Drinking Your Milkshake
Anthropic's 10x growth has to come from somewhere. It's coming from your budget.
I listened to Dario Amodei on Dwarkesh Patel’s podcast. He described Anthropic’s trajectory — zero to $30 billion in two years — and explained, calmly, that if growth drops to 5x the company goes bankrupt. He talked about revenue, compute, infrastructure. In over two hours, he never once mentioned the people whose budgets that revenue comes from. Daniel Plainview built the town a church too.
Drinking Your Milkshake
I listened to Dario Amodei’s interview with Dwarkesh Patel in February 2026 [1]. I expected a technology discussion. What I heard was an investor pitch.
Zero to $1 billion. $1 billion to $9 billion. $9 billion to $30 billion annualized by March. Trying for 20-30x growth per year. And then this: if growth comes in at 5x instead of 10x, “there’s no force on earth that could stop me from going bankrupt if I buy that much compute.”
He said this casually. The way a person describes weather. Then he moved on to how Anthropic is being careful — spending “somewhat less” than competitors, sizing compute to avoid ruin if timelines slip. The responsible one. The measured one.
What I did not hear, at any point in the conversation, was where the $30 billion comes from. Not who pays for the tokens. Who used to do the work the tokens replaced.
Whose Milkshake
I spent nine years in telecom enterprise software. In that world, when someone says “we grew 6% this year,” the first question is: what did the market grow? If the market grew 2%, your 6% means you took share from a competitor. Someone shrank so you could grow.
It’s simple math. If your growth exceeds the market’s growth, the difference came from someone else’s budget.
So: what market is Anthropic in? The standard answer is “the AI market.” But where does that market’s money come from? It comes from corporate budgets that used to pay for something else. R&D budgets. IT budgets. Labor budgets. Every API call, every Copilot seat, every Claude Code subscription is paid from a line item that previously paid a person, or a team, or a contractor.
In There Will Be Blood, Daniel Plainview explains how oil extraction works. “I drink your milkshake.” He doesn’t create oil. He drains oil that was already there, under someone else’s land, through a very long straw. He doesn’t hide it. He explains it calmly, to the person whose oil he is draining. They just don’t understand what he is telling them.
That is what I heard on the podcast.
The Math
There are roughly 28 to 37 million professional software developers in the world [2]. The global average salary is approximately $71,000 [3]. Total global developer compensation: roughly $2 to $2.5 trillion.
S&P 500 companies collectively spend roughly $500 to $600 billion per year on R&D [4].
Anthropic hit $30 billion in annualized revenue in March 2026 [5]. About 80% of that comes from enterprise accounts [8]. The other 20% is individuals — people paying $20 or $200 a month out of their own pockets, money that might otherwise go to Netflix. That 20% is not milkshake. It is people buying a tool they want. But the 80% — the $24 billion — comes from corporate budgets. And corporate budgets are labor budgets with a different name.
OpenAI is at roughly $25 billion [6]. Add Microsoft’s AI revenue target of $25 billion [7], Google Cloud’s AI-driven growth, and the dozens of smaller players, and the aggregate AI industry revenue is approaching $100 billion.
One hundred billion dollars is roughly 17-20% of total S&P 500 R&D spending. It is roughly 4-5% of total global developer compensation.
From companies that had zero revenue three years ago.
That money comes from headcount. When Block cut 4,000 people and its stock surged 24%, the stock market was telling you where the milkshake comes from. Not new products. Not new markets. Fewer people doing the same work. The savings flow partly to AI subscriptions and partly to margins. Both feed the straw.
The Ceiling
If Anthropic’s revenue hits $300 billion — which is what 10x from $30 billion implies — that would exceed the total R&D budget of every S&P 500 company combined. At that point, you are not taking share. You are the market. You have consumed the entire labor budget available to you.
Going from $3 billion to $30 billion means finding $27 billion in new budget. Going from $30 billion to $300 billion means finding $270 billion. That money has to come from somewhere, and there is a finite amount of labor budget on earth. The milkshake has a bottom.
Amodei knows this. That is why he talks about being careful with compute purchases. He is not worried about whether AI works. He is worried about whether the milkshake runs out before the infrastructure is paid for.
The Church
Amodei wrote a 30,000-word essay called “Machines of Loving Grace” about AI curing cancer, solving climate change, lifting all boats. I have no reason to doubt his sincerity. He probably believes every word.
Daniel Plainview built the town a church. He funded the school. He brought prosperity. He also drained every drop of oil under the land and left nothing behind. The church and the extraction were not contradictory. The church was what made the extraction possible. The townspeople leased the drilling rights because they were getting a church.
“Machines of Loving Grace” is the church. The $30 billion in revenue extracted from labor budgets is the drilling operation underneath. One does not disprove the other. But only one of them shows up on the balance sheet.
I’ve written before about what happens when companies pull the lever — when one AI-augmented engineer replaces a team of nine and the savings go to the employer, not the engineer. That article looked at the individual case. This one is about the aggregate: every company pulling that lever, the total savings flowing from labor budgets to AI companies, through a very long straw, into the milkshake cup of a man who explains exactly what he is doing in a calm voice on a podcast. And nobody in the room asks: whose milkshake is that?
The Part Where I Am the Straw
I use Claude every day. Last month I spent $33 and got 4,586 lines of production Go in under an hour. That work would have taken a developer a week. I know where that $33 came from — it came from the next developer who doesn’t get hired. I handed Anthropic the straw and pointed it at someone else’s cup.
The technology is real. The productivity gains are real. This is not an argument that AI is a scam.
This is an observation about a man who sat on a podcast, described a $30 billion revenue stream extracted from the labor budgets of knowledge workers, explained that his company will go bankrupt if the extraction slows down, and never once mentioned the people on the other end of the straw. Not because he was hiding it. Because it did not occur to him that it needed to be said.
Plainview did not think of the townspeople as victims. He thought of them as land that happened to have oil under it.
REFERENCES
[1] Amodei, D. (2026). Interview with Dwarkesh Patel. Dwarkesh Podcast.
[2] SlashData (2025). Global Developer Population Trends 2025. https://www.slashdata.co/post/global-developer-population-trends-2025-how-many-developers-are-there
[3] Gini Talent (2025). Global Software Engineer Salary Guide 2025. https://ginitalent.com/global-software-engineer-salary-guide-2025/
[4] Visual Capitalist (2025). The World’s Top Companies Fueling R&D Spending Growth. https://www.visualcapitalist.com/the-worlds-top-companies-fueling-rd-spending-growth/
[5] SaaStr (2026). Anthropic Just Passed OpenAI in Revenue. https://www.saastr.com/anthropic-just-passed-openai-in-revenue-while-spending-4x-less-to-train-their-models/
[6] PYMNTS (2026). OpenAI’s Annual Recurring Revenue Tripled to $20 Billion in 2025. https://www.pymnts.com/artificial-intelligence-2/2026/openais-annual-recurring-revenue-tripled-to-20-billion-in-2025/
[7] Campaign US (2026). Big Tech’s AI Spend in 2026: Following the Money. https://www.campaignlive.com/article/big-techs-ai-spend-2026-following-money/1949168
[8] Axios (2026). Anthropic Turns the Tables on OpenAI in Critical Revenue Category. https://www.axios.com/2026/03/18/ai-enterprise-revenue-anthropic-openai