Anthropic Prepares for IPO on SpaceX Scale; Citigroup Joins the List and the Prospectus Arrives in Days
Anthropic may file its S-1 by the end of August, with projected annual revenue nearing $65 billion and a valuation of $965 billion. Four banks are advising; the prospectus will list AI backlash as a risk.
Anthropic is days away from filing its public offering prospectus, according to Bloomberg reports from August 20. Citigroup has been added to the top of the advisory bank list alongside Morgan Stanley, Goldman Sachs, and JPMorgan. The company projects revenue between $190 billion and $200 billion by 2028, according to people close to the process cited by Reuters.
The financial context is well known. Anthropic raised $65 billion in a funding round in May, with a valuation of $965 billion. The annualized revenue from sales of access to Claude reached approximately $65 billion by the end of July, according to individuals familiar with the figures. If market conditions remain stable, this listing has the potential to be the largest tech IPO in history, surpassing the $86.2 billion mark set by SpaceX.
What the S-1 Will Force to Discuss
On August 21, CNBC reported that Anthropic intends to list AI backlash, or the widespread public rejection of AI, as a risk factor in the prospectus. This is the first time a current generation company formally acknowledges, in a binding document with the SEC, that public opinion on artificial intelligence is a material variable in its business model. Not about competition from OpenAI. Not about token pricing. But about the chance that society might decide, through regulatory or labor means, that the product should not exist at the projected scale.
This item forces a discussion that institutional investors have been avoiding. Buying Anthropic at the IPO is buying exposure to AI, but it is also buying long-term reputational risk in a sector where European regulators, American journalists, and labor organizations in various countries converge, for different reasons, on the same conclusion that the pace of adoption needs to slow down.
Who Buys $100 Billion in a Window
If Anthropic raises between $40 billion and $50 billion in primary shares, plus a significant secondary block, the addresses that can absorb this volume are few. American pension funds, particularly CalPERS and Texas Teachers, have signaled appetite for direct exposure to AI outside of hyperscalers. Sovereign funds, notably Norges Bank Investment Management, the Saudi Public Investment Fund, and Singapore's GIC, are on the natural list. European managers such as Amundi and Allianz Global Investors would allocate on behalf of clients already seeking exposure to AI without going through Nvidia.
The volume, in practice, is distributed across three geographies that require the same thesis: asymmetric return on a technology asset with defensible operating fundamentals. If Anthropic can demonstrate net retention above 130% on the enterprise API and low churn in Fortune 500 accounts, the thesis holds. If not, the banks must adjust the price mid-roadshow.
What the Thesis Against Anthropic Must Acknowledge
The strongest skepticism about AI today comes from Gary Marcus and analysts at Sequoia Capital, who have expressed concern over the gap between capex spending and revenue generated. The reading is that we are in an AI infrastructure bubble lacking equivalent return fundamentals. Applied to Anthropic, this thesis translates as follows: $190 billion in 2028 revenue is a projection, not history, and relies on a growth premise that no software company has sustained without breakdowns.
The counter-argument is not that Marcus is wrong. It is that Anthropic is not a startup burning VC funding like Character.ai was. It is a company already charging for enterprise tokens, with multi-year contracts in banks and consultancies, and with real annualized revenue, although projected to grow at an aggressive pace. The distinction that the market debate has yet to clarify is between AI labs that are classic VC unicorns and AI labs that are already effectively large-scale B2B providers. Anthropic falls into the second category; Character.ai did not. The same caution applies in reverse: not every projection of $200 billion in 2028 revenue is comparable.
There is also evidence that weakens the optimistic thesis. The figure of $65 billion in ARR at the end of July is the annualization of a current month, not revenue recognized in firm contracts. If the base of enterprise accounts begins to revise token consumption in quarterly cycles, ARR could contract before the prospectus even reaches investors.
What Will Define Pricing in the Roadshow
The discussion that will matter in September is not about multiples. It will be about the retention curve of the token and the share of revenue that comes from multi-year enterprise contracts versus variable revenue from consumption. What Anthropic decides to disclose in the prospectus will determine how much institutional funds are willing to pay. If the number is low, pricing is aggressive. If it is opaque, banks need to readjust. And if the market perception of AI shifts before the roadshow concludes, the difference between a historic listing and an IPO stalled mid-process may be a matter of two weeks.