What a week this has been, and what a week we have ahead.

With Anthropic’s much-anticipated IPO all but certain to go live this Thursday, we can expect another eventful set of days. Ever since the first IPO rumors this summer, our team has been working on an Anthropic deep dive to cover it. I’ve been told we’re on track to get it live this Wednesday.

If you need something to read in anticipation, five of our deep dives explain where Anthropic’s money comes from and where it goes:

  1. Anthropic’s yearly revenue pace grew from $1 billion at the end of 2024 to about $81 billion at the end of October. Back in May, when it stood at $44 billion, A Primer on the Agentic AI Economy traced most of it to businesses paying for AI agents that write and ship code, and found that the lasting value will likely go to whoever owns the task an agent does, because that owner can switch models.

  2. Among the businesses whose card spending Ramp tracks, 1% pay about four-fifths of the money that goes to OpenAI and Anthropic. As we found in Why AI Is Booming, but Productivity Isn’t, most companies bolted AI onto old ways of working and have earned little back, so the number to watch is whether the other 99% start spending.

  3. In the first half of 2026, Anthropic kept about 65 cents of every dollar it charged for running its models, after losing 94 cents on each dollar in 2024, by SemiAnalysis’s estimate. Free models trailed the best ones by about four months in June, and The Open vs. Closed AI Race found what lets a company that sells only models keep charging more: its users and what it learns from their prompts.

  4. Anthropic has signed about $895 billion in commitments, almost all for computing power and data centers, including $44.6 billion with Nscale and $41.3 billion with Lambda, two companies that rent out AI chips. Those two renters are the subject of Anthropic Just Committed $80B in Five Days, which found that the ones that last deliver on time, and Lambda has already missed a lease deadline that lets either side cancel $34.9 billion of its deal, though Anthropic says it doesn’t plan to.

  5. In May, Anthropic agreed to pay SpaceX $1.25 billion a month to use its Colossus supercomputer, and either side can cancel on 90 days’ notice after an initial period. SpaceX chose the short term so it could take the computers back, as our SpaceX deep dive found in June, and if new data centers keep getting delayed, it could charge two to three times as much.

Caught My Eye…

1) Anthropic Just Crossed Its AI Threshold

On November 10, Anthropic told prospective shareholders that it has met the threshold in its safety policy for AI that speeds up AI research. Its automated R&D threshold can be met in one of two ways:

  • Substitution: models could fully replace the whole set of Research Scientists and Research Engineers, at competitive cost (within a factor of 5)

  • Acceleration: AI dramatically speeds up research, doubling the rate of progress in overall AI capability

The filing points to the second threshold being met. This would mean two years of AI progress in one, with the pace still rising. Anthropic warned that its models may advance faster than its safety measures, and said it is still evaluating the implications.

Five days earlier, METR, a nonprofit that studies advanced AI systems, published an assessment after 11 weeks inside Anthropic. Most of METR's staff on the project believe Anthropic is improving its models at around twice the rate it managed from March 2024 to October 2025, with an uncertainty range of 1.3 to 3 times. The same week, Anthropic released its latest model, Claude Fable 5.2, and took steps to limit its skill at the research tasks that could most directly accelerate AI development.

The AI Futures Project, the forecasters behind the AI 2027 scenario, responded to METR's estimate. Absent a deliberate slowdown, all of its authors now forecast a Superhuman AI Researcher (SAR) by November 2027: an AI that, with no human help, is better than the best human AI researcher.

By Nov 2027, a forecast has AI running 95% of Anthropic's model R&D

Another trend to keep an eye on is the Remote Labor Index (RLI). It covers a broad range of paid projects sourced from freelancing platforms. Within a year, AI's automation rate jumped from 3.8% to 41.3% across game development, product design, architecture, data analysis, video animation, and more. If Epoch AI's capability index keeps rising at its current pace, the RLI automation rate reaches 90% in late 2027.

If recent gains hold, late 2027 brings 90% on the Remote Labor Index

These findings tie into some of the insights we will cover in our Anthropic deep dive this Wednesday. Here's the chapter overview (subject to change before publishing):

Subscribe to keep reading

This content is free, but you must be subscribed to Social Capital Research to continue reading.

Already a subscriber?Sign in.Not now