Anthropic Kicks Off IPO Roadshow: Up to $64B at a $1.26–1.41T Valuation

Anthropic has officially launched its roadshow: the stage of an IPO when a company meets with investors and explains why they should buy its shares. For the industry, this is more than financial news. One of the leading developers of frontier models is showing the market for the first time what the business looks like when models get more expensive to train while tokens get cheaper to sell.

What the numbers say

According to Anthropic's announcement, the company is offering 825 million shares in a range of $70–78. The total offering will be $57.75–64.35 billion, and the company's valuation is roughly $1.26–1.41 trillion. The Nasdaq ticker is ANT.

The deal structure is as follows. Anthropic itself will receive about $56.1–62.5 billion before fees and expenses. Another $1.66–1.84 billion goes to existing shareholders, who are selling part of their holdings. In other words, this is mainly a capital raise for the company rather than an exit for early investors. The money is needed for compute and for training the next models.

The numbers are internally consistent: 825 million shares multiplied by $70 and $78 give exactly $57.75 and $64.35 billion.

Where the economics may not add up

The author of an analysis in the Telegram channel AI Product rightly notes that IPO money does not eliminate a structural problem. It has several layers.

Layer one. Models are getting more powerful, while the price per token is falling. Revenue per unit of compute is shrinking, and this has to be offset by volume.

Layer two. Volume is indeed growing, but costs are growing with it. Users run not one assistant, but dozens or hundreds of agents on a single task. Each of them consumes compute that has to be purchased and maintained. At the same time, money must be spent on training the next generation of models.

Layer three. Customers know how to count. According to the author, companies like Revolut are moving engineers to the open GLM model because it is cheaper. The market increasingly discusses that, with large token bills, it is more cost-effective to invest in your own hardware and your own model. This is so far an observation from the channel, not confirmed statistics, but the trend is clear to anyone who has looked at an API bill.

Layer four. Subscriptions to products such as coding assistants are partially subsidized. While growth continues, this pays off in market share gained. At a public company, such subsidies will be watched much more closely.

What this means for developers and product managers

An IPO at this valuation has practical consequences, even if you have no plans to buy ANT.

• Being public brings transparency. Financial reporting and details on margins, compute costs and revenue structure will most likely appear. This is a rare chance to see the real unit economics of a major model provider.
• Don't build your product on a single vendor. If your margin depends on one model from one provider, build in the ability to switch. An abstraction layer over the model and your own evaluation task sets make migration cheap.
• Count the cost of the task, not the token. If an agentic scenario makes hundreds of calls, the final price depends on the number of steps, not on the price per million tokens. For high-volume but simple tasks, open models may already be the more sensible choice.
• Keep a plan B on pricing. Public market pressure may change rates and limits. Subsidized plans are not forever.

Limitations

This material relies on a Telegram post and a link to Anthropic's announcement. The price range and offering size may change after the roadshow, and the source provides no details on financial metrics. The author's assessment — "an interesting show, but I wouldn't bet on it unconditionally" — is an opinion, not a forecast.

The main takeaway: the market gets a chance to assess how much leadership in models costs when tokens are getting cheaper and customers are counting their money ever more carefully. Building the best model is not enough. You also have to convince customers to pay specifically for it, and now this will be visible in public numbers.