What the S-1 says about growth
Anthropic’s S-1 filing was the subject of this week’s Friday 15. Andy and Jared, with Brian out for the second week, went through the figures that stood out. First-half revenue was $16.3 billion, which Andy described as roughly 14 times the prior year. Against that, he cited about $80 billion in losses, which Jared characterized as largely an accounting figure. The filing also shows $131 billion in reserves, and a compute efficiency measure of 2.08, meaning each dollar of compute produced about two dollars of revenue.
Jared noted that he gets nervous whenever non-GAAP numbers appear next to GAAP financials, since companies often use them to cover something up. He read the efficiency number as an attempt to show a path to scale.
If you squint, you can see the path to how this grows to be a wildly profitable business at some point.
Jared Blank
How to read an S-1
Jared helped prepare S-1 filings early in his career and explained what the document is for. It is the disclosure companies file with the SEC and walk investors through, laying out the upside, the risks, and the challenges. The long risk factor section protects the company: if one of the listed risks occurs, an investor cannot claim the company failed to warn them. The filing also has to make the case for growth, which is why it repeats that adoption is at the very beginning of the curve.
Numbers that stood out
Andy highlighted several items from the filing. About 90 percent of contributions to Anthropic’s code base in September were written by Claude. Revenue per employee is $6.6 million, which he compared to $1.5 million to $2 million at Google and Microsoft, $3 million at Apple, and roughly $7 million to $7.5 million at NVIDIA. On market size, the filing cites a $40 trillion total addressable market and says Anthropic’s revenue is less than 1 percent of Fortune 500 budgets. Jared pointed out that when the denominator is human work, the market is very large.
Andy also noted an adjusted operating profit of $1.6 billion, with operating profit turning positive in the second quarter of 2026 at $974 million. Jared said that growing as fast as Anthropic is while approaching profitability is close to unheard of, because companies normally invest in compute and a sales force ahead of the revenue.
Uninsurable harm and “too big to fail”
One risk factor drew the most discussion. The filing says the scale of potential AI-related incidents could be large enough that the company could not obtain commercial insurance to cover the liability. The hosts called that a first. They compared it to the banks in 2008 being treated as too big to fail, and wondered whether the language signals that a government already taking stakes in companies would step in if things went wrong.
A B2B positioning
Andy found the areas the filing names for Anthropic’s impact the most interesting part, including biology and medicine, broad economic growth, and government services. None is a consumer business. He read this as positioning Anthropic as the global B2B enterprise AI leader, ahead of whatever OpenAI says in its own filing. The hosts compared OpenAI’s consumer path to Netscape in the 1990s, and said enterprise revenue is harder to win but more reliable, with less churn.
Both noted how fast leadership shifts. OpenAI led, then Claude pulled ahead for many users, and OpenAI has since improved its product. They saw the pace as driven by use cases more than technology: people find a tool that does something they did not know was possible, until a better one appears.
The compute overhang
Anthropic has $500 billion in non-cancellable compute contracts, out of $657 billion in infrastructure agreements, according to what Andy read. The first several risk factors in the filing concern the risk of not getting enough compute. Anthropic does not own its data centers and buys compute from companies that are also competitors, including Google, Amazon, and Meta. Andy said he would be less concerned about vertical integration, and that the filing’s message is that those agreements cover the company. Jared added that everything is cancelable in a bankruptcy.
What it means for B2B leaders
Andy offered several takeaways. First, avoid lock-in to a single AI vendor. A tool people cannot live without this week may be replaced next week, and model agnosticism, including open-weight models, preserves choice. Second, watch unit economics. Tokens behave like oil prices, and companies that told employees to spend freely have since received the bill. Third, the technology is not the main constraint. Gains will come when companies reimagine the work instead of swapping AI in for an existing task. Few companies can yet point to a transformed business, though everyone expects a major impact.
The hosts joked that Accenture, McKinsey, and Deloitte will gladly sell help with rethinking the work.
Will the product get worse?
Jared’s closing worry was about what happens after the IPO. Once investors and partners require profits, he said, products tend to get worse, as Google search did when it leveled off. He cited Cory Doctorow’s concept of enshittification and said he wonders how it plays out for AI. Andy named two counterexamples, Amazon and Elon Musk, who ignored Wall Street’s short-term demands. His view of where Anthropic and OpenAI are likely to land:
History suggests they are Salesforce and not SpaceX.
Andy
Andy also reminded listeners that registration is open for the Master B2B Mindshare Summit, March 16-17, 2027, at Convene in Chicago, at summit.masterb2b.com.