Buried in the SpaceX S-1 — the filing that priced the largest IPO in market history at $135 a share on Friday — is a disclosure that has very little to do with rockets. Per TechCrunch’s coverage of the prospectus, Anthropic is paying $1.25 billion a month and Google is paying $920 million a month in pre-IPO compute commitments tied to Elon Musk’s xAI division. Combined: $2.17 billion a month. Roughly $26 billion a year. Disclosed inside a rocket-company prospectus.
If those numbers hold up post-prospectus scrutiny, this is not a footnote. It is the most expensive frenemy disclosure in the modern AI economy — and the clearest sign yet that the compute crunch has structurally inverted the relationship between model labs and their nominal competitors.
Anthropic’s $1.25 billion monthly commitment to xAI infrastructure is a tell. Anthropic already has multi-year compute partnerships with Amazon (the largest, Trainium-anchored) and Google (TPU access). Adding xAI on top — at a run rate of $15 billion a year — only makes sense if existing supply is binding.
That points directly at the Colossus cluster: the Memphis-based GPU supercluster xAI scaled to over 200,000 H100/H200-class GPUs in 2024-25. Until now, the assumption was that Colossus existed to train Grok. The S-1 disclosure suggests the more accurate framing is that Colossus is a two-sided compute platform — training Musk’s own models and renting capacity to whoever can pay.
If Anthropic is willing to write a 10-figure monthly check to a competitor’s data center, the inference is straightforward: it can’t get enough capacity anywhere else fast enough.
Why Google Is Also on the List
Google’s $920 million monthly commitment is the stranger line item. Google has TPUs. Google has Google Cloud. Google has more in-house compute than any company on Earth except possibly Microsoft. Why would it pay xAI nearly $11 billion a year for capacity?
Two plausible answers, neither flattering to the conventional narrative:
One:Google’s internal AI workloads have outgrown even TPU supply. Gemini training, Veo, the agent infrastructure, the AI Overviews stack — all running at scale that exceeds organic capacity growth. Renting from xAI is faster than building.
Two:Google is hedging the model race by underwriting a competitor’s data center economics in exchange for option value — first look at capacity, partnership leverage, or strategic intelligence on Colossus utilization. The DeepMind playbook has always been to be everywhere in the value chain.
Three structural things, all visible from this one disclosure:
1. Compute is the only scarce input that matters in 2026. Talent has loosened (compensation arbitrage is closing). Data is plentiful (synthetic generation works). Capital is abundant for any lab with a credible scalingstory. But power-adjacent GPU capacity at scale is still the bottleneck — and the people who own clusters are the ones collecting rent.
2. xAI is becoming a compute landlord, not just a model lab. A model lab that rents $26 billion a year of capacity to its competitors looks economically more like a cloud provider than an AI startup. That is a different business — different margins, different valuation comps, different competitive dynamics.
3. The SpaceX IPO is partly a Musk-empire balance-sheet event. By disclosing xAI’s compute deals inside the SpaceX prospectus, the filing makes those revenue streams visible to public-market investors for the first time — without xAI itself going public. That is convenient. It also revives speculation, flagged by the S-1’s own language on “significant equity dilution” in future transactions, that the next step is a Tesla–SpaceX–xAI consolidation.
The Takeaway
If the numbers in the prospectus hold, the most important fact in the SpaceX IPO is not the $135 share price or the $75 billion proceeds. It is that the two most strategically credible AI labs outside OpenAI — Anthropic and Google — are writing combined $2.17 billion-a-month checks to Elon Musk’s compute cluster. That is the compute crunch made legible in a public filing.
The rest of the prospectus is about rockets. This part is about the AI economy’s actual power structure.
Gennaro is the creator of FourWeekMBA, which reached about four million business people, comprising C-level executives, investors, analysts, product managers, and aspiring digital entrepreneurs in 2022 alone | He is also Director of Sales for a high-tech scaleup in the AI Industry | In 2012, Gennaro earned an International MBA with emphasis on Corporate Finance and Business Strategy.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
