Coding platforms swap default models, and lab revenue looks rentable

Raw lab revenue figures and compute deal lists are already covered, by Epoch AI for free, by Sacra and Dealroom as paid estimates, and by The Information for scoops and its data center database. A new entrant cannot win by just collecting run-rate numbers. What nobody appears to sell yet is the referee layer: a documented, auditable restatement of each lab's revenue on one common basis, meaning net of cloud resale with the hyperscaler revenue share backed out (for example Bank of America's estimate of up to $6.4B for Anthropic) and quarterly booked revenue rather than a single month annualized. That restated revenue would sit next to the lab's compute commitments as a coverage ratio, and the product would trace each normalized number back to source documents. The October 2026 Bloomberg story on gross versus net booking and OpenAI's memo arguing the gap is about $8B make this timely, and it would be most valuable in the window before S-1 filings force audited disclosure. Once the labs file S-1s, much of this advantage disappears, so the product needs to pivot afterward into comparing filings across labs. Concentration Watch is harder. The best existing signal, Ramp, misses large enterprises, and the Meta and Microsoft pullback surfaced through The Information's sourcing, not through data. A credible edge would mean stitching together hyperscaler marketplace data, job posting and tooling-adoption signals, and channel checks. Alternative-data vendors that sell to hedge funds could add this as a feature, and I did not check them in this research.

• Epoch AI – AI Companies Data Hub

  A free dataset under a CC-BY license, downloadable as CSV and updated daily. It tracks revenue, funding, compute spending and usage for frontier model companies, and is the closest thing to an open Lab Ledger. It doesn't appear to restate revenue on a gross versus net basis or tie revenue to individual compute contracts, and it isn't built as an investor product.

• Sacra

  A paid research service that sells revenue, valuation and growth estimates on private companies, including Anthropic and OpenAI, to investors. It already sells estimated revenue for private labs. This link is a university library mirror. I could not see its current figures or whether it normalizes the gross versus net question.

• Dealroom

  Publishes Anthropic versus OpenAI run-rate charts, for example $74.3B against $41.3B as of July 22, 2026. Its news notes also cover customer pullbacks, such as Microsoft cutting internal Claude spending. These are side features of a broad private-company data platform, not a normalized ledger.

• The Information (AI Data Center Database and lab finance reporting)

  Paywalled. It runs an AI data center database and breaks most of the lab revenue, compute deal and customer spending stories, including Anthropic's $517B in compute deals and the Meta and Microsoft Claude pullback scoop. In practice it is the early-warning signal that Concentration Watch wants to be, delivered as journalism rather than structured data.

• Ramp AI Index

  A monthly index of AI vendor spend built from card and bill-pay transactions at more than 70,000 US businesses. Investors already watch it as a spending signal for OpenAI and Anthropic. It skews toward startups and small and mid-sized companies, so it misses large enterprise contracts and spend that flows through cloud marketplaces.

• Presenc AI compute commitments tracker

  A vendor page listing multi-year compute commitments by lab, updated quarterly with the last update in June 2026. It overlaps with the compute side of Lab Ledger, but I could not verify its figures or methodology.

• Epistrophy Capital Research

  An independent newsletter that restates Anthropic revenue on a net basis. It is one of several analyst newsletters and blogs, alongside ValueAddVC and The Squid, doing the gross versus net normalization by hand, which suggests demand exists but nobody has productized it.

• Lambda Finance lab revenue dashboard

  A comparison page for OpenAI and Anthropic revenue whose figures sharply conflict with every other source. It is a weak competitor, but it shows that low-quality trackers are already showing up.

Checked 2026-10-11