Musk Gave Claude 220,000 GPUs: Is Altman Going to Collapse in His Chair Now?

The enemy of my enemy is my friend. Musk has truly mastered the art of war with this move.

Just yesterday, Anthropic dropped an announcement out of the blue, declaring a partnership with SpaceX.

You read that right. Anthropic, the AI company, and SpaceX, the rocket builder—two companies that seem completely unrelated—are suddenly in the same trench.

Specifically, Anthropic can now use the full computing power of the Colossus 1 data center in Memphis, owned by SpaceX.

What does that mean? Within a month, Anthropic will gain over 300 megawatts of computing power, equivalent to more than 220,000 Nvidia GPUs (including H100, H200, and the latest GB200) for them to burn through.

I have to say, Musk's move is pretty ruthless.

Considering how fiercely Musk's lawsuit against OpenAI is being fought, he turns right around and happily partners with Anthropic.

Whether OpenAI feels the pressure is unclear, but Anthropic is absolutely thrilled. Never having fought such a well-resourced battle, they generously raised Claude's usage limits, giving users a reason to celebrate.

Notably, shortly after the Anthropic-SpaceX partnership was announced, Musk responded on X that xAI would be dissolved as an independent company, its products merged into SpaceX, and even renamed to SpaceXAI.

This string of actions inevitably makes one wonder if Musk hates Altman so much that he came up with this trick. Some have even imagined a dramatic, soap-opera-level plot...

Others are egging on the drama: dissolving xAI and giving compute power to Anthropic—what happens to Grok?

Actually, things aren't as complicated as we think.

Anthropic's motivation for this partnership is pure: it's all about compute power. The announcement even specifically showed off the major compute deals they had signed.

What's more confusing might be Musk's 180-degree turn in attitude.

After all, Musk chasing Anthropic online to criticize them isn't a new thing; he's called them anti-human and hateful toward Western civilization.

But lo and behold, after meeting with the Anthropic team last week, he immediately changed his tune, saying they didn't trigger his "evil detector."

The speed of this about-face is so fast that even a master of Sichuan opera face-changing would have to buy a standing ticket overnight to learn from him.

Jokes aside, to understand why Musk rented compute power to Anthropic and suddenly dissolved xAI, we might need to clarify a few basic facts first.

First of all, the so-called "dissolution" of xAI is not a real dissolution.

As early as February this year, xAI had already been acquired by SpaceX. Musk's reply on X was merely revealing an internal restructuring: xAI transformed into SpaceXAI. The signboard changed, but the business continues as usual.

According to internal documents disclosed by CNBC, this record-breaking deal is the largest merger and acquisition in history.

At the time, SpaceX was valued at $1 trillion, and xAI was valued at $250 billion. With the merger, a $1.25 trillion super-giant was born.

But Musk's ambition clearly doesn't stop there.

Because SpaceX is about to IPO. According to reports from Bloomberg and Reuters, SpaceX might seek a valuation of $1.75 trillion and go public around June this year.

But boosting the valuation further in a matter of months is no easy task.

So at this critical juncture, Musk has to find a way to paint a $1.75 trillion super-rosy picture. Rather than slaving away building rockets and developing AI models, the concept of "space compute power" is much sexier.

This includes the mention in the agreement with Anthropic that they might collaborate to develop several GW of orbital AI computing capability, using a hot commodity to back up his claims.

One can only say that Musk is working incredibly hard to pave the way for his IPO.

Secondly, renting compute power to Anthropic doesn't mean giving up on Grok.

Many might not have realized it before, but Musk indeed holds the world's largest AI supercomputing center, Colossus.

Take Colossus 1, for example. This giant, jointly built by xAI and Nvidia two years ago, took only 122 days from construction to operation, with over 220,000 GPUs dedicated exclusively to Grok.

The awkward part is that even sitting on top-tier computing resources, Grok hasn't fully capitalized on this massive advantage; its skills have all been pointed at generating NSFW content.

Moreover, the compute power of Colossus 1 is actually overkill for Grok.

According to a previous leak by The Information, Grok's compute utilization rate on Colossus 1 is barely 11%. It simply can't squeeze the performance out of those 220,000 cards.

So, by packaging and renting Colossus 1 to Anthropic, is he leaving Grok out to dry?

Of course not, because Colossus 2 is freaking here.

At the beginning of this year, the world's first GW-level supercomputing cluster, Colossus 2, officially went into operation.

The scale of Colossus 2 is even more insane, packing 550,000 NVIDIA GB200/GB300 GPUs. The procurement cost for chips and related hardware alone soared to about $18 billion.

Currently, SpaceXAI's training tasks have migrated to Colossus 2, beginning the hard work on the next generation of Grok.

Actually, Musk is extremely shrewd. Colossus 1 is freed up; hundreds of thousands of cards sitting idle still cost money to run the data center. It's better to rent out the compute power and make some easy money.

In today's AI model circle, someone releases a Claude Opus 4.7 today, someone else pulls out a GPT-5.5 tomorrow, and a Gemini 4 might pop up the day after. This endless model race—who knows when it will end.

And to be fair, whether it's a "Big Three" or a "Dual Hegemony," Grok always seems half a beat slow, never catching the hot meal.

Musk, having tinkered with Grok for a while, probably realized this too.

Rather than going head-to-head with these fiercely competitive rivals in the arena, it's better to build an impressive AI supercomputing center and make a massive score.

Earning rent from Anthropic to recoup funds, feeding his own brainchild Grok, and simultaneously using Anthropic to constantly irritate OpenAI.

He makes money and vents his anger.

As for the overhyped space compute power, just listen to it for now; it's more like a pie Musk custom-baked for Wall Street to push toward that $1.75 trillion IPO.

One just wonders if, at this very moment, Sam Altman is slumped in his chair, looking at Anthropic's suddenly enriched computing resources.

Source: Chapingjun