If three months ago you told a Silicon Valley investor that Musk would lease the entirety of xAI’s largest training cluster, Colossus 1, to Anthropic, they’d likely have laughed out loud.

After all, in February, Musk was still denouncing Anthropic on X as “hating Western civilization,” and in March he coined the nickname “misanthropic” for the company. To Musk, Anthropic was practically the epitome of politically correct AI—an adversary on par with OpenAI that had to be dismantled.

Then on May 6, Anthropic and SpaceX jointly announced: Anthropic would receive the full compute capacity of Colossus 1—over 220,000 NVIDIA GPUs and 300 megawatts of power—delivered within one month. Anthropic explicitly stated this compute would directly enhance the service experience for Claude Pro and Claude Max subscribers.

Musk posted on X something that stunned everyone: He revealed deep engagement with Anthropic’s executives over the past week, calling them “impressive,” “highly capable,” and “doing the right things seriously.” He even said Claude “is probably going to be good.”

On the same day, he announced xAI would dissolve as an independent entity and rebrand as SpaceXAI.

This Is a Capacity Concession

Mainstream English media described this event as a “landmark moment of AI compute sharing”—but they missed a crucial fact:

Colossus 1 is xAI’s most critical training facility, not some “backup capacity.”

Let’s revisit the timeline. Colossus 1 was completed in Memphis in September 2024, constructed from ground-breaking to power-up in just 122 days—a miracle in data center history. This cluster is the primary engine behind training Grok 3 and Grok 4, and serves as the physical manifestation of Musk’s “compute is power” narrative. It houses over 220,000 GPUs, including H100, H200, and the latest GB200, and at its peak in late 2025 ranked among the world’s top three in scale.

Leasing such a massive training cluster entirely to a direct competitor is equivalent to TSMC renting out its entire 5nm production line to Samsung. Such a move has never happened in the semiconductor industry. Anyone familiar with market cycles knows this kind of action only occurs when capacity is underutilized.

SpaceXAI’s official statement claims Anthropic’s compute will “directly benefit Claude Pro and Claude Max subscribers.” That means Anthropic is using this capacity for inference—running models for paid Claude users, effectively serving requests for the very AI Musk detests most.

Labeling this as “customer collaboration” is misleading. In practical terms, control of Colossus 1 has, to some extent, changed hands.

Grok’s Story Can’t Justify Colossus’ Scale

Why would there be “excess capacity”?

The most direct answer lies in Grok’s user data.

According to Similarweb’s data from April, Grok’s global daily active users (DAU) dropped from 13.9 million in March to 12.2 million in April—a 12.5% MoM decline. In the U.S. market, the drop was steeper: from 1.4 million to 1.1 million, a -15.6% MoM. A year ago, it was the world’s second-largest AI app after ChatGPT; by April, it had fallen to fifth place, overtaken sequentially by Claude, Gemini, and DeepSeek.

Meanwhile, Claude’s DAU rose from 16 million to 23 million during the same period—a +44% MoM surge.

This is a stark contrast: In 2026, an era of universal AI app growth, Grok stands out as one of the few major products losing users. The reason is straightforward—Grok’s core use cases have remained tightly bound to the X (formerly Twitter) platform, functioning primarily as a “real-time search + sharp commentary” tool. Yet it has never achieved the “workflow stickiness” seen in Claude across standalone apps and web platforms. Numerous Reddit users complain about Grok gradually moving image and video generation features behind paywalls, compounded by regulatory investigations in multiple countries and Apple’s threat of banning the app—effectively killing its growth engine.

Even more troubling is what’s happening internally at xAI.

According to a Fast Company report in April, over 80 employees—including several co-founders—have left xAI in recent months. Meanwhile, a Financial Times report from February highlighted Musk’s persistent imposition of “unreasonable technical targets” on the team, pushing them to catch up with competitors—a classic symptom of leadership in a product’s decline phase.

When viewed together, the reason for Colossus 1’s surplus capacity becomes clear: It was originally built for a Grok much larger than the current one.

The Real Challenge for SpaceXAI Lies in Valuation Narrative

“Insufficient demand from Grok” is merely surface-level.

Beneath it lies a deeper logic: Musk needs a new story to justify SpaceXAI’s $1.25 trillion valuation.

Recall what happened in February. SpaceX acquired xAI via a stock-only deal, creating a merged entity valued at $1.25 trillion—the largest merger in history. Prior to the merger, xAI’s last funding round (Series E in January) was $20 billion at a $230 billion valuation. Folding xAI into SpaceX essentially used SpaceX’s rocket business cash flow to sustain xAI’s burn rate—still losing $1.46 billion per quarter.

But even with SpaceX’s financial infusion, SpaceXAI faces a sharp question: Why is it worth this price?

OpenAI’s latest valuation: $852 billion, ARR ~$24–25 billion, valuation-to-revenue ratio ~35x. Anthropic’s ongoing talks for a $90 billion valuation, with $30 billion ARR, implies a 30x valuation-to-revenue ratio.

What about xAI? Revenue in Q3 2025: $107 million, net loss: $1.46 billion. Even if we optimistically project Grok’s 2026 revenue at $2 billion, SpaceXAI’s implied valuation-to-revenue ratio would still far exceed those of OpenAI and Anthropic. In other words, Musk urgently needs a new cash flow narrative for SpaceXAI. Growth driven by Grok’s user base won’t work. Enterprise API revenue won’t either.

Leasing Colossus 1 to Anthropic is precisely the beginning of that narrative.

It instantly repositions SpaceXAI from a “model company” to an “AI cloud infrastructure provider”—a player akin to CoreWeave, but on a significantly larger scale with greater power supply capacity. In the world of narrative-based valuation, cloud providers are worth more than model companies. Cloud vendors can present long-term contracts and predictable cash flows—something pure model companies struggle to offer.

Combined with the vague “orbital compute center” memo shared by Anthropic and SpaceX, which expresses interest in exploring the deployment of multi-gigawatt AI data centers in space, the direction becomes clear. This is a new balance sheet crafted for SpaceX’s IPO. Rockets, Starlink, ground data centers, orbital compute—all bundled into a single, massive infrastructure story. Grok itself has become irrelevant. What matters now is the GPUs, power, and launch slots in Musk’s possession.

The True Meaning Behind Musk’s 180-Degree Shift

Under this framework, Musk’s dramatic change in stance toward Anthropic takes on a different meaning.

It’s a transaction.

Anthropic offers SpaceXAI more than rent—it brings credibility endorsement. By publicly endorsing Colossus 1’s availability, scalability, and operational quality, Anthropic effectively grants SpaceXAI entry into the “compute infrastructure club”—a group including AWS, GCP, Azure, and CoreWeave. Prior to this, xAI had no reputation in the cloud services market; it had only ever used compute to train its own models, never operating commercially for external clients.

For Anthropic, this deal is equally compelling. It’s currently fundraising at a $90 billion valuation, possibly preparing for an IPO in October. Its public need is for 5 gigawatts of training compute. While SpaceX’s 300 megawatts may seem modest, its value lies in immediate delivery: power-on within a month, directly alleviating immediate inference pressure on Claude. In April, Anthropic openly admitted that Claude’s reliability and performance were affected during peak hours due to “infrastructure strain.” The emergency capacity of 300 megawatts carries value far exceeding its nominal figures.

This is a two-way narrative transaction. Anthropic gains service stability; SpaceXAI gains a valuation story.

Who made concessions? Musk himself clearly bowed—he did business with his longtime rival and praised them publicly. But on a deeper level, the concession was made by Grok. As a product, as a model company, as Musk’s flagship weapon against OpenAI and Anthropic, Grok is being quietly downgraded to just another ordinary business line within SpaceXAI’s portfolio. The fact that a core strategic asset like Colossus is now being leased to customers signals Musk no longer sees “in-house models” as the main battlefield.

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

Industry Inflection Point: Compute Capacity Concentrating Among Few Players

Zooming out further, the broader industry implications of this event may be even larger than currently apparent.

Throughout 2024 and 2025, the AI compute market was in a state of “full-scale scramble.” OpenAI competed, Anthropic competed, xAI competed, Mistral competed, sovereign wealth funds competed. GPUs became hard currency; data center locations became geopolitical issues; power supply became a national security concern. Under such universal scarcity, no one would rent their training clusters to rivals, because every GPU hour leased today could be the key compute edge tomorrow.

Now, xAI has done it.

This signals the first crack in the AI compute market’s homogeneity: Top-tier model companies (OpenAI, Anthropic, Google DeepMind) continue seeing exponential growth in compute demand, while mid-tier and lower-tier model companies are starting to show signs of capacity slack. This divergence appears in the mid-to-late stages of every capacity expansion cycle—from solar panels to EV batteries to Bitcoin mining rigs. The script is nearly identical: early stage, everyone lacks; mid-stage, excess capacity emerges for second-tier players; late stage, first-tier players integrate upstream and downstream, while second-tier players either pivot to infrastructure services, get acquired, or fail.

CoreWeave is the perfect case study. Originally an Ethereum mining farm, it seized the GPU oversupply window in 2018 to pivot into AI cloud computing and went public in 2024 with a $60 billion valuation. Its existence proves the path “if your model fails, switch to compute” is viable. SpaceXAI is now retracing this path—more aggressively, selling not just ground-based compute but also extending into space.

The true signal of the AI bubble’s peak may be when mid-tier model companies begin transforming into cloud service providers. When the core industry narrative shifts from “I have the best model” to “I have the most GPUs,” it typically signals the end of differentiation-driven competition.

A telling detail: During Colossus 1’s construction in Memphis, xAI deployed dozens of natural gas turbine generators to power the facility, claiming “temporary use” and requiring no federal permits. Local residents have been protesting ongoing air pollution issues—protests that remain unresolved to this day.

Now, these GPUs powered by gas turbines will run Anthropic’s Claude—one of the most ethically rigorous labs in AI, especially on climate and ethics issues.

Even more absurdly, Anthropic and SpaceX expressed “interest” in deploying multi-gigawatt AI compute in orbit. Musk’s logic: Earth’s power and cooling will eventually be insufficient; the future of AI is in space.

Between the gas turbines in Memphis and the orbital solar panels in Musk’s PPT slides lies a vast valuation expectation. Leasing Colossus 1 to Anthropic is Musk’s first new story for that expectation.

xAI transformed from adversary to supplier in just three months. Who will be next to be revalued?

Author: Written by Xiao Bing, DeepTide TechFlow

Source: DeepTide TechFlow