Anthropic's IPO Is Being Priced on 2028 Revenue

Anthropic, the maker of Claude and OpenAI's closest American rival, is preparing what could be one of the largest IPOs on record, and the bankers pricing it are not working from 2026 numbers. They are anchoring the valuation on a 2028 revenue projection of roughly $190 billion to $200 billion, two people familiar with the company's financials told Reuters. That figure is about four times the run rate Anthropic publicized as recently as May, near $47 billion, a gap that defines the bet investors are being asked to underwrite.

Pricing an IPO on revenue two years out is uncommon. Banks normally anchor software valuations to the current year or the next twelve months. High-growth names can justify forward multiples, but a 2-year-forward multiple is rarer, and the reason is straightforward: the further out the forecast, the more it depends on assumptions about cost, demand, and competitive position. Four sources told Reuters that bankers and investors are using enterprise value-to-revenue multiples on a 2028 base, the methodology they are settling on as the company approaches its analyst day. Anthropic declined to comment.

The second problem is the comp set. There is no mature, pure-play AI software company in the public markets yet, so the bankers are reaching for analogues. The comparables in play are Cloudflare, Palantir, and SpaceX, per Reuters sources briefed on the process. Cloudflare is an internet-infrastructure business built around hosting and security, not model serving. Palantir is a data-software operator that only recently turned consistently profitable and trades more on government and enterprise contracts than consumer AI. SpaceX is a private hardware company that priced its own tender offer this June on 2029 projections. None of them are a clean read-across for a frontier-model lab, an unprofitable company whose products are large language models sold to developers and enterprises, and that is still scaling training and inference spend.

If a mature native comp existed, the bankers would not be reaching across categories. Recent precedents confirm the pattern. Cerebras' backers cited 2028 revenue expectations in the runup to the AI chipmaker's 2026 listing, per the Reuters reporting. SpaceX priced on a 2029 view this June at a record valuation. The forward-year math is becoming the standard escape hatch when the present business does not yet justify the price, and it is most often used for AI-adjacent issuers where the public market has not yet caught up to the operating reality. For Anthropic, that means the comps list will be re-litigated with every quarterly print until at least one native anchor emerges, and a single serious quarterly miss could blow up the math before the IPO prices.

The 2028 number has to absorb something the 2026 number cannot: the cost of the compute, training runs, and hires needed to reach it. AI infrastructure spending has already rattled the most popular tech stocks in recent months, with the largest names pulling back on capex guidance as questions about returns mounted. Anthropic is not insulated from that pressure. Its margins are still being worked on by the cost of running frontier models, hiring top researchers, and absorbing the GPU bill that comes with training and serving at scale. The bet embedded in a 2-year-forward multiple is that revenue compounds faster than those costs, and that margins expand fast enough to justify the multiple when 2028 arrives. If that bet does not hold, the math stops working well before the IPO prices.

The next public anchor is the analyst day, where the company will publish the reference set in print. Until then, the comps list, the 2028 base, and the capex assumptions are the working draft of the math, and investors are being asked to sign off on all three at once.