SpaceX Just Paid $60 Billion for the Layer Everyone Call(ed) a Commodity
The week the application layer won — and what a $1.75 trillion rocket company taught me about net worth, founder debt, and the Knicks.
There’s a race to the bottom happening with AI models. The frontier ones are still expensive, but more and more it’s the application layer — the applied AI sitting on top — that’s becoming the thing worth paying for, not the models underneath.
Take Rogo AI. It’s now valued at over $2 billion (a fresh $175M round), and it’s used by the banking and financial-services teams at Goldman, JPM, BofA, Wells, and Barclays — 35,000 users and growing. Here’s the simple math: Rogo has no unique advantage at the model layer. It plugs into OpenAI, Anthropic, and everyone else — and so does every one of its customers. The entire value lives in the application: the workflow, the trust, and the ability to close a Tier-1 bank in months instead of years, then expand seat by seat inside each account.
That’s the future I see. Companies like Rogo become high-margin businesses on the back of their application — not on the compute they’re subsidizing as arbitrage for their clients. The model is a commodity. The product is the moat. They have to keep becoming a better application, or the whole thing collapses into the price of tokens.
SpaceX Just Bought Its Way Into the Application Layer
If you wanted proof the value is moving up the stack, look at what happened this week: SpaceX is acquiring Cursor for $60 billion in an all-stock deal — four days after the largest IPO in history. SpaceX (now merged with xAI) didn’t need more compute; it needed a product. Cursor — by most accounts the third-best coding tool behind Claude Code and Codex — instantly makes Grok a real player in the one place it was losing: the application layer where developers actually live.
You don’t spend $60 billion on the layer you think is a commodity. You spend it on the layer you think is the prize.
Uber Burned Its Entire 2026 AI Budget by June
Meanwhile, everyone else is learning the cost of the model layer the hard way. Companies are burning through AI credits at record speed — Uber torched its whole 2026 AI budget by June and is now rationing tokens across teams. I think model prices fall over the long run, but in the short term they’re climbing as “tokenmaxxing” loses its shine.
This feels like the right correction. Burning tokens for the sake of burning tokens was never the point — the productivity has to be there. And it’s already minting a new product category: companies like Ramp now sell AI cost-monitoring tools to track the spend. The picks-and-shovels of the token economy.
The Fintech Tell: Ramp at $44 Billion
Speaking of Ramp — you’d almost have missed it in the noise, but they raised another $750 million at a $44 billion valuation last week, on north of $1 billion in annualized revenue and ~170% purchase-volume growth. I’m bullish on Ramp as a 2027 IPO candidate, and possibly still undervalued in a public market that’s been starved for a fintech that actually delivers.
Here’s what I can’t stop thinking about: Ramp is now worth roughly eight times Mercury (~$5.2B) — and Brex just got acquired by Capital One for $5.15B. Same category, same era, wildly different outcomes. What did Ramp get right that the other two missed? That’s the real question, and I don’t think the answer is “spend.” I think it’s the same lesson as Rogo: the company that wins is the one that turned its product into the thing customers can’t live without.
What Is Net Worth, Anyway?
Then there’s the rocket. SpaceX went public this week at roughly $1.75 trillion — the biggest IPO ever, raising about $75 billion, instantly one of the most valuable companies on Earth on less than $20 billion of revenue. Elon Musk appears to have become the world’s first trillionaire in a matter of days, his net worth ballooning by billions with every dollar the stock ticks up.
It feels like Musk has turned net worth itself into a meme — a number untethered from anything I can explain to my kids. How do you teach a child about money in a world where a company doing under $20 billion in revenue is worth $1.75 trillion and climbing? No one man should have all that power… and yet here we are.
Founder Debt
Here’s the part nobody says out loud at the SF dinner parties. Yes, everyone seems to be getting rich, and we just minted our first trillionaire. But most founders go into massive debt — financial, emotional, reputational — to build anything at all. Musk is no different; he just operates at a scale where the debt is invisible until it isn’t.
The risk it takes to succeed as a founder has never felt more real. As capital swings to fewer and fewer companies, starting something new gets riskier, not safer — the outcomes are violently skewed. To even step onto the field, you have to accept a baseline of risk most rational people won’t. And then you have to do the genuinely irrational part: believe your outcome will be 10 to 1,000x bigger than any sober analysis would allow. Delusion isn’t a bug in founding. It’s priced in.
Knicks in 6 Would’ve Been Better
I’ll end somewhere unprofessional. This is the worst day of my year — the NBA season ended Saturday, and I had the bad luck of not being in New York for it. Jalen Brunson went full Curry-meets-Jordan to close it out with 45. My real gripe is the Spurs, who led for 72% of the series and still couldn’t force a Game 6 or 7.
I miss the Warriors run as an SF resident, and I respect that we’ve had eight different champions in eight years. But if San Antonio makes a few moves this summer, they could be building the next dynasty. That’s the thing about betting on the future — in basketball and in markets, the boring, fundamentals answer is usually the one that ages best.
What I’m reading this week
The High Costs of the IRS’s Centralized Authorization File System — the unglamorous government plumbing that decides whether small-business loans close. (More on this soon.)
Ben Thompson, Anthropic’s Safety Superpower (Stratechery)
Satya Nadella, A frontier without an ecosystem is not stable — the application-layer argument, from the guy running one.
Tyler Cowen, Sometimes it is hard to solve for the equilibrium (Marginal Revolution)
Anthropic disables Fable and Mythos models after the U.S. barred foreign access (Fortune) — the AI-export story is just getting started.
See you Thursday.
— Matt