Happy Monday.
The data this week: A company that runs more than 1,200 hobby forums just signed its first two deals with AI labs, and neither one is a flat fee. One lab pays for the training data it keeps, counted in tokens. The other pays each time the forums' posts help answer a question. Two days later, a new publisher collective signed Yahoo to a license that grows with Yahoo's AI revenue.
Underneath: Mercor raised $750 million at a $20 billion valuation, Mirendil raised a $1 billion Series A, Anthropic signed up seven tech companies to buy $26 billion of its stock at the IPO price, and SpaceX said it will spend more than $74 billion on two AI computing campuses.
The Lead
The big picture: For the past two years, licensing data to AI companies mostly meant a fixed check for access. Reuters reported in 2024 that Google pays Reddit about $60 million a year, a figure neither company has confirmed. Reddit's IPO filing put its first licensing contracts at $203 million in total, running two to three years. Those deals are running out. Reddit's latest quarterly filing shows $62.4 million left on its long-term contracts, mostly content licensing, with $30.5 million due in 2027 and nothing scheduled after that. Bloomberg reported last year that Reddit wants the next ones priced on how much AI answers rely on its posts.
Driving the news: On November 10, VerticalScope, a Toronto company that owns more than 1,200 enthusiast forums with over 2 billion posts, said it had signed deals with two frontier AI labs. It didn't name them or disclose terms. The two deals are priced differently. The first is a non-exclusive training license. VerticalScope delivered the data in early November, and it gets paid on the number of tokens (the word fragments models read) that the lab accepts. The second lab pays a variable fee, which the lab itself calculates, whenever VerticalScope content contributes to one of its AI answers. It started this month on a handful of forums. CEO Christopher Goodridge called it "real cash" on the earnings call, according to GuruFocus' summary.
On November 12, the Internet Licensing Collective, a nonprofit group of more than 3,000 publishers formerly called RSL Collective, signed Yahoo's AI search tool Scout as its first customer. Yahoo pays one blanket fee tied to its own revenue from the product, and the collective splits the money based on how each publisher's content gets used and whether it gets cited.
Why it matters: Pricing by usage means somebody has to count the usage, and in both VerticalScope deals the lab does the counting. "Accepted" tokens are whatever survives the lab's own filters, after it drops duplicates, spam and anything it doesn't want. The answer fees are, in VerticalScope's words, "determined by the counterparty." The forum owner can't see inside either number. I think that's the real shift in this market. A flat fee pays for permission. A meter pays for how much the buyer depends on you, and whoever controls the meter has the leverage.
VerticalScope needs the money. Third-quarter revenue fell 11% to $13.1 million, and monthly users dropped to 76.8 million from 83 million as search traffic slid. It counted 115 million attempts by AI systems to read its forums in 30 days, none of which brought ad revenue, and it is still suing OpenAI. Its CFO wouldn't size the deals but said the training license "could be a meaningful part of the quarterly results."
The rest of the market is building meters too. Cloudflare's pay-per-crawl, still in closed beta, charges AI crawlers a set price each time they fetch a page, starting at a tenth of a cent. In September, Cloudflare opened a beta that charges for each use instead of each visit. And as I wrote in October, about 100 publishers now get paid when Gemini and AI Overviews use their work.
What's next: VerticalScope expects the lab to settle the token count by year-end, and its CFO said data licensing will probably get its own line in the results. If it does, the fourth-quarter report will give us one of the first public prices for forum data sold by the token. The bigger test is Reddit. In July, the Wall Street Journal reported Reddit had discussed cutting off Google's AI access, and its latest filing still lists renewing its licensing deals as an open question. If Reddit gets paid per use, the flat-fee era is close to over.
The other path is to pool supply. The Korean Publishers Association did a version of that last week, which I covered Thursday: it now acts as the single channel for books going to five AI data firms. A collective like ILC gives publishers someone on their side of the table doing the counting. A single forum company negotiating alone has to take the lab's word for it.
Read VerticalScope's announcement.
See how the Internet Licensing Collective pays publishers.
Read Cloudflare's case for charging per use.
Data, Knowledge & Retrieval
Mercor raised a $750 million Series D at a $20 billion valuation, about double its October 2025 price, with Benchmark, Felicis, Nvidia and funds managed by Apollo affiliates among its investors, Reuters reported.
Databento raised a $100 million Series C at a $1.1 billion valuation led by Fifth Down Capital for its financial market data platform.
Models, Training & Developer Tools
Mirendil raised a $1 billion Series A co-led by Kleiner Perkins and Clearlake Capital, and says one of its researchers used its AI agents to build a base model in about 3.5 months for $10 million.
LangChain raised a $300 million Series C at a $3 billion valuation led by CapitalG for tools that run, test and improve AI agents.
OpenAI has raised $15 billion at a $1.25 trillion valuation, CNBC's David Faber reported, up from the roughly $11 billion reported a week earlier.
Anthropic agreed to sell $26 billion of stock at the IPO price to AMD, Broadcom, Google, Micron, Microsoft, Nvidia and TSMC, and plans to price on November 18 and start trading November 19, according to people familiar with the details.
Compute, Energy & Infrastructure
SpaceX said it will spend more than $74 billion on two AI computing campuses, up to $24 billion near Starbase in Texas and about $50 billion in Cochise County, Arizona.
Fractile raised a $720 million Series C backed by Thrive Capital, Founders Fund, Lightspeed and Redpoint for AI chips built to run models after they're trained.
Torus raised $500 million led by Magnetar for Foundry, prebuilt units that put 1 to 20 megawatts of AI computing at sites that already have grid power.
Enterprise Deployment & Distribution
Rox raised a $100 million Series B at a $1.3 billion valuation, co-led by General Catalyst and March Capital, for AI agents that run enterprise sales work.
Blue J raised a $176 million Series E led by JMI Equity for AI tax research software.
Physical AI & Robotics
Memories.ai launched Memories Upload, which pays gig workers to wear its camera while they cook, stock shelves and repair equipment, and says it recorded more than 1 million hours of video in its first month of beta.
Path Robotics raised a $300 million Series E at a $2.45 billion valuation led by Georgian for AI welding robots.
Health
Oura agreed to buy Counsel Health for $625 million, adding an AI primary care service backed by licensed doctors.
Charta Health raised a $55 million Series B led by Insight Partners for AI that checks medical charts before claims get billed.


