Anthropic’s vendor costs for data quality and performance evaluation services rose by $506 million year over year in the July-to-September quarter alone, about twice the $251 million rise across the whole first half, by our arithmetic from its amended Form S-1 filed with the SEC on 2 November and its first S-1 of 26 October.
Wing partner Tanay Jaipuria flagged the move on X late on 2 November, writing that it was “Interesting to see a big jump in data vendor spend this past quarter for Anthropic”. His chart estimates the spend at a run-rate of about $2.4 billion in September, up from about $900 million in June. Those levels are his estimates: the filing discloses only the year-over-year increases.
The line sits inside “Other operating and administrative” expenses, which the S-1/A says “primarily consist of charitable donations expense related to our donation matching program, vendor costs related to data quality and performance evaluation services, information technology and cloudsoftware costs”, among other items. The filing gives no standalone dollar figure for the data line and names no vendor.
For the nine months to 30 September, the S-1/A lists “a $757 million increase in vendor costs related to data quality and performance evaluation services” against the same period of 2025. The first S-1, covering the six months to 30 June, listed “a $251 million increase in vendor costs related to data quality and performance evaluation services”.
Subtracting one from the other leaves $506 million of the increase in the third quarter alone, by our arithmetic. For full years, the S-1/A reports “a $165 million increase” in the same vendor costs for 2024 over 2023, and again for 2025 over 2024. The third-quarter rise is about 1.5 times those two annual increases combined, by our arithmetic.
Year-over-year increase in Anthropic’s vendor costs related to data quality and performance evaluation services, as disclosed in its Form S-1 (26 October 2026: first half) and Form S-1/A (2 November 2026: nine months and full years). The third-quarter figure is the nine-month increase minus the six-month increase, our arithmetic. The filings do not disclose the cost levels.
Business Pill · Verification Cost
An evergreen Business Engineer explainer on why checking AI output costs more than making it. It explains the general idea only and does not discuss Anthropic or its filings.
The key insight: As we read the two filings, Anthropic’s spending on outside data quality and evaluation work accelerated faster than its compute bill in the third quarter: the year-over-year increase was $506 million in that quarter alone, about twice the first half’s $251 million, while the compute and infrastructure increase was about 1.2 times the first half’s, by our arithmetic.
What Was Posted on X
Interesting to see a big jump in data vendor spend this past quarter for Anthropic.
Much more in my Anthropic S-1 breakdown which goes out shortly.
Tanay Jaipuria, a partner at Wing who writes Tanay’s Newsletter, posted this on 2 November at 23:01 UTC with a chart titled ‘Anthropic data vendor spend run-rate’. The chart’s values are his estimates from the S-1/A (run-rate = last quarter times four). This piece checks them against the filing, which gives year-over-year increases for these vendor costs but not their dollar level: his shares of compute for December 2025, June 2026 and September 2026 are consistent with the filing’s compute and infrastructure line, while his 2023 and 2024 points cannot be checked against it.
Data Spend Against Compute
Compute and infrastructure, the largest cost line, also rose faster in the third quarter, but by less in relative terms. The quarterly table shows $9,863 million for the July-to-September quarter against $2,047 million a year earlier, an increase of $7,816 million, by our arithmetic. Across the first half the year-over-year increase was $6,611 million, so the third quarter’s compute increase was about 1.2 times the first half’s, against about 2 times for the data vendor line.
Within the wider line, the $757 million is about 16% of the $4.8 billion nine-month rise in other operating and administrative expenses, by our arithmetic; the filing attributes “a $2.1 billion increase in non-cash charges related to donations made under our donation matching program” to the same rise. Its quarterly commentary says “The overall increase reflected vendor costs related to data quality and performance evaluation services, information technology and cloudsoftware costs, payment processing and platform fees, and professional services costs as our business scaled”.
How Tanay Jaipuria’s Estimates Line Up
Jaipuria’s chart, titled “Anthropic data vendor spend run-rate”, shows about $40 million in December 2023, $240 million in December 2024, $530 million in December 2025, $900 million in June 2026 and $2.4 billion in September 2026. Its source line reads “Estimates from Anthropic S-1/A data quality & evaluation vendor costs. Run-rate = last quarter x 4”.
Each bar also carries a share of compute: about 6%, 7.5%, 4.5%, 4% and 6%. Applied to the filing’s compute and infrastructure line, those shares reproduce his run-rates: 6% of $9,863 million times four is about $2.37 billion, 4% of $5,558 million times four about $889 million, and 4.5% of $2,945 million times four about $530 million, by our arithmetic. His 2023 and 2024 points cannot be checked against the quarterly table, which starts in 2025.
On his estimates, the run-rate rose about 2.7 times from June to September, while compute and infrastructure rose about 1.8 times from the second quarter to the third, by our arithmetic. His full Anthropic S-1 Breakdown, published on his newsletter about half an hour after the post, is for subscribers; its description reads “Breaking down the growth, profitability, margins and compute needs and commitments behind the largest IPO ever”.
Who Gets Paid
The filing does not say which companies receive the money. Earlier reporting has linked two data suppliers to Anthropic. SiliconANGLE, citing Reuters in July 2025, said Surge AI’s “premium, high-end data labeling services” are “also used by Anthropic PBC”.
The Information reported in July 2026, citing internal documents, that Mercor “booked roughly $614 million in gross revenue in the first half of 2026”, with Anthropic among its named customers, according to a summary by AI Weekly. Neither report ties a dollar amount to Anthropic, and the S-1/A does not name Surge AI, Mercor or any other supplier.
As we read it, the disclosure is a line inside a line: the vendor costs sit within other operating and administrative expenses, and the filing reports only how much they rose, never how large they are. Tanay Jaipuria’s $2.4 billion run-rate is his estimate built on that disclosure, and his shares of compute reproduce from the filing’s compute and infrastructure line, by our arithmetic.
As we read it, the third-quarter jump came in the same quarter as a $4.3 billion step-up in compute and infrastructure expense from the second quarter ($5,558 million to $9,863 million in the quarterly table), so the data work and the compute behind it grew together, with the data line growing faster on Jaipuria’s estimates.
Anthropic PBC, Form S-1/A, 2 Nov 2026
“The overall increase reflected vendor costs related to data quality and performance evaluation services, information technology and cloudsoftware costs, payment processing and platform fees, and professional services costs as our business scaled”
Three Implications
FOR DATA AND EVALUATION SUPPLIERS The S-1/A puts a $757 million nine-month increase on this cost line, and earlier reports link Surge AI and Mercor to Anthropic; the filing itself names no supplier.
FOR READING ANTHROPIC’S COST BASE The vendor-cost increase is about 16% of the $4.8 billion nine-month rise in other operating and administrative expenses, by our arithmetic, next to a $2.1 billion increase in non-cash donation-matching charges in the same line.
FOR THE RUN-RATE FIGURE The $2.4 billion is Jaipuria’s estimate (last quarter times four), not a filing number; the third-quarter figure derived from the two filings is the $506 million year-over-year increase, by our arithmetic.
The framework reads: “Each moat is built by a different data pipeline.” One of its four is “The Verifier Moat — built by RLVR on domain-specific reward signals, building into a new asset class for reasoning-heavy verticals.”
As we read it, the S-1/A puts a price signal on that pipeline at Anthropic: the vendor costs for data quality and performance evaluation services rose $506 million year over year in one quarter, by our arithmetic. The framework was published in July 2026 and does not discuss the S-1/A.
The filing gives increases, not levels, so the size of Anthropic’s quarterly data vendor bill is not disclosed; the $2.4 billion run-rate is Jaipuria’s estimate. The third-quarter figure is our subtraction of a six-month increase in one filing from a nine-month increase in another. The filing does not split the cost between data quality and evaluation work, or say what it bought. We did not contact Anthropic or Jaipuria.
This Week In The AI Supercycle
Anthropic’s Reality Check, The Capital Squeeze, & The New World Order
This week’s issue: for two years the AI build-out was priced as a growthstory. This week it started being priced as a credit story.
Anthropic’s two IPO filings show its vendor costs for data quality and performance evaluation services rose by $506 million year over year in the third quarter, about twice the $251 million rise in the first half, by our arithmetic. Tanay Jaipuria estimates the spend at a $2.4 billion run-rate.
A note on sourcing. Anthropic figures and quotes are from its SEC Form S-1 of 26 October 2026 and Form S-1/A of 2 November 2026; interim figures are unaudited. Tanay Jaipuria’s run-rates and shares of compute are his estimates, read from the chart in his X post of 2 November; chart-read values are approximate. Supplier links are from SiliconANGLE (citing Reuters) and The Information (via AI Weekly). The third-quarter increase and all ratios are our arithmetic.
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.
