Anthropic’s S-1 Takeaways: From Giant Compute Commitments to Complex Lock-Ups

Anthropic publicly filed a preliminary prospectus for its planning initial public offering, providing a look inside its business for the first half of this year, as well as new details about its governance and more than 90 pages of risk factors.

The S-1’s filing on Monday sets the stage for Anthropic to start its investor road show in about two weeks, leading up to an IPO that Anthropic is expected to target for the week of Nov. 16, ahead of the Thanksgiving holiday. The stock is to be listed on the Nasdaq Stock Market under the ticker symbol ANT.

All told, the preliminary prospectus revealed the breadth of its compute commitments, its efforts to squeeze more revenue out of every dollar of spending and a still large concentration of customers—though the filing also left many specific figures blank, and didn’t include full financials for the third quarter

The IPO is expected to value Anthropic at $1.5 trillion or more, although the preliminary prospectus didn’t disclose data about expected pricing of Anthropic’s shares.

Here are the key takeaways from this document.

Fast Growth and an Inkling of Profitability

Anthropic generated $16.3 billion in revenue in the six months ending June 30, including $11.6 billion in the second quarter. That means the company generated 2.5 times as much revenue in the second quarter as it did all of last year.

Much of Anthropic’s business is concentrated in just a few customers: Its top five customers represented around 25% of its revenue for the nine months through September. No single customer represented more than 10% of its revenue, the company said, although in the first half of the year, one customer accounted for 11%. Concentration risk has eased somewhat: In 2025, two customers each represented 12% of its revenue.

It’s also reliant on partnerships with its cloud providers to sell its AI. Anthropic generated around $7.8 billion, or 48% of its revenue, through cloud providers in the first half of 2026. Anthropic counts resale of their models through cloud providers such as Google and Amazon as revenue, then treats the payouts to the cloud providers for the revenue share as sales and marketing expenses.

The company recorded a net loss of $80.7 billion in the six months ending June 30, far more than the $42 billion it lost in all of 2025. However, the vast majority of the losses came from accounting charges tied to the structure of some of the investment it has received. Backers including Amazon have invested through convertible notes that Anthropic records as liabilities, and it books a charge as the price of the shares tied to those notes increases along with jumps in its valuation.

Excluding those accounting charges, Anthropic’s net loss in the first half was about $8.4 billion, a bit more than the $7.7 billion total for all of 2025, despite a much larger jump in revenue, reflecting improvement in managing its margins.

Indeed, Anthropic recorded an operating profit of $974 million in the second quarter of this year, up from an operating loss of $1.7 billion in the first quarter.

Operating cash flow for the first half of this year was $3.9 billion, compared to negative $1.5 billion in the same period last year and negative $2.6 billion for all of 2025.

Improving Compute Efficiency

Hardware costs for running and training its models, along with compensation for its researchers, make up the lion’s share of its expenses.

Anthropic recorded $2.2 billion in stock-based compensation in the first six months of 2026, more than seven times higher than the same period a year ago. Spending on compute and infrastructure rose by 3.8 times, to $8.9 billion in the first half of this year.

Anthropic has managed to significantly improve its compute efficiency over time, generating more revenue per dollar of compute spend. The company’s spending on compute amounted to 48% of revenue by the end of the second quarter, down from 72% in the first and 251% for the first quarter of last year

Those improvements, as well as Anthropic’s aggressive fundraising efforts in the past year, have helped it build up a hefty cash coffer. As of June 30, Anthropic had $112 billion cash and cash equivalents.

The company listed current liabilities of $7.2 billion for that period, and $108.6 billion in convertible notes payable. It said it could raise more debt going forward to meet its financing needs. It has a revolving credit facility of up to $15 billion.

A Complicated Lock-Up Period

Anthropic is asking shareholders to agree to a staggered lock-up structure, similar to what SpaceX and Cerebras did in their IPOs this year—but significantly more complicated.

Some Anthropic insiders, excluding the co-founders, are allowed to sell up to 10% on the first trading days. They can also sell up to another 35% after full-year 2026 results. Their lock-ups end after Anthropic’s first quarter results in 2027 or 180 days, whichever comes first.

The prospectus also indicated that some shareholders would be able to sell at the IPO price, as The Information previously reported.

Meanwhile, Anthropic’s seven cofounders can begin selling up to 29% of their locked-up shares after the earlier of either 180 days after the IPO or the second full trading day following the release of first-quarter 2027 financial results.

Anthropic’s shareholders have agreed to a long lock-up period, meaning they can’t sell all their shares for up to 180 days or 365 days after the IPO. However, they are allowed to sell some shares in designated windows. Different shareholders agree to different provisions, and it isn’t clear who is allowed to sell at what time.

Over $650 Billion in Compute Deals

Anthropic spent $8.9 billion on cloud server rentals in the first half of this year. The S-1 filed Monday didn’t spell out the total amount of its planned spending on compute, but it laid out cloud computing agreements that put it on the hook to pay around $657 billion over the next seven to ten years, $417 billion of which is uncancellable.

Anthropic plans to spend $114.5 billion on Google cloud computing contracts, $110 billion with Amazon Web Services and $44.9 billion on Microsoft, an increase from the $31 billion commitment to Microsoft it had disclosed previously; none of those deals can be canceled unless the cloud providers breach their contracts.

And the S-1 outlined spending plans for compute from SpaceX that are larger than previously disclosed, saying it plans to spend $120.5 billion with the Elon Musk company by 2029, most of which is cancellable 90 days in advance. That will amount to about $3.1 billion a month once all capacity has been delivered.

Anthropic also disclosed its intention to spend $20 billion on a compute deal with AMD, $44.6 billion with Nscale and $41.3 billion with Lambda Labs, which are only cancellable if those firms fail to meet delivery deadlines.

And the company said it planned to lease Google chips from Broadcom totaling $161 billion over five years, which is non-cancellable. It also said it could spend an additional $125 billion to $150 billion to secure Google chips from other providers, a deal that’s still under negotiation, but noted that such spending is subject to availability and not included in its contractual commitments.

Based on all disclosures in the current version of this prospectus, Anthropic could spend up to $806 billion if it fulfills all of its compute commitments.

Anthropic said in the filing that to meet its demands for compute, it is shifting from relying on fixed cloud computing agreements to a hybrid model where it can also lease chips directly and put them into data centers dedicated to Anthropic’s usage.

Unusual Governance Structure to Include 5 Classes of Stock

The S-1 adds details about Anthropic’s unusual corporate governance structure as a public benefit corporation—which means its leaders can pursue societal goals in addition to trying to make money for shareholders—that has delegated significant power to its Long-Term Benefit Trust, a group of non-shareholders who advise the company on safety and social issues.

Anthropic will have five classes of common stock, the S-1 says:

• Class A: for regular shareholders, with 1 vote per share
• Class N: Primarily for strategic business partners, these have no voting rights but will “generally” convert to Class A stock upon sale or transfer to an unaffiliated third party. Anthropic is also using these to match its donation policy for employees.
• Class E: To be issued after the IPO to employees, excluding co-founders, these will be entitled to vote on certain limited matters and have no economic rights.
• Class F: A single share to be issued immediately before completing the IPO to Founder LLC, a limited liability company organized by its co-founders.
• Class T: Held by the Long-Term Benefit Trust, giving it authority to elect and remove a majority of Anthropic’s our board of directors. No general voting rights or economic rights.

Under the plan, Founder LLC will receive Class F stock that grants Anthropic’s seven co-founders a combined 50.1% of voting power over key corporate matters, as long as three of the seven co-founders retain a minimum number of shares in the company. The Information first reported on the arrangement in September.

The company disclosed two separate “sunset” provisions for its Class F and Class T shares. The trust’s Class T shares have no automatic expiration date, but can be eliminated with approval from 80% of the board and 80% of the trustees.

The Class T shares can also be eliminated even if the trustees object in certain cases. One of those cases is if there’s unanimous founder approval and majority approval from employees and public shareholders within three years after the IPO. The second case has no time limit and occurs if there is two-thirds founder approval, 75% approval from employees and majority approval from public shareholders.

The ability to remove Anthropic’s trust could raise alarm bells for safety advocates and others concerned about concentration of power. The members of the Long-Term Benefit Trust have no financial interest in Anthropic and are therefore meant to help the company make decisions that might go against its commercial interests. However, the company has previously signalled, even in its original blog post announcing the trust, that the structure of the trust could be amended in the future.

Meanwhile, the seven co-founders’ Class F share arrangement, which provides them with 50.1% of voting power, can be eliminated voluntarily with a supermajority of founders or automatically if more than four cofounders no longer are “continuing founders,” which happens in cases like a cofounder stepping down, dying or selling too much stock. When there are fewer than three remaining continuing founders for 90 consecutive days, a 450-day transition period begins, and at the end of that period, the special Class F shares are eliminated.

Ownership and Business Partner Relationships:

Google and Amazon are among the biggest shareholders in Anthropic, as the tech giants invested tens of billions into the Claude maker over the years and struck large cloud computing and chip deals with Anthropic. While the preliminary prospectus doesn’t disclose how much those corporate shareholders currently own, both have limits on how big their stakes can get: Amazon can’t hold more than 12% of Anthropic’s stock and Google cannot hold more than 14.99%.

Convertible notes that some investors own can transform into equity once Anthropic goes public, unless that would make the companies’ stakes exceed their ownership caps. In that case, the companies have to wait to convert the remainder of their notes if they wish to do so. Anthropic has $6.2 billion in outstanding notes held by Amazon. The preliminary S-1 doesn’t list notes outstanding to Google or another party. Such notes are just a portion of the company’s debt obligations.

The S-1 also sheds light on Amazon’s important mediating role in Anthropic’s relationship with the government. Government customers outside the U.S. and customers with workloads in classified environments access Claude models via Amazon Web Services’ Bedrock service.

Anthropic also disclosed that it pays a distribution fee to cloud providers that resell its AI models to customers; the company said such fees totaled roughly 17% of its revenue in the first half of 2026.

Anthropic’s seven co-founders, including siblings Dario Amodei, the company’s CEO, and Daniela Amodei, its president, each own roughly 2% of the company.

Dario and Daniela Amodei each received a salary of $654,167 last year, but with stock awards and other items, total compensation rose to just under $18 million for Dario and about $16.4 million for Daniela.

Spark Capital Added to Holdings

Few venture capital firms are mentioned in the filing. One exception is Spark Capital, which first invested in Anthropic in 2023 when it led the Series C funding round. The firm added to its stake in the AI company in every subsequent round. Spark has invested nearly $800 million across six funding rounds since 2023 according to the filing.

Most recently, Spark plunged almost $440 million into Anthropic’s Series H in June, snapping up roughly 747,000 more shares in the company.

Spark general partner Yasmin Razavi led the firm’s investments into Anthropic, and she is the only venture capitalist on the company’s board of directors.

Goldman Sachs to Lead the IPO

The filing mapped out the pecking order among more than 20 banks working on Anthropic’s IPO, which is expected to deliver windfalls in fees for Wall Street.

Goldman Sachs won the lead left role, putting it in the top position again after the bank won the top job for SpaceX. Morgan Stanley is also leading the IPO, and is named stabilization agent for the IPO, a coveted role overseeing the early hours of trading and making sure the stock gets off to a smooth start on public markets.

Anthropic is following SpaceX’s playbook by selling its shares internationally, including to investors in Australia, Canada, Europe, Japan, Switzerland and the United Kingdom. However, Anthropic didn’t disclose how much of the offering it’s allocating to small, individual investors.

Valida Pau is a reporter at The Information covering tech M&A and IPOs. Based in New York, she can be reached at [email protected], Signal @validapau.59, and on X @PauValida.

Stephanie Palazzolo is a reporter at The Information covering artificial intelligence. She previously worked at Business Insider covering AI and at Morgan Stanley as an investment banker. Based in New York, she can be reached at [email protected] or on Twitter at @steph_palazzolo.

Aaron Holmes is a reporter covering tech with a focus on enterprise and cybersecurity. You can reach him at [email protected] or on Signal at 706-347-1880.

Catherine Perloff is a reporter covering Amazon. You can reach her at [email protected] or on Signal at catherineperloff.05

Dakin Campbell is a reporter at The Information, where he covers the money behind the artificial intelligence boom, and other financial topics. He can be reached by email at [email protected] or on Signal at dakin.11.

Julia Hornstein is a venture capital reporter at The Information. Reach her at [email protected] or on Signal at juliah.22.