Musk Dissolves xAI, Leases Data Center to Anthropic

Elon Musk has announced that xAI, his artificial intelligence venture, will be dissolved as an independent entity and absorbed into SpaceX under a new banner called SpaceXAI. In what may be the most ironic twist in the AI industry this year, SpaceX will lease xAI's crown jewel — the massive Colossus 1 data center — to Anthropic, the very company Musk publicly branded as 'misanthropic and evil' just 3 months ago.

The announcement, made on May 6, comes exactly 33 days before SpaceX's highly anticipated June 8 IPO roadshow, a timing that analysts say is anything but coincidental.

Key Takeaways

• xAI is no more — the company has been merged into SpaceX and rebranded as SpaceXAI
• Colossus 1, xAI's flagship data center with roughly 300 megawatts of compute capacity, will be leased to Anthropic
• The move comes just 33 days before SpaceX's planned IPO roadshow on June 8
• Musk called Anthropic 'evil' and 'misanthropic' as recently as 3 months ago
• xAI had been systematically dismantled over the prior 3 months, making this the final step
• SpaceXAI will no longer pursue next-generation frontier AI models, pivoting instead to a compute-as-infrastructure role

From 'Evil' to Business Partner: Musk's Stunning Reversal on Anthropic

The sheer audacity of this deal cannot be overstated. In early 2025, Musk was publicly excoriating Anthropic — the maker of Claude — as a company that 'hates humanity' and poses existential dangers. He positioned xAI and its Grok model as the ideological counterweight to what he characterized as the misguided safety-theater of Anthropic and OpenAI alike.

Now, Musk is effectively becoming Anthropic's landlord. SpaceX will collect rent on the very infrastructure that was supposed to power Grok's challenge to Claude, GPT-4, and Gemini.

The reversal is jarring but not inexplicable. When the numbers are large enough, ideological battles tend to fade. Anthropic desperately needs compute capacity to keep up with OpenAI and Google DeepMind in the frontier model race, and Colossus 1 represents a ready-made, massive-scale facility that would take competitors years to replicate from scratch.

The Strategic Logic: Why xAI Had to Die for SpaceX to Thrive

The dissolution of xAI did not happen overnight. Sources indicate that the company had been systematically dismantled over the preceding 3 months, with teams reassigned, projects shelved, and infrastructure earmarked for alternative uses. The writing was on the wall long before the official announcement.

The core strategic logic is straightforward and ruthlessly financial:

• As an independent entity, xAI carried its own valuation — one that was increasingly hard to justify against competitors like OpenAI (valued at over $300 billion) and Anthropic (valued at approximately $60 billion)
• Merged into SpaceX, xAI's assets — particularly its compute infrastructure — become part of SpaceX's balance sheet without the burden of standalone AI company metrics
• SpaceXAI has no direct comparable in public markets, making it nearly impossible for analysts to apply traditional valuation discounts
• The Colossus 1 lease transforms a cost center (running AI training) into a revenue-generating asset (collecting rent from Anthropic)

This is valuation engineering at its finest. By killing xAI and rebirthing its assets under SpaceX, Musk creates a narrative where SpaceX is not just a launch company but a diversified technology conglomerate with AI infrastructure exposure — all without having to prove that Grok can compete with Claude or GPT.

IPO Timing: 33 Days That Explain Everything

The timing is the skeleton key that unlocks the entire strategy. SpaceX's IPO roadshow is set to begin on June 8, and every move Musk has made in the AI space over recent months now appears calibrated to maximize the company's pre-IPO narrative.

Investment bankers love simple, powerful stories. 'SpaceX launches rockets AND owns AI infrastructure that Anthropic pays to use' is a far more compelling pitch than 'SpaceX launches rockets and also has a money-losing AI subsidiary trying to catch up with OpenAI.'

The Colossus 1 lease provides something even more valuable than revenue — it provides validation. When one of the world's most valuable AI companies is willing to pay for access to your data center, it signals that the infrastructure has genuine, market-tested value. This is precisely the kind of signal that IPO investors want to see.

Compared to other recent tech IPOs, SpaceX is positioning itself uniquely. While companies like Arm Holdings went public on semiconductor design prowess and Reddit IPO'd on community data value, SpaceX would be the first major IPO to blend aerospace and AI infrastructure into a single valuation story.

What Anthropic Gets: 300 Megawatts of Desperately Needed Compute

For Anthropic, this deal addresses an existential bottleneck. The company has been compute-constrained for months, struggling to allocate sufficient GPU hours across its expanding product lineup. Reports indicate that Claude Code — Anthropic's AI coding assistant — had its user quotas adjusted specifically because compute resources were being redirected to handle Colossus 1 integration.

300 megawatts of data center capacity is enormous. To put this in perspective:

• Microsoft's largest data centers typically range from 200-400 MW
• Google's largest facilities operate at similar scales
• A single 300 MW facility could house approximately 100,000+ GPUs depending on configuration
• This represents enough power to train multiple frontier-scale models simultaneously

Anthropic's willingness to lease from a Musk-controlled entity — despite the very public ideological tensions — underscores just how desperate the compute situation has become in the frontier AI race. Principles are expensive; GPUs are priceless.

The deal also signals a broader industry trend: the separation of AI model development from AI infrastructure ownership. Just as cloud computing separated software companies from data center operations, the AI industry may be entering an era where the companies that train the best models are not necessarily the ones that own the hardware.

The End of Grok's Frontier Ambitions

Perhaps the most underreported aspect of this announcement is what it means for Grok, xAI's flagship large language model. Under the new SpaceXAI structure, the AI division will no longer pursue next-generation frontier models.

This is a significant retreat. Just 6 months ago, Musk was positioning Grok as a serious competitor to Claude 3.5 Sonnet, GPT-4o, and Gemini Ultra. The model had made genuine progress, particularly in its integration with the X platform (formerly Twitter) and its willingness to engage with topics that other AI models refused.

But competing at the frontier requires not just talent — it requires massive, sustained compute investment. By leasing Colossus 1 to Anthropic, Musk has effectively conceded that the economics of frontier AI model development do not work for his portfolio right now. The compute is worth more as a rental asset than as a training resource.

Grok will likely continue to exist in some form within the X ecosystem, but its days as a frontier contender appear to be over. The model may be maintained and incrementally updated, but the ambition to build the next GPT-5 or Claude 4 competitor has been quietly shelved.

What This Means for the AI Industry

This deal reshapes the competitive landscape in several important ways:

• Anthropic gains a significant compute advantage, potentially accelerating Claude's development timeline by months
• OpenAI loses a competitor — with xAI effectively exiting the frontier race, the field narrows further
• AI infrastructure becomes a recognized asset class, separate from AI model development
• The compute shortage narrative intensifies — if even billionaires are choosing to rent out GPUs rather than train models, the economics of frontier AI are more challenging than most investors realize
• IPO markets get a new playbook — SpaceX's approach could inspire other companies to acquire AI assets purely for valuation purposes

For developers and businesses building on AI platforms, the immediate implication is positive. More compute for Anthropic likely means better Claude performance, more API capacity, and potentially lower prices as the company can spread infrastructure costs across a larger resource base.

Looking Ahead: The Next 33 Days and Beyond

All eyes now turn to the June 8 IPO roadshow. The success of Musk's gambit depends entirely on whether investors buy the SpaceXAI narrative — a rocket company that also happens to be a major AI infrastructure player.

If the IPO prices well, expect other companies to follow the playbook: acquire AI assets, lease them to model developers, and use the 'AI exposure' narrative to boost valuations. If it stumbles, the xAI dissolution will be remembered as one of Musk's rare strategic miscalculations.

Meanwhile, Anthropic must rapidly integrate 300 MW of new compute capacity — no small operational challenge. The company's ability to translate this infrastructure windfall into model improvements and product features will be closely watched by the entire industry.

One thing is certain: in the AI industry of 2025, the line between technology strategy and financial engineering has never been thinner. Musk, ever the dealmaker, appears to understand this better than anyone.