In May 2026, Anthropic signed a historic agreement with SpaceX to access the 220,000 Nvidia GPUs of Elon Musk's Colossus 1 supercomputer, located in Memphis. The immediate result: Claude Code usage limits have been doubled, peak-hour restrictions removed, and API volumes considerably increased.
This rapprochement between Anthropic and Musk's empire marks a major turning point in the race for AI computing power.
The SpaceX-Anthropic deal: 220,000 GPUs for Claude Code
The announcement was made on May 6, 2026, on stage at the Code with Claude developer conference in San Francisco. Anthropic, the creator of Claude, officially announced a partnership with SpaceX to leverage the entire computing power of the Colossus 1 data center, installed in Memphis, Tennessee.
Key figures of the partnership
The scale of the agreement is unprecedented in the industry for a bilateral contract between an infrastructure provider and an AI lab:
| Feature | Details |
|---|---|
| Number of GPUs | Over 220,000 Nvidia GPUs |
| Chip types | H100, H200 and GB200 (new generation) |
| Energy capacity | Approximately 300 megawatts |
| Consumption equivalent | 300,000 American households |
| Location | Memphis, Tennessee (former Electrolux site) |
| Construction time | 122 days |
| Activation timeline | Less than a month after the announcement |
The Colossus 1 supercomputer now belongs to SpaceXAI, the new entity born from the merger between SpaceX and xAI, the AI startup founded by Musk. This strategic regrouping positions SpaceX no longer just as a space player, but as a large-scale AI computing infrastructure provider.
Why Anthropic needed this deal now
On Anthropic's side, the situation had become critical.
The company acknowledged as early as April 2026 that the explosive demand for Claude Code and its agentic tools had generated what it described as "inevitable tension" on its servers. Developers were experiencing frustrating rate limits, slowdowns during peak hours, and increasingly frequent service interruptions.
According to Anthropic, the average developer now spends at least 20 hours per week using Claude Code. This massive adoption has put infrastructure under unprecedented pressure.
Dario Amodei, Anthropic's CEO, publicly admitted on stage: "We have had difficulties with compute. We are sorry if sometimes it takes time, but we will continue to acquire as much capacity as possible."
Executives at the company had reportedly even considered removing Claude Code from the $20 per month Pro offering, due to insufficient capacity to meet demand.
Concrete improvements for Claude Code users
The effect of the SpaceX agreement was felt immediately. On the very day of the announcement, Anthropic deployed a series of improvements that directly address the frustrations expressed by developers for weeks.
Doubled usage limits on all plans
Claude Code rate limits on the five-hour windows have been multiplied by two across all paid plans:
- Claude Pro: doubled limits on the 5-hour window
- Claude Max: identical doubling with priority access
- Claude Team: same increases for teams
- Claude Enterprise: substantial increase in dedicated quotas
Removal of peak-hour restrictions
Before the agreement, Pro and Max subscribers experienced automatic reductions in their limits during peak hours. This mechanism, particularly frustrating for developers working during the day in the United States, has been completely eliminated. You can now use Claude Code at full capacity regardless of the time.
Massive increase in API limits for Claude Opus
Developers who use the Claude API directly in their applications also benefit from a substantial increase in authorized request volumes, particularly on the Claude Opus models. This increase concerns all API access tiers and represents a significant jump compared to previous quotas.
These improvements played a direct strategic role: some developers had begun migrating to OpenAI Codex, which benefited from more generous limits thanks to compute agreements signed earlier by OpenAI. The SpaceX deal allows Anthropic to reclaim this ground.
Elon Musk's spectacular reversal
The most surprising aspect of this partnership is not technical, but human. Elon Musk and Anthropic had maintained an openly hostile relationship until very recently. Understanding this reversal sheds light on the current power dynamics in the AI industry.
Musk's public attacks on Anthropic
In February 2026, just three months before the agreement was signed, Musk was posting on X (his own social network) direct attacks against Anthropic.
He called the company "misanthropic" (a play on words with the company's name) and its models "evil". He accused them, without providing evidence, of racial and gender bias in Claude's responses.
This hostility was part of a broader context: Musk owned his own AI company, xAI, and his Grok chatbot was in direct competition with Claude. Attacking a competitor on social media is a tactic Musk has used many times in other sectors.
The turnaround in just a few days
The week before the deal announcement, Musk spent several days with Anthropic's leadership team. The result was a radical change in tone. On X, he posted a message in which he said he was "impressed" by the competence and ethical commitment of Anthropic's teams, adding that everyone he met "deeply cared" about the good of humanity.
This reversal is characteristic of Musk's style, who frequently oscillates between extreme positions. But beyond the billionaire's personality, two strategic factors explain this change:
- SpaceX's IPO: scheduled for the month following the deal, this IPO is estimated between $1.75 trillion and $2 trillion. Showcasing Anthropic as a reference customer of Colossus 1 considerably strengthens the case presented to investors, by demonstrating that SpaceXAI's infrastructure generates concrete revenue.
- The monetization of Colossus 1: the Memphis data center represents a colossal investment. Having a customer the size of Anthropic, whose revenues are growing at an annual rate of 80x, guarantees a recurring and predictable revenue stream.
The compute race: Anthropic catches up with OpenAI
This SpaceX deal is just one piece of a much larger puzzle. Anthropic has significantly accelerated its acquisition strategy for computing power, adopting an approach that its own CEO had nevertheless criticized at OpenAI a year earlier.
The irony of Amodei's strategy
In 2025, Dario Amodei publicly questioned Sam Altman's eagerness to sign massive compute agreements. He suggested that OpenAI was prioritizing quantity over quality and taking risky financial commitments. A year later, Anthropic is reproducing exactly the same pattern, driven by demand that has exceeded all forecasts.
This reversal illustrates a reality that has become unavoidable: in the AI industry in 2026, computing capacity is the main limiting factor. Having the best model is useless if you can't run it at scale.
Anthropic's portfolio of agreements
The SpaceX agreement adds to a series of commitments totaling tens of billions of dollars:
| Partner | Capacity / Amount | Timeline |
|---|---|---|
| Amazon (AWS) | Up to 5 GW (including ~1 GW operational) | End of 2026 |
| Google + Broadcom | 5 GW + TPU | 2027 horizon |
| Microsoft + Nvidia (Azure) | $30 billion | Multi-year |
| Fluidstack | $50 billion | US AI infrastructure |
| SpaceX (Colossus 1) | 220,000 GPUs / 300 MW | Immediate |
| Google Cloud (TPU) | $200 billion | Multi-year |
| Amazon (total) | $100 billion over 10 years | 2036 |
The scale of these commitments reflects Anthropic's growth trajectory. If the current 80x annual rate is maintained for even one more year, the startup would become one of the highest-revenue companies in the world.
The entire ecosystem under tension
The phenomenon does not only affect Anthropic. At the Code with Claude conference, the tension on compute was palpable throughout the entire ecosystem:
- A startup CEO confided having personally called a Google executive to beg for more Gemini tokens
- An executive at Cursor, the popular AI code editor, was seen crossing his fingers while waiting for the activation of his own compute agreement with SpaceX
- Tech companies of all sizes expressed the same urgent need for additional AI capacity
An important technical point: capacity transfer between customers is relatively simple in modern data centers, because most use the same Nvidia GPUs. Anthropic confirmed that access to SpaceX compute would be operational in less than a month.
The impact on developers: what changes concretely
Beyond the strategic announcements, this deal has very concrete consequences for developers who use Claude Code on a daily basis.
Before and after the SpaceX deal
Here is a comparison of the developer experience before and after the agreement:
| Aspect | Before the deal | After the deal |
|---|---|---|
| Request limits (5h) | Standard quotas | Doubled quotas |
| Peak hours | Automatically reduced limits | No restrictions |
| Claude Opus API | Restrictive rate limits | Massive volume increase |
| Service reliability | Frequent interruptions | Improved stability |
| Latency | Degraded during peak periods | More consistent |
Competition with OpenAI Codex
The timing of the agreement is not insignificant. OpenAI had signed its own compute agreements earlier, which allowed it to offer more generous limits on Codex, its competing AI coding service. The result: some developers had begun migrating from Claude Code to Codex, not out of preference for the model, but simply because they could use it more.
By doubling its limits and removing peak-hour restrictions, Anthropic sends a clear signal: Claude Code is back in the race, with the capacity to meet demand. For developers who had migrated, the question now arises of returning.
The environmental and social stakes of the Memphis data center
The SpaceX-Anthropic deal also raises important questions that go beyond the technological framework. The Colossus 1 data center does not exist in a vacuum: it is installed in a real community, with tangible consequences.
The energy bill and emissions
With 300 megawatts of consumption, Colossus 1 represents the equivalent of the electricity consumption of 300,000 households. The infrastructure is powered by gas turbines, a fossil energy source that generates significant emissions of CO₂ and other pollutants.
Memphis residents have already expressed their dissatisfaction. Complaints have been filed regarding emissions, and a protest took place in front of a SpaceX investor event the month before the Anthropic deal.
These local tensions illustrate the real social cost of the compute race: while developers in San Francisco benefit from doubled limits, it is Memphis residents who breathe the corresponding emissions.
The orbital data center project
In a more futuristic vein, Anthropic has expressed its interest in a project for orbital data centers in partnership with SpaceX.
The idea consists of moving part of the computing power into space, where cooling is naturally provided by the vacuum of space and where solar energy is continuously available.
This project, still at the exploratory stage, shows how willing the industry is to push all limits to satisfy compute demand.
Anthropic's hypergrowth: what the conference reveals
Beyond the SpaceX deal, the Code with Claude conference offered a striking glimpse of the speed at which Anthropic is evolving. Several journalists present noted signs of almost uncontrollable growth.
An unprecedented pace of development
Stephen Council, AI reporter at Business Insider, summed up the situation as follows: major tech conferences are generally annual, with products carefully prepared.
At Anthropic, the one-year cycle seems almost obsolete. Employees who arrived a few weeks earlier were giving demonstrations; others admitted they could no longer keep up. Features go from research preview status to public beta in record time.
The revenue trajectory
Anthropic maintains a revenue growth rate of around 80x per year. To put this figure in perspective: if this trajectory is maintained for even one more year, Anthropic would become one of the highest-revenue companies on the planet. It is this potential that justifies the tens of billions of dollars in infrastructure investments, and that makes a partnership with SpaceX not only logical, but necessary.
One of Anthropic's senior executives privately admitted that the company had underestimated demand. When usage exploded much faster than expected in 2026, the teams had to improvise urgently to handle the load.
What this deal means for the future of AI
The SpaceX-Anthropic partnership is not just a technological anecdote. It reveals deep trends that will structure the AI industry for years to come.
The end of ideological boundaries
Anthropic was founded by former OpenAI members who wanted to create a safer and more ethical AI. Elon Musk, for his part, has publicly attacked almost every player in AI, including OpenAI (which he co-founded) and Anthropic. The fact that these two parties sign a massive commercial agreement demonstrates that, in the compute race, ideological considerations give way to economic realities.
As a Frandroid journalist summed up: "At this stage of the AI race, no one looks at who they sign with anymore, as long as there are megawatts at the end of it."
The consolidation of AI infrastructure
The AI infrastructure landscape is consolidating around a few major players capable of providing hundreds of megawatts of computing power. SpaceXAI joins AWS, Google Cloud, and Microsoft Azure in this very exclusive club. For AI startups like Anthropic, diversifying their compute sources across multiple providers has become a survival strategy, not a luxury.
Compute as a decisive competitive advantage
The episode of developers migrating from Claude Code to OpenAI Codex, motivated solely by usage limits, demonstrates a new fact: model quality is no longer enough. If a developer cannot use Claude Code during their working hours due to rate limits, they will switch to a competitor, even if the underlying model is considered inferior. Compute has become the decisive competitive advantage.
Frequently asked questions
Why did Elon Musk sign an agreement with Anthropic when he was criticizing the company?
The reversal can be explained by two main factors.
First, SpaceX is preparing an IPO estimated between $1.75 trillion and $2 trillion, and having Anthropic as a reference customer strengthens the IPO case.
Second, the Colossus 1 data center represents a massive investment that needs high-consumption customers to be profitable. Musk also stated that he had been "impressed" by Anthropic's safety work after spending several days with their teams.
What are the new Claude Code limits after the SpaceX agreement?
Claude Code usage limits on the 5-hour windows have been doubled for all paid plans (Pro, Max, Team, Enterprise). The restrictions applied during peak hours have been entirely removed for Pro and Max subscribers. API request volumes for Claude Opus models have also been considerably increased. These changes took effect on May 6, 2026.
Is Claude Code better than OpenAI Codex after this deal?
The SpaceX deal does not change the intrinsic quality of the Claude model: it improves availability and usage limits.
The comparison between Claude Code and OpenAI Codex depends on your specific needs.
What changes is that Claude Code is no longer penalized by usage restrictions more constraining than those of the competition. Developers who had migrated to Codex solely for quota reasons now have a reason to reconsider Claude Code.
What is SpaceXAI and what is the link with xAI?
SpaceXAI is the entity born from the dissolution of xAI as a standalone company and its merger with SpaceX. This restructuring, announced in early 2026, transforms SpaceX into a dual player: space and AI infrastructure. The Colossus 1 data center in Memphis, initially built to train Grok (xAI's chatbot), is now operated by SpaceXAI and rented to third-party customers like Anthropic.
Does the SpaceX-Anthropic agreement have an environmental impact?
Yes, the Colossus 1 data center consumes approximately 300 megawatts, which is equivalent to the consumption of 300,000 American households. It is powered by gas turbines, a fossil energy source. Memphis residents have expressed concerns about emissions and air quality. Protests have taken place to oppose the site's environmental impact. Anthropic has mentioned an interest in orbital data centers as a future solution, but in the short term, the carbon footprint of this additional compute is significant.