Anthropic tests an unprecedented IPO and shakes up Nvidia
Anthropic is preparing an IPO with a target valuation of up to $1.4 trillion. At the same time, the company claims to reduce its dependence on Nvidia through a multi-chip strategy and the use of its own models to shift computing loads onto AMD.
Anthropic, the developer of the Claude assistant, is preparing an IPO that could value the company up to $1.4 trillion, according to documents cited by several sources. The San Francisco group is also targeting a total raise of up to $90.4 billion, including $26 billion via an anticipated private placement. At the same time, it claims to have reduced in a few days a porting to AMD chips that usually takes months, a sign that the AI battle is played as much in software as in silicon.
Une IPO gigantesque, mais sous forte contrainte de capital
The operation is expected to open around November 19 on Nasdaq under the symbol ANT, after a price setting expected on November 18, according to the amended filing. The announced range goes from $70 to $78 per share, for a total of 825 million Class A shares. Anthropic plans to sell about 801.4 million, while current shareholders would sell 23.6 million shares.
This stock window comes at a time when the company is pushing very far the idea of an enterprise AI capable of automating a large part of office work. Goldman Sachs estimates that Anthropic could reach about $685 billion in annual revenue by 2030, a jump of nearly 150 times compared to the $4.59 billion in sales recorded in 2025. According to the bank, enterprise clients would weigh $586 billion, or nearly 85% of the expected total.
The trajectory is spectacular, but it relies on an equally spectacular addition of costs. Goldman Sachs anticipates a return to positive annual free cash flow only in 2030. Until then, IT infrastructure expenses should approach $400 billion by the end of the decade. The kind of bill that reminds us that in AI, growth loves very high bandwidth.
Le vrai sujet : qui paie les puces, les data centers et les promesses ?
The regulatory documents give a rawer picture of the machine. For the nine months ended September 30, Anthropic reported $33.2 billion in revenue, including $17 billion in the third quarter alone, against $18.8 billion in computing and infrastructure costs. Over the same period, the company posted a net loss of $104 billion, mainly due to unrealized fair value adjustments on convertible bonds linked to its rising valuation [to be verified].
Anthropic also revealed about $900.9 billion in long-term commitments for leases and purchases, mostly for computing capacity. Among them, $451.6 billion would be due by the end of 2030. These amounts show how much cutting-edge models consume infrastructure even before generating a clear financial return.
Investors will have to decide between technological promise and economic discipline. On one side, the company wants to embody the next major generative AI platform for enterprises. On the other, it asks the market to finance a heavy industrial base, with costs rising almost as fast as ambitions.
Anthropic veut aussi desserrer l’étau Nvidia
On the chip front, Anthropic claims a multi-vendor strategy that aggregates AWS Trainium, Google TPUs, Nvidia GPUs, and AMD GPUs as an interchangeable compute pool. Tom Brown, co-founder and chief compute officer, explains that the company trains and runs Claude on the four major chip families, after betting early on this diversification. The idea is simple: better distribute the load, choose the best price-performance ratio, and reduce dependence on a single supplier.
The AMD case serves as a demonstration. According to Anthropic, its frontier model Mythos wrote or ported nearly 100% of the kernels during the adaptation to AMD GPUs in just three days, where it often takes six to twelve months with traditional methods. The figures come from an Anthropic investor presentation and have not been independently verified. If confirmed, they suddenly reduce one of Nvidia’s main competitive advantages: CUDA, its software environment that historically makes migrations long and costly.
The company complements this strategy with "compute multipliers," advances intended to increase the yield of each compute unit. It is also developing its Anthropic Direct program, which aims to rent chips and operate them in data centers designed for it, to reduce total cost of ownership. Even major chip suppliers now have a direct interest in the matter: Nvidia must participate in the conditional private placement of $26 billion, while AMD announced up to $5 billion in strategic investment and a deployment that could reach 2 gigawatts of Instinct MI455X GPUs starting in the first half of 2027.
Key points
- Target valuation: up to $1.4 trillion.
- Goldman Sachs targets $685 billion in revenue by 2030.
- Claude reportedly ported to AMD in three days, according to Anthropic.
- Nvidia participates in the conditional private placement of $26 billion.
- Nasdaq aims for listing under the symbol ANT on November 19.
By the numbers
- $90.4 billion — potential total amount of the operation, according to sources.
- $33.2 billion — revenue for nine months ended September 30, according to regulatory filing.
- $18.8 billion — computing and infrastructure costs over the same period.
- $900.9 billion — long-term lease and purchase commitments announced by Anthropic.
- 2 gigawatts — AMD capacity targeted via the program with AMD, starting in 2027.