Insider Brief
- Anthropic’s proposed IPO highlights rapid revenue growth and narrowing operating losses alongside substantial computing commitments and limited control for public shareholders.
- According to the story, first-half revenue rose roughly fourteenfold to $16.29 billion, while its $80.75 billion net loss largely reflected noncash valuation charges.
- The offering could influence valuations and infrastructure investment across the AI industry, with pricing, computing obligations and customer concentration remaining key uncertainties.
Anthropic PBC made its finances public for the first time on Oct. 26, filing a registration statement with the Securities and Exchange Commission that shows the maker of the Claude AI models grew revenue roughly fourteenfold in a year while carrying an accumulated deficit of $132.9 billion and a governance structure that leaves public shareholders with little say over how the company is run.
The San Francisco company plans to list Class A common stock on the Nasdaq Global Select Market under the ticker ANT, according to the filing. Goldman Sachs and Morgan Stanley lead a syndicate of 22 underwriters. The share count and price range were left blank in the initial filing, which is standard for a first public S-1.
The document arrives months after Anthropic disclosed a confidential draft filing in June, when it said a $65 billion Series H round had valued it at $965 billion, NPR reported. That valuation makes the offering one of the largest technology listings on record and a test of how public markets will price frontier AI developers, a group whose economics have until now been visible only through private funding rounds and secondhand reporting.
The S-1 supports two very different ways of looking at the company and it’s future — as well as the future of the AI industry. Supporters can point to revenue growth, improving unit economics and a balance sheet holding more than $130 billion in cash and short-term investments. Skeptics can point to multidecade compute commitments, heavy dependence on cloud providers that are also competitors, a novel safety-first governance model and a stated willingness to delay products for safety reasons.
What do the numbers in Anthropic’s S-1 show?
Anthropic reported revenue of $16.29 billion for the six months ended June 30, 2026, compared with $1.16 billion in the same period of 2025, according to the filing. Full-year revenue was $41 million in 2023, $386 million in 2024 and $4.59 billion in 2025.
Costs rose quickly but more slowly than sales. Compute and infrastructure expense, the company’s largest cost line, reached $8.95 billion in the first half, up from $2.33 billion a year earlier. Sales, marketing and partnership spending climbed to $3.17 billion from $120 million, and personnel costs rose to $3.15 billion, including $2.2 billion of stock-based compensation. The result was a first-half operating loss of $693 million, or an operating margin of negative 4%, compared with negative 184% a year earlier.
The headline net loss looks very different with Anthropic reporting a GAAP net loss of $80.75 billion for the first half, after recording a $72.39 billion charge for changes in the fair value of convertible notes, derivative liabilities and simple agreements for future equity, plus $8.76 billion of other expense. The filing attributes the fair-value charges primarily to the rising value of outstanding convertible notes as the company’s valuation climbed. These charges are noncash, and the company says the convertible notes held by Amazon Web Services will convert into Class N shares before the offering closes.
On the company’s own non-GAAP measures, which strip out certain items the filing reconciles in its MD&A section, Anthropic posted adjusted operating income of $1.62 billion and an adjusted operating margin of 10% in the first half. Adjusted net income was $2.67 billion.
The balance sheet held $131.65 billion in cash, cash equivalents and short-term investments as of June 30, according to the filing. The company says it has raised $141.1 billion in total since its founding in 2021. Other figures in the filing include more than 11,900 customers generating at least $100,000 in trailing 12-month revenue, more than 1,900 generating at least $1 million, about 250 generating at least $10 million, and net revenue retention of about 900% as of Sept. 30.
What is the bull case for Anthropic’s IPO?
The bull case begins with growth that has few precedents at this scale. Revenue for the three months ended June 30 grew 1,368% from a year earlier while compute and infrastructure expense grew 300%, according to the filing. If that gap holds, each additional dollar of revenue requires less compute than the last, which is the central question for any business built on renting or buying AI accelerators.
The margin trend supports that reading as operating margin improved from negative 1,344% in 2023 to negative 176% in 2025 and negative 4% in the first half of 2026. A company that reaches GAAP operating breakeven while still growing at triple-digit rates would sit in a small group of software businesses.
Customer behavior is another bullish sign. Net revenue retention of roughly 900% means that the qualifying customers present a year earlier were spending about nine times as much in the latest quarter, according to the company’s definition, which excludes fixed-price consumer subscriptions. The base of customers spending $1 million or more a year now exceeds 1,900. Anthropic also reports trailing revenue per employee of more than $6.6 million as of Sept. 30, compared with $1.4 million a year earlier, and says many aspects of its own research, training and testing now involve Claude.
The company is also selling a product whose price per unit of work is falling. The filing says Claude Opus 5.5 reduced customer cost per task by about 40% compared with Claude Opus 5, and that lower costs tend to increase usage. Bulls would argue that this is the same dynamic that expanded cloud computing, where falling unit prices grew total spending rather than shrinking it.
Finally, the balance sheet gives Anthropic room to keep investing without returning to private markets. More than $130 billion in cash and short-term investments, before IPO proceeds, is large relative to first-half operating expenses of about $17 billion. The company’s multicloud strategy, which spreads training and inference across Amazon’s Trainium chips, Google’s TPUs and Nvidia and AMD GPUs, is presented as a hedge against any single supplier’s pricing or capacity limits.
What is the bear case for Anthropic’s IPO?
The bear case starts with the size and duration of Anthropic’s spending obligations. The company says it has entered into compute and infrastructure purchase commitments and leases that run through 2045, with one portion expected before the end of 2030 and substantially all of the remainder between 2031 and 2045. The dollar totals were left blank in the initial filing, and investors will need those figures before they can judge whether current revenue growth can carry fixed commitments that extend almost two decades.
Growth itself is the second concern, according to the filing, which warns that the company is early in its revenue growth and that historical growth rates are not indicative of future performance. Net revenue retention near 900% is a measure of a market in rapid adoption, and it will decline as the customer base matures. The question for valuation is how quickly it falls and where it settles.
Another red flag concerns Anthropic’s dependance on a short list of partners that are also rivals. The risk factors state that a substantial portion of revenue and accounts receivable comes from a limited number of customers and distribution channels. The company buys most of its compute from cloud providers that develop competing models, and it acknowledges that it has sought more compute from those providers than they were willing to supply. Amazon is at once a compute supplier, a distribution channel and an investor whose convertible notes convert to equity at the offering.
The filing describes technical risks — perhaps an understatement — that are unusual in an IPO prospectus. Anthropic says it has experienced failed or underperforming training runs, that scaling relationships between compute and capability may plateau, and that its reliance on its own models to build successors could propagate errors across model generations. It also says it has delayed or limited model releases for safety reasons, citing its Mythos model and Project Glasswing, and that such measures may become more frequent and costly and could slow it relative to competitors that do not adopt similar constraints. There’s also some hints of global destruction that we should file under the mega-bear scenario.
Finally, legal and accounting items add noise. The 2025 results include a $1.5 billion charge for the Bartz copyright settlement. Stock-based compensation reached $2.2 billion in the first half, and the gap between GAAP losses and the company’s adjusted profit figures is large enough that investors will scrutinize which measure better reflects the business.
How does Anthropic’s governance structure affect public shareholders?
Anthropic’s capital structure separates economic ownership from control more sharply than most dual-class technology companies. Buyers of Class A shares will receive one vote per share, but the filing describes two bodies that together can elect the entire board.
The Long-Term Benefit Trust, a panel of five financially disinterested trustees, holds a single share of Class T stock that carries the right to elect and remove a majority of the board. A limited liability company owned by the seven co-founders will hold a single share of Class F stock carrying 50.1% of total voting power on matters put to shareholders generally, including election of the remaining directors. Neither special class has economic rights. Strategic business partners primarily hold nonvoting Class N stock, which converts to Class A when sold to an unaffiliated party, and employees other than the co-founders will receive Class E stock that votes only on limited matters and carries no economic rights.
As a Delaware public benefit corporation, Anthropic’s directors are required to balance shareholder financial interests against the interests of people affected by the company and its stated purpose, the responsible development of advanced AI for the long-term benefit of humanity. The filing acknowledges that this duty may conflict with short-, medium- or long-term financial performance. Anthropic will also qualify as a controlled company under Nasdaq rules, and says it will use that status to exempt its nominating and governance committee from full independence requirements.
The structure also carries a dilution mechanism that public investors will need to model. Anthropic has contributed shares equal to 5% of its fully diluted stock to the Anthropic Public Benefit Fund, a wholly owned subsidiary. Its board may issue up to an additional 2% of outstanding shares to the fund each time the share price rises 50% above the prior milestone, beginning at the IPO price. The company argues in the filing that its safety commitments carry particular weight with enterprise buyers choosing a vendor for critical workloads, which would make the governance structure a commercial asset. Critics will see a structure in which public holders supply capital but cannot replace the board.
What would the bull case mean for the AI industry?
If Anthropic trades well, the most immediate effect would be on the IPO pipeline. Wedbush Securities analysts described Anthropic’s June filing as a reopening of an IPO market that had been largely dormant for several years, with SpaceX and other large technology companies also moving toward listings, NPR reported. A successful ANT debut would give other frontier developers a public valuation benchmark and a path to liquidity for employees and early investors.
A strong reception would also validate the enterprise-first model. Anthropic’s disclosures describe a business built on metered API consumption, coding agents and workplace automation rather than consumer advertising. Public-market approval would likely push rivals to emphasize enterprise metrics such as net revenue retention and large-customer counts in their own disclosures.
The supply chain would feel the effect as well. Anthropic’s long-dated commitments flow to chipmakers, cloud providers, data center developers and power suppliers. Evidence that a model developer can approach operating breakeven while expanding compute spending would strengthen the argument that AI infrastructure investment is backed by end demand rather than speculation.
A successful offering could also lend credibility to safety-oriented governance. If investors accept a public benefit corporation with an independent trust controlling the board, other AI developers may face less pressure to choose between mission-based structures and access to public capital.
What would the bear case mean for the AI industry?
A weak debut or a sustained decline after listing would reach well beyond one company. Anthropic is among the few frontier developers whose audited financials are now public, and its numbers would become the reference point for how investors value every private AI lab. A discount to the $965 billion private valuation would likely pressure later-stage private rounds across the sector and make it harder for smaller developers to raise the capital that frontier training requires.
Infrastructure suppliers would face the most direct exposure. Chip and cloud companies have built capacity plans around multiyear commitments from model developers, and NPR reported that shares of chipmakers and other AI supply-chain companies had risen sharply as Anthropic and OpenAI expanded data center spending. If public investors conclude that developer revenue cannot keep pace with those commitments, the market could reprice the entire chain, from accelerator makers to data center landlords and power providers.
The bear case would also sharpen scrutiny of circular financing. Anthropic’s largest suppliers are also investors and distribution partners, a pattern also found in other AI partnerships. A poor reception could prompt investors and regulators to look more closely at arrangements in which cloud providers invest in model developers who then spend much of that capital on the same providers’ compute.
Finally, a failed listing could weaken the case for safety-first corporate design. If investors penalize Anthropic’s public benefit structure or its willingness to delay releases, competitors may conclude that such commitments carry a measurable cost of capital, which could reduce the incentive to adopt them.
What should investors watch next?
The amended S-1 filings that typically follow an initial registration will fill in the details now missing, including the price range, the number of shares offered, the dollar value of compute commitments and the size of the public benefit share pool. Those figures will determine the implied market value and how much of the company public buyers will own.
Beyond the offering terms, four disclosures will carry the most weight. The first is third-quarter revenue, which the filing references but leaves blank. The second is the trajectory of net revenue retention as the customer base grows. The third is the ratio of compute spending to revenue, which the bull case depends on. The fourth is customer and channel concentration, including how much revenue flows through the three major cloud platforms.
Investors will also watch how Anthropic reports under the scrutiny of quarterly earnings. The company has said it may delay or restrict models for safety reasons. How it communicates those decisions to public shareholders, and how the market responds the first time it does, will show whether a public benefit corporation controlled by a trust and its founders can operate as a frontier AI developer and a listed company at the same time.
This article is for informational purposes and is not investment advice.
