nb1t.sh

Anthropic × SpaceX and Why Claude Is Now Running on Musk’s Supercomputer

Thu Apr 09 2026 · Nitin Bansal

#ai#space#partnerships#anthropic

Table of Contents

What You Need to Know

Anthropic has struck a major compute deal with SpaceX, giving Claude access to SpaceX’s Colossus 1 data center in Memphis — a massive AI facility reportedly packing more than 220,000 NVIDIA GPUs and over 300 megawatts of compute capacity [2][3][4][5][6][7][8].

On paper, this is a compute deal. In reality, it is more interesting than that.

It gives Anthropic badly needed capacity for Claude, especially Claude Code, whose growth has reportedly been constrained by rate limits and infrastructure bottlenecks [1][3][6]. It also gives SpaceX a high-profile AI customer just as the company is reportedly preparing for an IPO [1][2][3][5][6].

But the real story is stranger: Anthropic is now relying on infrastructure controlled by Elon Musk, whose own AI company, xAI/Grok, directly competes with Claude. At the same time, Musk has gone from publicly calling Anthropic’s models “misanthropic” and “evil” to saying he was impressed by Anthropic’s team and comfortable leasing them Colossus 1 [1][6][9].

This is not just an AI infrastructure story. It is a story about compute scarcity, strange alliances, IPO narratives, frontier model competition, and the beginning of a much bigger pitch: orbital data centers.

What Actually Happened

On May 6, 2026, Anthropic and SpaceX announced a compute partnership that gives Anthropic access to SpaceX’s Colossus 1 data center in Memphis, Tennessee [2][3][4][5].

The facility is described as one of the largest AI supercomputing clusters in the world, with:

Item Reported Detail
Facility Colossus 1
Location Memphis, Tennessee
Capacity 300+ megawatts
GPUs 220,000+ NVIDIA GPUs
GPU types H100, H200, and GB200 accelerators
Primary use Claude inference and capacity expansion
Financial terms Not disclosed
Duration Not disclosed
Exclusivity Unclear

Anthropic said the additional capacity would support Claude Pro, Claude Max, Claude Code, and Claude API products [4][8]. Tom Brown from Anthropic also indicated that Claude inference would begin ramping on Colossus “in the next few days” after the announcement [9].

That last part matters. This does not sound like a vague future memorandum. It sounds like near-term capacity relief.

Why Anthropic Needed This

The simple explanation: Claude is growing faster than Anthropic’s infrastructure can comfortably handle.

Claude Code appears to be one of the biggest pressure points. Developers have reportedly complained about rate limits and service disruptions, and one source says developers spend an average of 20 hours per week using Claude Code [3]. Another source claims Anthropic’s services were running at around 99.1% uptime, far below the “five nines” standard often expected from mission-critical infrastructure [6].

After the SpaceX deal, Anthropic announced several usage-limit improvements, including:

  • doubled five-hour Claude Code rate limits for Pro, Max, Team, and seat-based Enterprise users;
  • removal of peak-hour limit reductions for Pro and Max accounts;
  • higher Claude Opus API limits, including a reported Tier 4 increase to 10 million input tokens per minute and 800,000 output tokens per minute [4].

That tells us something important: this capacity was not just for future model training. It was also about day-to-day product reliability.

For users, this means fewer rate-limit walls. For Anthropic, it means the company can keep pushing Claude Code harder without immediately breaking the user experience.

Why SpaceX Wanted This

SpaceX gets something extremely valuable here: validation.

If SpaceX is trying to pitch itself as more than a rocket company — as a future AI infrastructure company — then landing Anthropic as a customer is a strong signal. Anthropic is one of the most important AI labs in the world. If it is willing to run Claude workloads on SpaceX infrastructure, that helps SpaceX tell a much bigger story to investors [1][3][6].

The timing is also important. Several sources describe SpaceX as IPO-bound, with one source suggesting a possible IPO as soon as June 2026 [3]. A major AI infrastructure customer makes that story more credible.

There is also a practical reason: SpaceXAI had reportedly moved its own training workloads from Colossus 1 to Colossus 2, freeing Colossus 1 for external use [6][9]. If true, leasing Colossus 1 to Anthropic is a straightforward way to monetize capacity that was no longer central to SpaceXAI’s own training needs.

In other words: Anthropic gets compute. SpaceX gets revenue, validation, and an IPO-friendly AI infrastructure narrative.

The Weird Part: xAI, Grok, and SpaceXAI

The strangest part of the deal is the competitive structure.

Around the same time, Musk announced that xAI would be dissolved as a separate company and folded into SpaceX under the SpaceXAI brand [5][6][8][9].

That means the same corporate universe now contains:

  • SpaceX, the infrastructure provider;
  • SpaceXAI/xAI, the AI division;
  • Grok, a Claude competitor;
  • Anthropic, now a major customer of that infrastructure.

This creates a very unusual “frenemy” setup. Anthropic is leasing compute from a company connected to a direct competitor. The sources do not explain how data isolation, competitive firewalls, security protections, or IP safeguards are structured.

That does not mean the deal is unsafe or badly designed. It means the most important contractual details are invisible from the outside.

And those details matter.

Musk’s Sudden Reversal on Anthropic

One of the most entertaining parts of the story is Musk’s shift in tone.

As recently as February 2026, Musk had publicly criticized Anthropic’s AI as biased, “misanthropic,” and “evil” [1][3][6][8][9].

Then, after meeting senior members of Anthropic’s team, Musk posted that he was impressed by them, that they cared about doing the right thing, and that “no one set off my evil detector” [9].

He also directly connected that meeting to the compute lease, saying he was okay leasing Colossus 1 to Anthropic after spending time with the team [9].

This is funny, but it is also revealing.

A decision involving one of the world’s largest AI supercomputers was publicly framed around Musk’s personal trust assessment of Anthropic’s leadership. Maybe that is just Musk being Musk. But it also raises a broader question: who gets access to frontier-scale compute, and on what basis?

Was this a serious alignment-driven change of heart? A pragmatic business move? An IPO-friendly partnership? Probably some combination of all three.

The Bigger Bet: Data Centers in Space

The most speculative part of the announcement is the mention of orbital data centers.

Multiple sources say Anthropic and SpaceX have expressed interest in developing multiple gigawatts of space-based AI compute capacity [1][2][3][4][5][6][7][8].

The pitch is obvious: AI compute demand is growing so fast that terrestrial power, land, and cooling may become bottlenecks. Space offers abundant solar energy, and SpaceX has the launch capacity and orbital operations experience to at least make the idea sound plausible.

But right now, this still looks more like a strategic vision than a concrete engineering plan.

The available sources do not disclose:

  • a timeline;
  • cost estimates;
  • binding commitments;
  • technical architecture;
  • launch economics;
  • maintenance plans;
  • latency assumptions;
  • thermal-management strategy;
  • radiation-hardening requirements;
  • regulatory details.

So the ground-based Colossus 1 deal is real and immediate. The orbital data center angle is exciting, but still very much in the “big narrative” category.

The Red Flags

There are a few important caveats.

First, the financial terms are completely unknown. No source discloses what Anthropic is paying, how long the agreement lasts, whether it involves cash, equity, revenue sharing, or some other structure.

Second, exclusivity is unclear. Some sources use language suggesting Anthropic gets all or full access to Colossus 1, while others simply say Anthropic gets access to the cluster [1][2][6][7]. That distinction matters a lot.

Third, Colossus 1 has environmental baggage. CNBC reported that xAI had used natural gas-burning turbines at the Memphis facility, drawing local pollution concerns and protests [5]. Anthropic’s own announcement emphasizes broader commitments around energy and responsible infrastructure, but the tension between those commitments and the Colossus controversy remains unresolved [4][5].

Fourth, Anthropic reportedly faced Pentagon blacklisting in March 2026 as a supply chain risk and sued the Trump administration over it [5][8]. That becomes more interesting because SpaceX is a major defense contractor, and xAI/Grok has reportedly been embraced by the Defense Department as an alternative [8].

None of this necessarily breaks the deal. But it makes the partnership much more complicated than a simple “Claude gets more GPUs” headline.

What This Means

The Anthropic–SpaceX deal says a lot about where the AI industry is heading.

First, compute scarcity is still one of the biggest constraints in AI. Even a company with massive partnerships across Amazon, Google, Microsoft, NVIDIA, Fluidstack, and others still needed a near-term injection of capacity from SpaceX [4][6].

Second, AI infrastructure is becoming more fragmented. The future will not be only AWS, Azure, and Google Cloud. It may include specialized AI clusters, private supercomputers, sovereign compute deals, and eventually maybe even orbital infrastructure.

Third, model companies may increasingly depend on their competitors’ infrastructure. This is already normal in cloud, but frontier AI makes it more sensitive because model weights, inference traffic, user data, and product reliability are all strategically important.

Fourth, SpaceX is clearly trying to move into the AI infrastructure story. Whether orbital data centers become real or not, the Anthropic deal gives SpaceX a credible opening chapter.

Finally, Claude Code seems important enough that Anthropic is willing to make unusual infrastructure moves to support it. That may be the most grounded takeaway of all: AI coding demand is now large enough to reshape compute strategy.

Open Questions

The most important unanswered questions:

  1. How much is Anthropic paying SpaceX?
    No source discloses the financial terms.

  2. Is Anthropic getting all of Colossus 1 or just priority access?
    The wording varies across sources.

  3. How long does the agreement last?
    No duration has been disclosed.

  4. How is Anthropic’s data isolated from SpaceXAI/xAI/Grok?
    This is the biggest competitive-safeguard question.

  5. What exactly is Colossus 2?
    Sources say SpaceXAI moved training there, but specifications are not provided.

  6. Is orbital compute a real engineering roadmap or an IPO narrative?
    The ambition is clear. The execution details are not.

  7. How does the Pentagon blacklisting issue affect this partnership?
    Especially given SpaceX’s defense ties and xAI/Grok’s government positioning.

  8. Will Memphis environmental concerns become Anthropic’s problem too?
    Anthropic may not own the facility, but it now benefits from it.

  9. Does this weaken Grok’s frontier-model ambitions?
    Maybe — but only if Colossus 1 was still strategically important to Grok. If Grok fully moved to Colossus 2, the impact may be minimal.

  10. Is this a one-off compute lease or the beginning of a deeper Anthropic–SpaceX relationship?
    The orbital data center language suggests both companies want to keep that door open.

References

  1. Anthropic unveils 'dreaming' feature to help its AI agents self-improve - https://money.usnews.com/investing/news/articles/2026-05-06/anthropic-unveils-dreaming-feature-to-help-its-ai-agents-self-improve

  2. Elon Musk's SpaceX will give Anthropic access to its massive Colossus 1 AI data center - https://nbcnews.com/tech/tech-news/anthropic-spacex-partnership-compute-power-rcna343902

  3. Anthropic and Elon Musk's SpaceX said on Wednesday that the two entities have signed an agreement - https://wired.com/story/anthropic-spacex-compute-deal-colossus

  4. Anthropic: Higher limits with SpaceX compute partnership - https://anthropic.com/news/higher-limits-spacex

  5. Anthropic strikes deal with Elon Musk's SpaceX for data center capacity, ending years of criticism - https://cnbc.com/2026/05/06/anthropic-spacex-data-center-capacity.html

  6. Anthropic Strikes SpaceX Deal to Fuel Claude Code Growth and for Data Centers in Space - https://morningstar.com/news/marketwatch/20260506224/anthropic-strikes-spacex-deal-to-fuel-claude-code-growth-and-for-data-centers-in-space

  7. Anthropic Compute Partnership - https://x.ai/news/anthropic-compute-partnership

  8. Anthropic, SpaceX announce compute deal that includes space development - https://reddit.com/r/space/comments/1t5joco/anthropic_spacex_announce_compute_deal_that

  9. Elon Musk on X: Colossus 1 lease to Anthropic - https://x.com/elonmusk/status/2052069691372478511

Related