Part 1 | Anthropic Toward Wall Street: Valuation, IPO, and Demand

On 28 May 2026, Anthropic raised 65 billion dollars at a post-money valuation of 965 billion, while reported annualised revenue had exceeded $47 billion.3 Nine months earlier, at the Series F, the valuation stood at $183 billion and annualised revenue had only recently crossed $5 billion.1 The first figure rose 5.27 times; the second by more than 9.4 times. The crude ratio of valuation to annualised revenue consequently fell from under 36.6x to under 20.5x.

The figure alters the most immediate reading. The valuation has grown at an exceptional pace, but the numbers disclosed by the company paint a more nuanced picture than simple multiple expansion: revenue has advanced more rapidly than the implied value of the capital. The counterargument remains substantial. Anthropic's disclosure remains below the standard of a listed company: gross margin, inference cost, cash flows, net financial position and the economic rights attaching to the various share classes are less observable. The ratio should therefore be read as a rough estimate, distinct from a standardised EV/Sales multiple — that is, a multiple of enterprise value to revenue.

On the financial side, the key question is whether Anthropic's growth will prove sufficient to absorb over ten gigawatts of capacity and multi-year infrastructure commitments.

Valuation races ahead, multiple compresses

The sequence of the last three fundraising rounds shows a company accessing increasing amounts of capital while scaling its commercial operations.

At the Series G, annualised revenue stood at 14 billion, not $20 billion.2 Two months later it had exceeded $30 billion; by May, $47 billion.34 Claude Code also grew from over $500 million in annualised revenue in August 2025 to over $2.5 billion in February 2026.12 The proprietary nature of these figures distinguishes them from an accounting series subject to the audit of a public set of financial statements. The trajectory remains, however, too pronounced to be dismissed as mere financial enthusiasm.

In practical terms, the compression of the multiple signals that reported operational growth has outpaced the increase in price. The attractiveness of the valuation nonetheless requires further scrutiny of margins, cash and infrastructure obligations. At $965 billion, the capital embeds a near-ideal scenario: persistent demand, capacity available on schedule, a declining unit cost of inference and stable access to international markets.

From private valuation to the public-markets test

On 1 June 2026, Anthropic confidentially filed with the Securities and Exchange Commission a draft registration statement on Form S-1 for a possible initial public offering. The filing leaves the number of shares, the transaction price and the timing of the listing yet to be determined. The IPO will depend on the SEC's review, market conditions and the company's final decisions.17

The development is nonetheless significant. A valuation of $965 billion has thus far been established through private negotiations between the company and a select group of investors. A listing would transfer that valuation to a market in which the price is tested daily and in which Anthropic would be required to disclose a far greater volume of financial information.17

The prospectus could make observable such items as recognised revenues, gross margin, inference costs, stock-based compensation, customer concentration, cloud commitments, infrastructure agreements and capital requirements. It would also allow a comparison between the annualised revenues reported by the company and those recognised in individual accounting periods.17

The IPO would therefore represent the first genuine public test of the valuation. The market would need to assess both Claude's growth and the technological capabilities of its models, and how much economic value remains after payments for chips, energy, storage, data centres and multi-year contracts.17

Real demand, with different economic value depending on the form of adoption

Anthropic reported more than 300,000 enterprise customers in September 2025.1 In February 2026, more than 500 customers were spending at least one million dollars on an annualised basis; by April, that figure had risen to over 1,000.24 Growth is therefore extending from individual subscriptions to workflow-level deployments. Claude is being integrated into the processes of enterprises that are procuring capabilities through application programming interfaces, Claude Code, cloud services and office productivity products.

The external landscape confirms broad adoption, but calls for a distinction between breadth and depth. The Stanford AI Index 2026 finds that 88% of the organisations surveyed use artificial intelligence tools and that 70% deploy generative AI in at least one function. The use of agents — systems capable of executing sequences of tasks with greater autonomy — nonetheless remains in the single digits across most business functions.6

This gap lends itself to two interpretations. It points to a still-substantial pipeline, as many enterprises have made a start while few have yet extended agents to core processes. It also serves as a reminder that a pilot project still offers limited visibility into the depth of monetisation. Assessing the quality of revenues would require data on renewals, consumption per customer, concentration, contract duration and the economic returns realised by end users.

Coding today represents the most measurable use case. Code can be verified, tested and integrated into a production pipeline. This is a plausible reason why Claude Code has scaled more rapidly than less structured applications. The advantage, however, may narrow if model performance converges and price becomes the primary competitive lever.

More than ten gigawatts to secure capacity and time

The capital raised is intended above all to make available the resource that constrains growth: computing capacity. With Amazon, Anthropic has reserved up to 5 GW and has committed to spending more than $100 billion over ten years on AWS technologies. Nearly 1 GW across Trainium2 and Trainium3 is expected to be operational by the end of 2026.5 With Google and Broadcom, it has agreed to a further 5 GW of TPU-based capacity — the processing units designed for machine learning — from 2027 onwards.34

In the image below, we look inside Project Rainier, the infrastructure through which AWS makes Trainium chip-based capacity available to Anthropic.

In May 2026, Anthropic also secured access to the full capacity of SpaceX's Colossus 1 data centre: over 300 megawatts and more than 220,000 Nvidia GPUs. The agreement was subsequently extended to Colossus 2 as well, although the additional capacity directly comparable to the first facility has yet to be disclosed for this second infrastructure.18

Anthropic also utilises SpaceX GPU capacity. The strategy distributes workloads across Amazon's Trainium chips, Google's TPUs and Nvidia's GPUs. This diversification reduces the risk associated with reliance on a single architecture, while still leaving the company dependent on a handful of providers across capital, chips, cloud and distribution.

The commitment to AWS takes the form primarily of a multi-year infrastructure contract, with a more modest upfront capital expenditure component for Anthropic. The distinction matters: outlays will follow capacity availability and utilisation, yet the arrangement nevertheless raises the monetisation threshold. Should revenues grow faster than the cost of serving requests, Anthropic could achieve operating leverage. Should pricing decline, energy costs rise, or capacity come online ahead of demand, that same scale would compress margins.

It seems like the text got cut off. Could you please provide the complete Italian text you'd like me to translate?13 billionafter the new injection of 5 billion and has the option to add up to 20 more.5This aligns interests, but also conflates them. AWS funds Anthropic, sells the compute capacity that Claude requires, and distributes the model through Bedrock. The customer, the vendor, and the strategic shareholder are all party to the same economic relationship.

In the following chart, we observe the acceleration in global computing capacity dedicated to artificial intelligence between 2022 and 2025, with Nvidia still dominant and a growing presence of proprietary architectures developed by major cloud operators.

"Content (text and/or images) produced with the assistance of artificial intelligence, under the editorial responsibility of the newsroom."

The Financial Spectator - Independent analysis of markets and the global economy