Anthropic IPO filing shows $895B spending plan, warns of self-speeding AI

A week after Anthropic filed to go public, its amended prospectus fills in two things the first version left out: a commitment figure of about $895.4 billion for compute and infrastructure, and a warning that the company may soon reach its own threshold for AI that dramatically speeds up AI research. Anthropic filed Amendment No. 1 to its Form S-1 with the US Securities and Exchange Commission on November 2, 2026. The previous Unicorn API report, Anthropic files to go public, showing $16.3B in half-year revenue, covered the original October 26 filing.

The update also brings revenue forward to the end of September and adds a striking efficiency statistic: since April 2026, the value of Claude used internally, priced at what Anthropic charges customers, is worth more than its payroll. As before, this report rests on news coverage that quotes and summarises the filing, not on the filing itself.

What the amended S-1 changes from the October 26 version

A Form S-1 is the document a company files to sell shares to the public for the first time; each “S-1/A” amendment corrects or completes it before the offering starts. The first S-1, filed on October 26, confirmed the ticker ANT on the Nasdaq Global Select Market and Anthropic’s status as a public benefit corporation. That original filing also set out five share classes (Class A, N, T, F and E), with Class A carrying one vote per share, Class N carrying none, and Class T held by so-called T directors.

Comparing the two texts, coverage of the amendment counts four changes. The main ones, as reported:

Item	S-1, October 26	S-1/A, November 2
RSP threshold for AI-accelerated research	Not mentioned	Possible “imminently” with future models
Latest revenue	$16.3 billion, first half of 2026	$33.2 billion, first nine months; $17.0 billion in Q3
Purchase and lease commitments (compute and infrastructure)	Left blank	About $895.4 billion
Price per share	Blank	Still blank

Table data as reported by Pasquale Pillitteri’s comparison of the two filings.

The blank price is expected at this stage. An amendment of this kind normally updates the figures to the latest closed quarter and answers SEC comments, while the price range arrives in a later amendment close to the roadshow, the series of presentations to institutional investors before listing.

About $895.4 billion in compute and infrastructure commitments

Shacknews describes the new disclosure as $895.4 billion in committed spending on AI expenditures, calling it “a massive amount of purchase obligation” for one of the major players in the industry. Pillitteri’s breakdown classifies the figure as purchase and lease commitments covering compute and infrastructure, a line left blank in the October filing. TechCrunch sums up the financial picture in the amendment briefly: revenue is soaring, but so are expenses.

On the revenue side, the amendment reports $33.2 billion for the first nine months of 2026, with $17.0 billion in the third quarter alone. Set against the $16.3 billion reported for the first half in the original filing, that implies the third quarter brought in more than the whole first half.

The sources reviewed do not say over how many years the $895.4 billion falls due, which suppliers it covers, or how it splits between purchases and leases. Those details matter for judging how much of the figure is near-term cash and how much is long-dated obligation, and they are not in the coverage available.

The new warning: AI that speeds up AI research

The most discussed addition sits in the risk factors, the section where a company going public must list what could cost shareholders money. TechCrunch quotes the full passage:

In addition, our increasing use of AI systems to accelerate our own research and development may cause the capabilities of our models to advance more rapidly than our ability to develop, validate, and implement corresponding safety, alignment, and security measures. As a result, we have in the past not identified, and may in the future not identify, a significant acceleration in our rate of development until after it has occurred. For example, our internal estimates and our expectations regarding future models, rather than our current models, indicate that we may imminently meet our Responsible Scaling Policy’s definition of “dramatically accelerated progress attributable to AI.” We have not designated any of our current models as exceeding this threshold, and we continue to assess the underlying data.

Two points stand out. First, the warning concerns future models, not anything currently deployed; Anthropic says no current model has been designated as past the threshold. Second, the company admits that much of its evidence on the pace of acceleration comes from lagging indicators, so in the past it did not notice a significant acceleration until after it had happened, and that could recur.

TechCrunch frames this as Anthropic approaching recursive self-improvement (RSI), the point at which AI becomes good enough at improving itself to trigger a sharp acceleration in capabilities. The filing itself uses the term in a separate passage: “Using AI to help build more capable AI could also pose a considerable safety risk, particularly if it sharply accelerates the pace of progress. This is one of the areas covered by our Responsible Scaling Policy, which sets out our plans to monitor and mitigate the risks as we approach what has been termed ‘recursive self-improvement.’”

Anthropic also commits to keeping people in the loop. “We do not anticipate being able to safely use Claude to fully autonomously design and build successive generations of frontier models without meaningful human supervision or understanding, if and when that becomes technically possible,” the filing says, adding that the company is “considering updates to our Responsible Scaling Policy to keep humans in control of frontier model development at Anthropic.”

What the Responsible Scaling Policy threshold means

The Responsible Scaling Policy (RSP) is Anthropic’s internal rulebook linking model development to safety measures. It works through thresholds, each tied to a dangerous capability ranging from chemical and biological weapons to automated research; once a model crosses one, specific obligations apply before release. Version 3.0 took effect on February 24, 2026, as a full rewrite of the 2023 text.

The AI research and development threshold marks the capability level beyond which a model markedly accelerates overall progress in AI. In version 3 of the policy, the example given was an AI able to compress into a single year the progress that took two years between 2018 and 2024. Anthropic has revised the wording on this threshold more than once, with a version 3.1 dated April 2 clarifying what “doubling the pace of progress” means.

Claude is already doing much of the work

The filing gives context for why the company expects acceleration. Business Insider reports that in April, the value of Claude used internally, calculated at customer prices, exceeded Anthropic’s monthly employee compensation costs excluding stock awards, and that by September internal usage had grown more than 250-fold since January. More than 90% of the code added to Anthropic’s software in September was written by Claude, according to the same report. Business Insider stresses that this is not cash: Anthropic is not spending more money on AI than on salaries, but valuing its own consumption at retail prices. Pillitteri lists the payroll comparison as one of the amendment’s new figures, though Business Insider published its report on November 1, the day before the amendment was filed, citing the IPO filing.

Claims versus evidence: how to read a prospectus warning

Everything above is Anthropic’s own account, filed in a legal document. There is no independent evaluation in the sources of whether future Claude models will cross the threshold, and the company’s internal estimates are not published in the coverage reviewed.

TechCrunch notes that skeptics may see an S-1 as both disclosure and marketing, and that claims of nearing a model able to build itself could fall into the second category. Pillitteri offers the other reading: the sentence is in the risk factors, where companies have a legal incentive to list anything material that could hurt shareholders. Both readings can be true at once, and the filing gives investors and the public a dated, on-record statement against which later claims can be checked.

What it means for builders using Claude

For developers on the Claude API, the amendment is not a product change, but it touches several things that shape the platform:

• Compute capacity. Commitments of about $895.4 billion for compute and infrastructure signal how much capacity Anthropic is lining up. The sources do not tie the figure to any change in rate limits, availability or pricing.
• Release cadence. Under the RSP, crossing a threshold triggers specific obligations before a model ships. If a future model is designated past the AI research threshold, those obligations would apply to it. The sources do not say what they would be or whether they would delay launches.
• Policy changes ahead. Anthropic says it is considering RSP updates to keep humans in control of frontier model development. Any revision could affect how and when new models reach the API.
• Customer base. The original filing showed that companies spending at least $100,000 with Anthropic in the 12 months to September 30 included more than half of the Fortune Global 500 and all of the top 10 in Fortune’s global ranking, which indicates how much enterprise demand sits behind the spending plans.

What to watch next

• The price range. A further amendment should add the share price range ahead of the roadshow.
• The commitment schedule. Whether later filings or reporting break down the $895.4 billion by year and supplier.
• A threshold designation. Whether Anthropic formally designates a future model as meeting its definition of “dramatically accelerated progress attributable to AI,” and what obligations follow.
• RSP revisions. The human-control updates Anthropic says it is considering.
• Governance details. How the five share classes and voting rights settle in the final prospectus; earlier IPO plans called for seven co-founders to hold 50.1% of voting rights.