The artificial intelligence arms race just produced one of its most unlikely alliances. Anthropic, the safety-focused AI company behind the Claude family of models, announced on May 6, 2026, that it has struck a deal with SpaceX to gain exclusive access to the entire compute capacity at the Colossus 1 data center in Memphis, Tennessee. The agreement, first reported by CNBC, delivers more than 300 megawatts of new capacity backed by over 220,000 NVIDIA GPUs — and it includes a forward-looking provision for developing orbital AI computing infrastructure in partnership with SpaceX.

The partnership is remarkable not just for its scale but for the corporate dynamics behind it. Colossus 1 was originally constructed to power xAI, the artificial intelligence venture founded by Elon Musk, who merged the operation into SpaceX earlier this year at a combined valuation of roughly $1.25 trillion. That merger effectively rebranded the AI division as “SpaceXAI,” and the Colossus facility — once dedicated exclusively to training Musk’s Grok models — is now being handed over in its entirety to a company that Musk publicly attacked just months earlier.

From Public Feud to Partnership

The speed at which the relationship shifted is striking. In February 2026, Musk posted on X that Anthropic was a company that, in his view, held deeply negative attitudes toward Western institutions. He questioned whether any technology firm could be more inconsistent in its public positions. Those remarks came amid his broader legal battle with OpenAI, the organization from which several Anthropic founders originally departed, and reflected a pattern of sharp criticism directed at multiple AI competitors.

Yet by early May, the tone had changed completely. After spending approximately a week meeting with senior members of the Anthropic team, Musk stated publicly that the people he encountered were highly skilled professionals who genuinely cared about responsible development. He noted that nobody he met raised any concerns through what he described as his personal judgment process for evaluating character. The turnaround suggests that commercial incentives — and the sheer demand for GPU capacity — have begun to override ideological differences in the AI sector.

From a business standpoint, the logic is straightforward. SpaceX is widely expected to pursue an initial public offering in the near future, and demonstrating that its data center infrastructure can attract premium tenants like Anthropic strengthens the revenue diversification narrative that prospective investors will want to see. For Musk, the deal positions SpaceX as more than a launch provider — it establishes the company as a serious AI infrastructure utility.

What Colossus 1 Brings to the Table

The Colossus 1 facility in Memphis is no ordinary data center. Equipped with more than 220,000 NVIDIA GPUs spanning the H100, H200, and GB200 accelerator families, it was purpose-built for the kind of massively parallel workloads that define modern AI training. The facility can handle large language model training, multimodal system development, scientific simulations, and generative AI inference at scale.

With over 300 megawatts of power capacity, Colossus 1 ranks among the largest single-site AI computing installations in the world. For context, a typical enterprise data center operates in the 10-to-50 megawatt range, meaning this single facility delivers the equivalent of six to thirty conventional data centers. The GB200 accelerators, NVIDIA’s newest generation, are specifically optimized for the kind of trillion-parameter model training that companies like Anthropic are pursuing as they develop their next generation of Claude systems.

Reports indicate that the arrangement includes references to a forthcoming model internally referred to as “Mythos,” though neither Anthropic nor SpaceX has provided official confirmation of that project. What is clear is that Musk himself characterized the Colossus infrastructure as uniquely well-suited for handling the immense parallel processing demands that Anthropic’s future systems will require.

It is worth noting that the Colossus 1 facility is distinct from the larger Colossus 2 complex, which continues to house the training infrastructure for Grok, SpaceX’s own AI platform. The separation ensures that both organizations can scale independently without competing for the same physical resources.

Immediate Impact on Claude Users

The deal is already producing tangible benefits for Anthropic’s customer base. Coinciding with the announcement, Anthropic rolled out several significant capacity improvements across its product lineup.

Claude Code, the company’s developer-focused coding assistant, saw its five-hour rate limits doubled across all paid tiers — including Pro, Max, Team, and Enterprise plans. For developers who rely on Claude Code for sustained programming sessions, this effectively removes one of the most common friction points that had driven complaints on forums and social media.

Perhaps more significantly, Anthropic eliminated peak-hour usage restrictions for Claude Pro and Claude Max subscribers. Previously, users on these plans experienced reduced access during high-demand periods, a throttling mechanism that frustrated paying customers who expected consistent availability. With the Colossus 1 capacity now flowing into the system, those constraints have been lifted globally.

On the API side, Anthropic announced considerable increases to rate limits for Claude Opus models, its most capable tier. While the company did not publish exact multipliers for every endpoint, the message was clear: the additional 220,000 GPUs are being deployed to address the capacity bottleneck that has constrained Anthropic’s ability to serve its fastest-growing customer segments.

For investors tracking the best AI stocks to buy in 2026, these capacity expansions signal that Anthropic is positioning itself to compete aggressively for enterprise customers who demand reliable, high-throughput AI access without service degradation during peak periods.

The Space Computing Frontier

The most forward-looking element of the agreement is Anthropic’s expressed interest in partnering with SpaceX to develop multiple gigawatts of orbital AI compute capacity. While details remain sparse, the concept involves deploying compute satellites in low Earth orbit that would harness solar energy for power, thereby reducing strain on terrestrial electrical grids.

The idea is not as far-fetched as it might initially seem. Data centers are among the fastest-growing consumers of electricity worldwide, and the power demands of AI training are accelerating that trend dramatically. A space-based compute layer could theoretically tap virtually unlimited solar energy while avoiding the permitting challenges, land-use conflicts, and grid interconnection delays that slow terrestrial data center construction.

SpaceX, with its Starlink constellation already comprising thousands of satellites and its Starship vehicle designed for heavy payload deployment, is arguably the only company on Earth with the launch economics to make orbital data centers feasible within the next decade. If the partnership progresses beyond the exploratory stage, it could fundamentally reshape how the industry thinks about AI infrastructure scaling.

The environmental dimension adds another layer. The Colossus 1 facility in Memphis has drawn criticism from local civil rights organizations over air quality concerns, as the site relies on natural gas turbines for power generation. xAI had operated under a temporary-use designation that it claimed exempted it from federal permitting requirements, a position that drew public protests. Moving some compute capacity to orbit would sidestep these terrestrial environmental disputes entirely.

A Crowded Infrastructure Landscape

The SpaceX deal does not exist in isolation. Anthropic has been assembling one of the most diversified compute supply chains in the AI industry, reflecting a deliberate strategy to avoid the single-vendor dependency that has constrained some competitors.

Recent agreements include a partnership with Amazon for up to five gigawatts of capacity, a separate arrangement with Google and Broadcom targeting five gigawatts with a 2027 launch timeline, a $30 billion Azure capacity agreement with Microsoft and NVIDIA, and a $50 billion infrastructure investment commitment from Fluidstack. The SpaceX deal adds another 300-plus megawatts to this portfolio, with the space computing component potentially contributing gigawatts more in the longer term.

This multi-cloud, multi-provider approach mirrors the broader trend among leading AI companies to secure compute resources wherever they can find them. As valuations in the sector continue to climb — a dynamic explored in depth in our analysis of AI startup valuations and whether they represent a bubble — the ability to actually deliver on product promises depends increasingly on raw infrastructure access rather than algorithmic innovation alone.

Competitive Implications

The timing of the announcement carries competitive significance. On the same day, OpenAI revealed a new partnership with PwC, one of the Big Four accounting firms, to deploy autonomous AI assistants in professional services. The parallel announcements underscore the intensity of the race to move AI from experimental technology to embedded enterprise infrastructure.

Anthropic also unveiled ten specialized AI agents designed for financial services — targeting banks, insurers, and asset managers — suggesting that the Colossus 1 capacity will support not just general-purpose Claude improvements but also vertical-specific enterprise products that demand dedicated compute resources.

For the broader technology market, the deal creates an interesting counter-alliance to the Microsoft-OpenAI partnership that has dominated AI infrastructure headlines for the past two years. By linking Anthropic’s safety-first development philosophy with SpaceX’s unmatched launch and infrastructure capabilities, the agreement creates a new axis of competition that neither Microsoft nor Google can easily replicate.

As big tech earnings demonstrate how strategic AI spending gets rewarded by the market, the Anthropic-SpaceX deal represents exactly the kind of bold infrastructure bet that Wall Street has been rewarding. Companies willing to lock in GPU capacity now, even at premium prices, are positioning themselves to capture market share as enterprise AI adoption accelerates through the second half of 2026.

What Comes Next

Several questions remain unanswered. The financial terms of the deal have not been disclosed, leaving analysts to speculate about the revenue implications for SpaceX’s anticipated IPO filing. The timeline for orbital compute development is similarly unclear, though SpaceX’s existing Starship production cadence suggests that hardware deployment could begin within two to three years if the partnership moves forward.

For Anthropic, the deal addresses the most immediate constraint on its growth — raw compute availability — while opening a potentially transformative long-term infrastructure channel. For SpaceX, it validates the company’s pivot toward becoming an AI infrastructure provider and not merely a transportation company that happens to own some GPUs.

The partnership between two organizations led by philosophical opposites — Anthropic’s constitutional AI framework emphasizing safety guardrails versus Musk’s historically aggressive stance on rapid AI deployment — may prove to be either a pragmatic masterstroke or an unstable alliance. What is certain is that the deal reshapes the competitive landscape of AI infrastructure in ways that will ripple through the industry for years to come.

How many GPUs does the Colossus 1 data center contain?

The Colossus 1 facility in Memphis, Tennessee, houses more than 220,000 NVIDIA GPUs. These span multiple accelerator generations including the H100, H200, and the newest GB200 chips. The facility delivers over 300 megawatts of compute capacity, making it one of the largest single-site AI computing installations globally.

Will this deal affect Claude subscription pricing?

Anthropic has not announced any pricing changes as part of the SpaceX deal. However, the company has already delivered tangible benefits to existing subscribers, including doubled rate limits for Claude Code, elimination of peak-hour throttling for Pro and Max plans, and increased API limits for Claude Opus models. These improvements suggest the additional capacity is being used to enhance service quality rather than to justify price increases.

What is the orbital AI computing component of the deal?

Anthropic has expressed interest in partnering with SpaceX to develop multiple gigawatts of compute capacity in space. The concept involves deploying compute satellites in low Earth orbit powered by solar energy, which would reduce dependence on terrestrial power grids. While still in the exploratory phase, SpaceX’s existing Starlink infrastructure and Starship launch vehicle make it uniquely positioned to pursue this goal.

Why did Elon Musk agree to work with Anthropic after criticizing the company?

Despite publicly criticizing Anthropic as recently as February 2026, Musk changed his position after spending approximately a week meeting with senior Anthropic staff. He stated that everyone he met demonstrated high competence and genuine concern for responsible development. The commercial logic is also compelling — the deal generates significant revenue for SpaceX’s data center division ahead of an expected IPO.

How does this deal compare to Anthropic's other compute agreements?

The SpaceX deal adds over 300 megawatts of immediate capacity, which complements Anthropic’s broader infrastructure portfolio. The company has also secured up to five gigawatts through Amazon, five gigawatts through a Google-Broadcom partnership launching in 2027, $30 billion in Azure capacity through Microsoft and NVIDIA, and a $50 billion commitment from Fluidstack. The SpaceX arrangement is unique in offering exclusive access to an entire facility plus a pathway to orbital computing.

What is the difference between Colossus 1 and Colossus 2?

Colossus 1 is the facility that Anthropic will now use exclusively under this deal. Colossus 2 is a separate, larger installation that continues to serve as the primary training infrastructure for Grok, SpaceX’s own AI platform. The separation ensures that both Anthropic and SpaceX can scale their respective AI operations independently without resource conflicts.