Written by: Xiao Bing, Deep Tide TechFlow

If three months ago you told a Silicon Valley investor that Musk would rent out the entire Colossus 1 training cluster of xAI to Anthropic, he would probably laugh out loud.

After all, in February, Musk was still insulting Anthropic on X for "hating Western civilization," and in March, he nicknamed Anthropic "misanthropic." In Musk's eyes, this company was practically a synonym for politically correct AI, an opponent that, like OpenAI, must be toppled.

Then on May 6, Anthropic and SpaceX jointly announced: Anthropic will receive the entire computing power of Colossus 1, over 220,000 NVIDIA GPUs, 300 megawatts of power capacity, delivered within a month. Anthropic clearly stated that this computing power will be directly used to improve the service experience for Claude Pro and Claude Max subscription users.

Musk posted something on X that left everyone stunned: he said he had deep interactions with Anthropic executives in the past week and was "impressed," noting, "They are all very capable and are serious about doing the right thing." He even stated that Claude "will probably be good."

On the same day, he announced that xAI would dissolve as an independent company and rename itself SpaceXAI.

This is a transfer of capacity

Mainstream English media wrote this off as a "landmark event in AI computing power sharing," but they missed a key fact:

Colossus 1 is xAI's core training facility, not some "spare capacity."

Let's review the timeline. Colossus 1 was completed in Memphis in September 2024, taking only 122 days from groundbreaking to powering up, a miracle in data center construction history. This is the main cluster used by xAI to train Grok 3 and Grok 4 and is the physical embodiment of Musk's narrative of "computing power equals power." It is equipped with over 220,000 GPUs, including H100, H200, and the latest GB200, with the cluster's scale once ranking in the top three globally by the end of 2025.

Handing over a training cluster of this scale to a direct competitor is equivalent to TSMC renting out the entire capacity of its 5-nanometer production line to Samsung. Such a thing has never happened in the semiconductor industry. Anyone who has observed the cycles knows that this kind of action only occurs under one circumstance: when you can't use it all yourself.

The official statement from SpaceXAI is that Anthropic's computing power will "directly benefit Claude Pro and Claude Max subscription users." In other words, Anthropic is using this portion of computing power for inference, running models for Claude's paying users, catering to the AI that Musk dislikes the most.

Summarizing this matter with the phrase "customer collaboration" is inaccurate; the actual control of Colossus 1 has, in a sense, changed hands.

The story of Grok does not support the size of Colossus

Why is there "unused capacity"?

The most straightforward answer lies in Grok's user data.

According to data released by Similarweb in April, Grok's daily active users (DAU) in global mobile applications dropped from 13.9 million in March to 12.2 million in April, a 12.5% month-over-month decline. In the U.S. market, the drop was more severe, from 1.4 million to 1.1 million, a -15.6% decline. A year ago, it was the second-largest AI application in the world, second only to ChatGPT; by April, it had fallen to fifth place, surpassed by Claude, Gemini, and DeepSeek in sequence.

During the same period, Claude's DAU increased from 16 million to 23 million, a month-over-month growth of +44%.

This presents a stark contrast: in 2026, when AI applications were generally experiencing rapid growth, Grok is among the few leading products that are losing users. The reason is simple: Grok's core scenarios have always been tied to the X (formerly Twitter) platform as a tool for "real-time search + spicy commentary." However, it has never formed the kind of "workflow stickiness" like Claude has on its standalone app and web version. Numerous users on Reddit have complained that Grok has gradually moved its image and video generation functions behind a paywall, and with regulatory investigations in several countries and threats of bans from Apple, its growth engine has nearly stalled.

More critically, within xAI.

According to a report by **Fast Company** in April, more than 80 employees have left xAI in the past few months, including several co-founders. A report by **Financial Times** in February mentioned that Musk has been putting "unreasonable pressure" on the team regarding technical metrics, trying to catch up with competitors, which is a typical reaction from a leader during a period of product declines.

Looking at these two events together, the answer to why Colossus 1 has excess capacity becomes clear: it was initially prepared for a Grok that was much larger than it is now.

The real challenge for SpaceXAI lies in valuation narratives

"Insufficient demand for Grok" is just the surface.

A deeper logic is: Musk needs a new story to support SpaceXAI's $1.25 trillion valuation.

Recall what happened in February this year. SpaceX acquired xAI entirely in stock, and after the merger, its entity valuation was set at $1.25 trillion, the largest merger in history. Prior to the merger, xAI's last round of financing was the Series E in January, for $20 billion at a valuation of $230 billion. Incorporating xAI into SpaceX is essentially using the cash flow from SpaceX's rocket business to keep xAI, which continues to lose $1.46 billion every quarter, alive.

But even with SpaceX's financial support, SpaceXAI still faces a sharp question: what justifies this valuation?

The most recent valuation of OpenAI is $852 billion, with an ARR of about $24-25 billion, resulting in a valuation/revenue ratio of about 35 times. The valuation that Anthropic is discussing is $900 billion, with an ARR of $30 billion and a valuation/revenue ratio of 30 times.

What about xAI? For the third quarter of 2025, the revenue is projected at $107 million, with a net loss of $1.46 billion. Even if we optimistically estimate Grok's expected revenue for 2026 at $2 billion, the corresponding valuation/revenue ratio for that portion of SpaceXAI would still far exceed that of OpenAI and Anthropic. In other words, Musk desperately needs to tell a new cash flow story for SpaceXAI. He cannot do it based on Grok's user growth, nor can it be done through enterprise API revenue.

Renting Colossus to Anthropic is the beginning of that story.

It instantly reoriented SpaceXAI from being a "model company" to an "AI cloud infrastructure provider," a player somewhat like CoreWeave but with larger scale and power supply. In the world of narrative valuation, cloud providers are worth more than model companies. Cloud providers can offer long-term contracts and predictable cash flow, which pure model companies struggle to provide.

Along with Anthropic and SpaceX's vague "orbital computing center" memorandum, where both parties agreed to "explore" deploying multi-gigawatt-level AI data centers in space, you can understand where all of this is headed. This is a new balance sheet being prepared for SpaceX's IPO. Rockets, Starlink, ground data centers, orbital computing — all bundled into a giant infrastructure story. Grok itself has become irrelevant; what matters are the GPUs, power, and launch sites Musk controls.

The real meaning of Musk's 180-degree turn in attitude

Within this framework, Musk's dramatic shift in attitude toward Anthropic gains a different explanation.

It is a deal.

Apart from rent, Anthropic provides something else to SpaceXAI: credit endorsement. By publicly endorsing the availability, scalability, and operational quality of Colossus 1, Anthropic effectively grants SpaceXAI a ticket to join the "computing infrastructure club." Members of this club include AWS, GCP, Azure, and CoreWeave. Prior to this, xAI's reputation in the cloud services market was virtually zero; it had only used computing power to train its own models and had never truly operated as an external service provider.

For Anthropic, this deal is also extremely cost-effective. It is financing with a valuation of $900 billion and may IPO in October. Its publicly disclosed need is for 5 gigawatts of training computing power; the 300 megawatts provided by SpaceX may not seem significant, but its value lies in "immediate delivery": powered within a month to directly alleviate the inference pressure on Claude. In April, Anthropic publicly admitted that Claude's "reliability and performance" were affected by "infrastructure pressure" during peak hours. The 300 megawatts of emergency capacity is worth far more than the numerical value on the books.

This is a reciprocal narrative transaction. Anthropic gains service stability, while SpaceXAI gains a valuation story.

Who made concessions? Musk indeed lowered his stance, he did business with an old rival and even spoke well of them. But on a deeper level, it is Grok that has made concessions. Grok, as a product, as a model company, as Musk's flagship weapon against OpenAI/Anthropic, is quietly being downgraded to an ordinary business within the SpaceXAI portfolio. The freeing up of core strategic assets like Colossus for customer use signifies that Musk no longer considers "self-developed models" as his main battlefield.

In this sense, May 6 marks the end of Grok's era as a "cutting-edge model company."

Sign of a turning point in the AI industry: capacity begins to concentrate among a few players

Zooming out a bit, the industry significance of this event may be larger than we can currently see.

Throughout 2024 and 2025, the AI computing power market is in a state of "all-industry frenzy." OpenAI is scrambling, Anthropic is scrambling, xAI is scrambling, Mistral is scrambling, and sovereign funds from various countries are scrambling. GPUs are hard currency, data center site selection is a geopolitical issue, and power supply is a national strategic matter. In this environment of universal shortage, no one will rent their training clusters to competitors because every GPU hour you rent out today could be critical computing power in chasing down competitors tomorrow.

Yet now, xAI has done it.

This means that the AI computing power market is beginning to show a first wave of differentiation: the computing power demands of leading model companies (OpenAI, Anthropic, Google DeepMind) are still growing exponentially, while second-tier and below model companies are starting to experience capacity loosening. This differentiation appears in the mid-to-late stages of all capacity expansion cycles, from solar batteries to electric vehicle batteries to Bitcoin miners; the script is almost identical. Initially, everyone lacks capacity, in the mid-stage, capacity begins to overflow to second-tier players, and in the late stage, first-tier players consolidate upstream and downstream, while second-tier players either transform into infrastructure service providers, get acquired, or go bankrupt.

CoreWeave is the best contrast. Originally an Ethereum mining operation, it seized the window of GPU surplus to transform into AI cloud in 2018, and in 2024, it went public with a market value of $60 billion. Its existence itself proves that the path of "if models don’t work, then turn to computing power" is viable. SpaceXAI is retracing this path, in a more radical version; in addition to selling ground computing power, Musk also aims to sell computing power in space.

The true signal of the peak of the AI bubble may be second-tier model companies all transforming into cloud service providers. When the core narrative of an industry shifts from "I have the best model" to "I have the most GPUs," it usually indicates that differentiated competition has reached its end.

A noteworthy detail is that in Memphis, where Colossus 1 is located, xAI deployed dozens of natural gas turbines to supply power to expedite construction, claiming "temporary use" requires no federal permit. Local residents have protested continuously over air pollution issues, and this matter has not calmed down to this day.

And now, these GPUs powered by gas turbines will be used to run Anthropic's Claude, one of the labs with the most stringent positions on AI ethics and climate issues.

Even more absurd is that Anthropic and SpaceX "expressed interest" in deploying multi-gigawatt-level AI computing power in orbit in their announcement. Musk's reasoning is that the power and cooling on Earth will eventually become insufficient, and the future of AI lies in space.

A huge expectation for valuation lies between Memphis's gas turbines and the solar panels in Musk's presentation. Renting Colossus 1 to Anthropic is the first new story Musk tells for this valuation expectation.

xAI transformed from an adversary to a supplier in just three months. Who will be the next to be repriced?