Cat Wu, who previously ran Claude Code at Anthropic, left the company days ago without a stated reason. She has now posted that she is joining OpenAI to lead product for core products. On the same morning, Austin Liu, who helped build Anthropic’s growth team, also posted that he was joining OpenAI. The product lead on Anthropic’s hit coding tool and an early growth hire both moved to the rival in the same week that Anthropic is preparing to sell stock to the public.
For months, discussion centered on a $2 trillion valuation. Big investors are now pushing that figure down to $1.5 trillion. The roadshow can start as soon as next week, and trading could begin as early as November 16th. The amended filing still shows about $895 billion in compute commitments. At the same time, headwinds exist for data-center growth following the last election, and growth has slowed from 144% to 47% in one quarter. People are leaving for the rival, the valuation is being talked down in public, and the compute bill continues to consume more of every dollar than the company makes.
Third-Quarter Numbers and Compute Spending
In the third quarter, Anthropic made about $17 billion. Compute took 58% of that revenue, up from 48% the quarter before—a 10-percentage-point increase in the share of revenue going to compute. The company spent about $5 billion on short-term compute that costs more than longer-term deals. Revenue per compute fell from $2.08 to $1.72. These figures come from Anthropic’s own paperwork and show revenue moving in the wrong direction relative to compute costs. Memory costs have also risen, adding further pressure.
Large customers are already routing work away from expensive models. Examples include AT&T, Pinterest, Firetiger, and Pelosa, among others. Smaller shops have done the same. At the start of the year, one software team was spending nearly $2,000 a month on Anthropic Claude Code bills; that figure has since fallen to roughly $300–400 in recent months.
Questions About Growth Figures and the S-1
The company appears to have done everything possible to snapshot growth on the S-1 and speed-run the entire total addressable market in six months. Much of the reported revenue growth came from distillation. Anthropic admitted that huge amounts of traffic were being paid for by parties in China who were distilling the model. The company allowed this activity while running up to the IPO and then stopped it at a point that allowed the revenue numbers to be presented in a certain way. Distillation continues even after those attempts to halt it.
One view is that Anthropic knew the full picture before posting the original S-1 and left the Q3 data out in order to frame the conversation. There was one quarter that looked profitable; much of that appearance rested on careful accounting choices. Q3 already shows the company back in the negative. If overall revenue continues to decline, the next quarter is expected to move further into the red.
Compute Commitments Through 2045 and Calls for Scrutiny
Michael Burry has noted that growth dropped from 144% to 47% and that operating income flipped from roughly $1 billion positive to a negative half-billion in a single quarter. Anthropic has committed to computing spend through 2045 that totals a trillion dollars. With growth slowing and quarterly revenue around $17 billion, there is no clear path for the company to cover that bill. Public companies must eventually operate profitably rather than relying indefinitely on new investor capital. Even using the top-line numbers, profitability in the near term looks unattainable; the long-term commitment is roughly twelve times annual revenue, and further accounting adjustments could make the gap larger. Combined with data-center headwinds from the last election, the revenue picture remains under pressure.
Anthropic has projected approximately $32 billion in unrecognized stock-based compensation expenses—an unprecedented figure. Stanford research has linked the practice of deferring stock-based compensation until after an IPO with a high risk of stock-price decline relative to the market. One interpretation is that the owners recognize the IPO may not succeed and are using the public offering to capture as much capital as possible from current hype before exiting. Cat Wu may already have had some shares locked in. The Securities and Exchange Commission has been urged to block the debut until the company provides two consecutive quarters of audited financial statements that disclose all material information about spending commitments. The books themselves have been described as cooked, and additional concerns include security issues and lawsuits already filed over content theft, with more expected once more of the company’s practices become visible.
Talent, Competition, and the Absence of a Moat
Claude Code has been a major cornerstone of the recent hyper-growth. Losing the person who ran it is therefore significant. The departures of Cat Wu and Austin Liu cluster around the same morning that Wu announced the move to OpenAI.
Anthropic reached the market early and fast with Claude Code, yet the ability to replicate the tool is now viewed as fairly straightforward. A small team has already launched an open-source Claude Code competitor called open mono agent. The project is free, has nearly 2,000 stars, and continues to add features. It is estimated to sit within 5–10% of Claude Code’s accuracy and skill and is said to outperform the version of Claude Code that existed a year ago. When a team of that size can produce such a result, the argument is that Anthropic no longer possesses a durable moat. Open-weight models are also described as close in capability. Quinn 4 is expected this week and is anticipated to be strong; the Quinn 3.8 Flash Next 180-billion-parameter model is already close to the behavior of paid frontier offerings. Frontier labs are said to be concerned, and Anthropic’s public offering is characterized as a cash grab timed to reward senior holders before the stock faces further pressure.
Broader View
The same assessment extends to OpenAI, which is viewed as struggling financially as well. Both companies are described as money-burning operations that risk eventual collapse. The concern is that an Anthropic IPO could pull other investors’ capital down with it. The speaker, who has spent twenty-five years in software development and a decade in executive technology leadership, builds custom software solutions and open-source tools that let companies run their own AI stacks rather than renting intelligence from frontier labs. He does not consider himself a stock analyst and is not offering financial or legal advice; the comments reflect observations drawn from the technology itself and from the filings and public statements available so far.
Whether the SEC will require fuller disclosure before any listing, how much further revenue will decline once distillation effects fade, and whether the compute commitments can ever be met under current growth rates remain open questions. The combination of executive departures, a reduced valuation discussion, rising compute intensity, and long-dated spending obligations has drawn attention precisely because Anthropic is seeking to go public in the coming weeks.
