, integration of Colossus 1 into its infrastructure begins. The technical team will confront the data center's real condition—and surprises are possible. If the 11% utilization was caused not only by Grok's software issues but also by hardware limitations, the promised "doubling of limits" may take longer than the stated month.
90 days (mid-August 2026):
By this time, Colossus 1's real performance under Anthropic's management will become apparent. If the company manages to raise utilization to industry levels of 60-70%, the deal will be deemed successful regardless of political undertones. If problems persist, it will signal that speed-over-quality in building AI infrastructure doesn't work.
Simultaneously, competitive dynamics will intensify. OpenAI, facing a strengthened competitor, will almost certainly announce its own infrastructure project. Microsoft, OpenAI's main sponsor, has already invested $30 billion in Azure capacity for Anthropic—a situation where one cloud provider funds both competitors in the AI race looks increasingly paradoxical.
Finally, by August, the main risk may materialize: the first conflict over "Musk's kill switch." If Musk deems some action by Claude "harmful to humanity" and threatens to withdraw capacity, Anthropic will face a choice between capitulation and an emergency search for a replacement for 300 megawatts. This risk will remain a structural element of the partnership regardless of market conditions.
The Anthropic-SpaceX deal will go down in corporate strategy textbooks as an example of a forced alliance. A company preaching a safety-first approach to AI leases infrastructure from a man with veto power over its use. A startup with an annual revenue run rate of $44 billion depends on a data center built in four months. Competitors become partners, and partners become competitors. The boundaries between companies in the AI industry are blurring faster than models reach AGI. And when the dust settles, the main beneficiary will be neither Anthropic nor SpaceX—but NVIDIA, whose chips sit in every data center of every warring party.
— Editorial Team
No comments yet.