Anthropic Is Now Paying Elon Musk $1.25 Billion a Month to Stay Alive

⚡ Originally published on my Substack — read the canonical version here. I publish daily AI + Big Tech hot takes there, short and sharp. Subscribe free.

Not a member?

On May 20, SpaceX filed its IPO prospectus. Buried inside, in a section most reporters skimmed, is a procurement contract that reads more like a hostage note. Anthropic — the lab that built its entire brand around being the safer alternative to OpenAI — has agreed to pay Elon Musk $1.25 billion every 30 days, for the next three years, just to keep Claude online.

The Twitter feud is over. The check has cleared. And the irony is so thick the filing should ship with a content warning.

The receipt

Here is exactly what SpaceX disclosed to the SEC on May 20:

• $1.25 billion per month from Anthropic to SpaceX, payable through May 2029.
• Roughly $15 billion a year. Up to $45 billion over the full term.
• Exclusive access to Colossus I and Colossus II — together more than 300 megawatts of power and over 220,000 NVIDIA GPUs.
• Anthropic’s compute chief Tom Brown confirmed in May the capacity is being used for inference, not training.
• Either party can terminate the agreement with 90 days’ notice, no cause required.

That last bullet is the most important number in the entire deal. We’ll come back to it.

For context: Anthropic just hit a $30 billion annualized revenue run rate in April. Roughly half their entire revenue is now committed to a single supplier — one majority-owned by Elon Musk, the founder of their most direct competitor.

Anthropic has never been more dependent on any single human. That human is the guy who, three months ago, called them evil on a public timeline of 200 million people.

Three months ago, he called them “misanthropic and evil”

On February 12, 2026, Anthropic announced a $30 billion Series G at a $380 billion post-money valuation — one of the largest startup raises in history.

Within hours, Elon Musk posted this on X:

“Your AI hates Whites & Asians, especially Chinese, heterosexuals and men. This is misanthropic and evil. Fix it. Frankly, I don’t think there is anything you can do to escape the inevitable irony of Anthropic ending up being Misanthropic. You were doomed to this fate when you chose your name.”

No examples. No data. No screenshots. Just the founder of a competing AI lab telling 200 million followers that Anthropic’s product was racist and evil and doomed by its own name.

That tweet is still up. It has 14 million views and counting.

It is the most expensive tweet of Elon Musk’s career — because three months later, the people he was insulting wired him $1.25 billion.

The Jackass Trophy

Here’s the part of this story almost nobody covering the S-1 has mentioned.

In 2018, when Musk lost his power struggle with Sam Altman and walked out of OpenAI, he held a farewell all-hands. An employee named Joshua Achiam pushed back during Q&A — telling Musk that prioritizing “the race to beat Google” over safety was the wrong move. Musk, in front of the entire company, called him a jackass.

The next day, Dario Amodei and David Luan — both still at OpenAI at the time — presented Achiam with a custom-made Jackass Trophy. For standing up to Musk on safety grounds. The trophy became an internal legend.

That was Dario.

Today, Dario Amodei is the CEO of Anthropic, the lab he founded specifically to avoid Musk’s “speed-over-safety” framing. And every 30 days, his finance team wires Elon Musk one billion two hundred and fifty million dollars.

The Jackass Trophy is still on a shelf somewhere. It just costs $1.25 billion a month to keep it.

Why Anthropic had no other option

You might reasonably ask: why sign with Musk specifically? Why not lean harder on Amazon, Microsoft, or Google?

The answer is they’re already maxed out everywhere else — and it’s still not enough.

• Amazon: up to $25 billion invested, securing up to 5 gigawatts of training and inference capacity. Claude runs on AWS Trainium chips.
• Microsoft + NVIDIA: $15 billion committed this year. In exchange, Anthropic locked in $30 billion of Azure compute plus up to 1 gigawatt of NVIDIA Grace Blackwell and Vera Rubin systems.
• Google: holds roughly a 14% stake from earlier rounds. Claude also runs on Google TPUs.
• SpaceX / Colossus (new): 300 MW, 220,000 GPUs, $45 billion over three years.

Add it up. Anthropic has, on paper, committed to over $120 billion in compute across four suppliers. And they’re currently in talks to raise another $30 billion at a pre-money valuation north of $900 billion.

This is not a company. It’s a megawatt-acquisition vehicle wearing a chatbot.

And even that wasn’t enough. Claude’s inference traffic — every API call, every Claude Code session, every Cursor agent running Opus 4.6 — is growing faster than the existing partnerships can provision GPUs. They needed another 300 MW. Yesterday.

The only seller with hundreds of thousands of unallocated H100s sitting in a warehouse is xAI — the stockpile Musk originally built for Grok, now sitting under the SpaceX corporate umbrella.

There was no one else to call. So Dario picked up the phone and called the guy who had spent three months calling him evil.

Musk now has a 90-day kill switch on Claude

Remember that termination clause? Either party. 90 days. No cause required.

What this means in practice: at any moment in the next three years, Elon Musk can serve Anthropic a 90-day notice and start clawing back GPU capacity. He doesn’t need a reason. He doesn’t need a court order. He just needs to feel like it.

Over the following 90 days, Anthropic would have to move billions of dollars of inference workload onto AWS Trainium and Azure — both already running near capacity. Or watch Claude’s latency, rate limits, and downtime visibly degrade for every Pro user, every API customer, every Cursor agent on the planet. Or both.

It wouldn’t destroy Anthropic. But it would absolutely tank the IPO they’re walking toward. And it would hand xAI an unmissable competitive opening: “Grok stayed up. Claude went down.”

That’s not a hypothetical. That’s a clause in the actual contract.

The single most powerful person in Anthropic’s near future is no longer Dario Amodei. It’s the guy who, three months ago, told 200 million followers that Anthropic was “doomed to this fate when you chose your name.”

What every Claude Pro user just signed

Here’s the part that doesn’t make it into the IPO filing.

If you pay Anthropic $20/month for Claude Pro or $200/month for Claude Max, a meaningful chunk of that bill now flows through Anthropic, out the door, and into SpaceX’s compute revenue line.

SpaceX uses that money to fund Starship development.

Which means, transitively, Elon Musk is using your Claude subscription to fly rockets.

He didn’t change his mind about Anthropic being “evil.” He just figured out how to monetize the fact that you disagree.

Recall: in March, the Pentagon demanded Anthropic sign off on autonomous-weapons use and mass surveillance. Anthropic refused, lost the contract, and got blacklisted by the Trump administration. OpenAI signed within hours. Sam Altman publicly apologized. And then — in one of the most unambiguous market endorsements in consumer-AI history — Claude surged to #1 on the App Store, on a wave of goodwill for the company that had said no.

Three months later, that same company — the principled holdout, the “safety-first” lab, the alternative to Big Tech AI — quietly signed the largest single-customer compute deal in Elon Musk’s career.

Principles are expensive. So is inference.

What the IPO filing will eventually say

When Anthropic goes public — and it will, probably in 2027 — page one of their S-1 will list their five largest cost commitments. Pages 30 through 50 will be the risk factors. And somewhere in there, in language drafted by very nervous lawyers, will be a sentence that reads something like:

“We depend on a single supplier for a material portion of our inference infrastructure. That supplier is also one of our most direct competitors.”

Fund managers will read that sentence. Some of them will laugh. Some of them will skip the IPO.

But none of them will be surprised. Because by 2027, every “safe AI” pitch in the industry will sound exactly like it does now — earnest, well-funded, and renting GPUs from someone who hates them.

If this resonated, the full version (with subscribe + share widgets) lives on my Substack — read it here and consider subscribing for free to get the next one in your inbox the day it ships.

If this hit, I’m writing daily AI + Big Tech hot takes on Substack now — short, sharp, the stuff Medium articles can’t move fast enough on. Subscribe free at https://noobprogrammer16.substack.com (recent posts: Meta firing 8,000 to hire 7,000 AI agents, OpenAI letting GPT-5.5 plan its own launch party, and Andreessen calling every AI layoff a lie).

References

• TechCrunch — Anthropic will pay xAI $1.25B per month for compute (May 20, 2026)
• Axios — Anthropic is paying SpaceX $15 billion per year (May 20, 2026)
• SEC.gov — SpaceX Form S-1 (FY2026)
• CNBC — Anthropic closes $30B funding round at $380B valuation
• Fox Business — Musk slams Anthropic AI as “misanthropic and evil”
• Axios — How Elon grew to love Anthropic
• OfficeChai — The Jackass Trophy story
• VentureBeat — Anthropic hits $30B revenue run rate