Anthropic generated around $16.3 billion in revenue in the six months ended 30 June 2026 and spent nearly $9 billion on computing capacity and infrastructure, according to a draft initial public offering prospectus viewed by The Wall Street Journal and reported on 20 October 2026. By our arithmetic on those figures, that is about $1.81 of revenue for every dollar of compute spending, or a little more, because the Journal says the spending was nearly, not exactly, $9 billion. Anthropic declined to comment to the Journal.
The same calculation on the year-earlier figures the Journal reported gives about $0.48 for the first half of 2025 and about $0.63 for all of 2025. Below is how each ratio is built, what it measures and what it leaves out.
Business Pill · RUN-RATE REVENUE
A one-minute explainer of run-rate revenue: what an AI company’s headline number does and does not mean. It teaches the general idea only and says nothing about any organisation in this story.
The key insight: As we read it, on our arithmetic from the Journal’s figures, Anthropic’s first-half revenue covered its compute spending about 1.8 times, against about 0.48 times a year earlier, while the commitments the Journal reported run years beyond the period.
All three ratios below are our calculation from figures the Journal reported: revenue divided by spending on compute, in the same period.
First half of 2026: around $16.3 billion in revenue divided by nearly $9 billion spent on computing capacity and infrastructure, about $1.81.
Full year 2025: $4.6 billion in revenue divided by $7.3 billion spent on computing capacity, about $0.63. A Financial Times account summarised on Yahoo Finance, which we covered on 15 October, gave the same roughly $0.63 figure for 2025.
First half of 2025: $1.1 billion in revenue divided by $2.3 billion, about $0.48. The Journal describes the 2025 half-year figure as spending “on infrastructure” and the 2026 figure as spending on “computing capacity and infrastructure,” so the two half-years may not be measured on exactly the same basis.
On those figures, revenue grew about 15 times between the two half-years and compute spending grew about four times, by our arithmetic.
Revenue per dollar of compute spending, our arithmetic on figures from Anthropic’s draft IPO prospectus as reported by The Wall Street Journal on 20 October 2026: about $1.81 in the first half of 2026 (around $16.3 billion in revenue / nearly $9 billion on computing capacity and infrastructure), about $0.63 for 2025 ($4.6 billion / $7.3 billion) and about $0.48 in the first half of 2025 ($1.1 billion / $2.3 billion, which the Journal describes as spending on infrastructure).
What the Ratio Measures
Revenue per compute dollar is a simple test of whether an AI company earns back what it spends on the chips and data centers that run its models. Below $1, each dollar of revenue costs more than a dollar of compute. Above $1, revenue covers the compute bill with something left over.
It is not a profit measure. It leaves out salaries, research, sales costs and the accounting charges in the prospectus. The Journal reported a first-half net loss of $80.7 billion, of which $72 billion resulted from accounting charges tied to changes in the value of convertible notes and other instruments.
It also measures spending in the period, not commitments. The Journal reported that Anthropic is on the hook for around $450 billion in cloud computing agreements over the next seven to ten years, the bulk of them noncancelable, and that its future infrastructure costs top $600 billion including a Broadcom chip lease.
A Second Measure, From Investors
A different figure came from Anthropic’s own presentation to prospective investors. The Financial Times reported, according to a BeInCrypto summary syndicated on Yahoo Finance, that Anthropic told investors it earned $1.70 in revenue for every $1 it spent on computing power in the third quarter, as we reported on 15 October.
That $1.70 is Anthropic’s own ratio for one quarter, as relayed in press accounts. The $1.81 is our arithmetic on half-year figures from the draft prospectus. The two may define compute spending differently, so we do not read the gap between them as a trend.
As we read it, the move from about $0.48 to about $1.81, by our arithmetic, came because revenue grew about 15 times while compute spending grew about four times between the two half-years the Journal reported.
As we read it, the ratio is a period measure: the around $450 billion in mostly noncancelable cloud agreements the Journal reported are future spending that this ratio does not capture.
As we read it, the two figures in circulation, the FT’s $1.70 for the third quarter and our $1.81 for the first half, may rest on different definitions of compute, which is why we do not read them as a trend.
Anthropic’s draft prospectus, as reported by The Wall Street Journal, 20 October 2026
“limited number of customers for the foreseeable future”
Three Implications
ABOUT $1.81 PER COMPUTE DOLLAR By our arithmetic, around $16.3 billion in first-half revenue divided by nearly $9 billion in compute spending is about $1.81, or a little more.
ABOUT $0.48 A YEAR EARLIER By our arithmetic, $1.1 billion in revenue divided by $2.3 billion spent on infrastructure in the first half of 2025 is about $0.48.
AROUND $450B COMMITTED The Journal reported that Anthropic is on the hook for around $450 billion in cloud computing agreements over the next seven to ten years, the bulk of them noncancelable.
The framework frames the economics this way: “Training was a cost center: episodic, concentrated among roughly twenty frontier model labs, amortizable across trillions of tokens. Inference is a revenue engine: continuous, distributed across millions of applications and agents, incurring marginal cost with every query.”
As we read it, revenue per compute dollar is the simplest way to watch that frame at one company: on our arithmetic from the Journal’s figures, Anthropic’s ratio moved from below $1 to about $1.81 between the first halves of 2025 and 2026. The framework was published in March 2026 and does not discuss Anthropic’s prospectus.
What Is Not Established
We have read the Journal’s article, not the draft prospectus, which is not public on the SEC’s EDGAR system as far as we could check. The ratios are ours. The Journal did not publish them, and the draft does not, as far as the Journal’s account shows, define a revenue-per-compute measure. A draft can change before a final filing.
Business Engineer Framework
The State of The Inference Economy
A Business Engineer map of the shift from training to inference: who makes money on AI compute, how, and where value concentrates.
By our arithmetic on figures The Wall Street Journal reported from Anthropic’s draft prospectus, Anthropic took in about $1.81 of revenue for each dollar of compute spending in the first half of 2026, against about $0.48 a year earlier and about $0.63 for all of 2025. As we read it, the ratio shows revenue now covering the compute bill in the period, while the long-dated commitments the Journal listed sit outside it.
A note on sourcing. This piece is built on The Wall Street Journal’s 20 October 2026 report on Anthropic’s draft IPO prospectus, read in full; we have not seen the draft itself. The revenue-per-compute ratios are our own arithmetic on the Journal’s figures. Earlier figures come from our own pieces citing Reuters and the Financial Times. The chart shows our arithmetic on figures the Journal reported. Nothing here is financial or investment advice.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
