Anthropic has made public its IPO prospectus (S-1), to start the roadshow with investors on 9 November and go public in the week of 16 November. Highlights from the prospectus include:
(i) Strong revenue growth: $41M in 2023; $386M in 2024; $4.585B in 2025 and $16.286B in 1H26.
(ii) Still not truly profitable: operating loss in 1H26 of -$693M, although revenues are growing faster than costs. In Q2 26 revenues +1,368% vs. +300% spending on computing and infrastructure;
(iii) High customer concentration: two customers account for 20% of revenues and the top 5 for more than 25%. Some of them have no long-term contracts;
(iv) Strong commitments to spending on computing and infrastructure: $450B in cloud (with $120.5B with SpaceXAI; $110B with AWS; $110B with Google Cloud and $31B with Microsoft Azure), for the next 7-10 years, with 80% non-cancellable; $150B in chip rentals and $150B currently under negotiation, with cash of $131.651B.
(v) They expect to raise around $100B, which would value the company in a range of $1.5 to $2 trillion.
Opinion of the Bankinter analysis team:
Despite Anthropic presenting very high revenue growth and possessing unique technology, we believe that the profitability-risk ratio is not particularly attractive at the valuations being considered (~$1.5/2 trillion). The risks that concern us the most are:
(i) The hardware is still not ready for computing needs, making it very unprofitable to train and operate these models due to high costs;
(ii) High dependence on a few customers, who are also their main investors/creditors and from whom they also depend to obtain the computing for their models (Amazon, Microsoft and Google), with the risk that at some point they might prioritise their own models, restricting access or changing prices for Anthropic;
(iii) Very high investment commitments in computing capacity (~$800B), largely non-cancellable, which could pressure margins if revenue growth slows, something we saw in Q3 36 (+45% q/q vs. +146% in Q2 26);
(iv) Strong competition from open-source models, mainly from China, with similar capabilities and much cheaper, which is putting downward pressure on token prices and therefore the margins of these companies.
In conclusion, we believe that the valuations being considered (> $1.5 trillion) discount a scenario in which execution would have to be perfect, in a context where the risks are increasing. We think there is more value and less risk in the earlier stages of the AI chain, such as semiconductors, energy, or even in any of the hyperscalers. Furthermore, in these types of IPOs, history shows us that despite the FOMO of the early days, patient investors tend to fare better and therefore we believe it is better to wait for the initial volatility to pass and wait to know more information.
