The Anthropic IPO filing puts a final price on its public offering, but the more revealing disclosures concern the infrastructure needed to build and serve Claude. The company priced 980 million Class A shares at $83 each and expects trading to begin on Nasdaq under the symbol ANT.
Update — November 19, 2026: Anthropic announced the final price after this article was first published. The company expects shares to begin trading November 19 and the offering to close November 20, subject to customary conditions. This update preserves the article’s original publication date.
Anthropic is selling 956,351,354 of the shares, while existing stockholders are selling 23,648,646. Underwriters have a 30-day option to buy up to 147 million additional shares. The SEC declared the registration statement effective on November 18; the company’s pricing announcement provides the final offering terms.
Beyond those conventional IPO terms, the filing describes a business whose growth depends on making enormous commitments for chips, data-center capacity and power years before the resulting compute is used. That makes the amended SEC registration statement useful evidence about the economics—and uncertainty—behind frontier AI.
What the Anthropic IPO filing says about the offering
At $83 per share, the base offering has a gross value of about $81.34 billion before underwriting discounts, commissions and expenses. That is not the same as the cash Anthropic will receive: selling stockholders receive the proceeds from their shares, and final net proceeds also depend on transaction costs and whether underwriters exercise their option.
Pricing completes an important step, but trading and closing remain separate events. Market prices can diverge from the offering price after shares begin trading, and the closing remains subject to customary conditions.
Anthropic’s 10 November roadshow announcement identifies Goldman Sachs and Morgan Stanley as sources for the preliminary prospectus. It also repeats the legal distinction between filing a registration statement and completing a sale.
Consumption drives most of Anthropic’s reported revenue
The filing divides revenue into consumption-based and subscription-based streams. Customers using the API or applications pay primarily for metered model usage, while subscriptions provide a specified level of access for a set period.
For the nine months ended 30 September 2026, Anthropic reports $26.3 billion in consumption-based revenue and $6.9 billion in subscription-based revenue. Consumption therefore represented 79% of the combined total reported for those two streams.
The company also uses a “Run-Rate Revenue” measure that annualizes a short recent period. Anthropic explicitly warns that this is not GAAP revenue or a forecast. Seasonal demand, price changes, lower-cost models and changing customer use can all make the annualized measure differ from revenue ultimately recognized.
That distinction matters when comparing a fast-growing AI company with a mature software business. A rate extrapolated from 28 days of consumption is not interchangeable with audited annual revenue, and rapid early adoption does not guarantee the same growth later.
What the Anthropic IPO filing reveals about compute commitments
Anthropic says it had entered purchase commitments and lease agreements with aggregate committed spending of about $900.9 billion as of and after 30 September. Roughly $451.6 billion is expected before the end of 2030, and about $70.7 billion is due during the following 12 months. Much of the rest extends from 2031 through 2045.
These are contractual commitments over many years, not one year’s operating expense. They also do not prove that the company will consume exactly the amount of compute it currently forecasts. The filing identifies both shortage and overcapacity as risks: underestimating demand can constrain products, while overestimating it can leave expensive capacity underused.
The company says compute is its largest operating cost. Historically, it relied heavily on cloud providers. It is now adding what it calls “Anthropic Direct”: directly leased chips hosted in dedicated facilities operated on its behalf. Anthropic argues that separating data-center capacity from accelerator purchases lets it secure long-lead-time sites before choosing all the hardware that will occupy them.
The structure also adds execution risk. Facilities, power, cooling, chips and financing must arrive on compatible schedules. Some infrastructure partners rely on third-party financing, and cloud providers may simultaneously be suppliers, investors, customers and competitors.
Future & Fact’s guide to AI infrastructure explains why the model, accelerator, data center, cooling system and electric grid form one supply chain rather than separate technology stories.
Why agentic workloads complicate the cost curve
Anthropic says multi-step agentic work can consume substantially more compute per user interaction than a single API call or chat response. Long reasoning chains, tool use and extended contexts may therefore make demand grow faster than the number of users.
Efficiency improvements can offset part of that demand. The filing emphasizes training methods, inference optimization and the ability to use several chip architectures. But efficiency does not automatically reduce total infrastructure spending if lower costs also stimulate more use or if newer workloads require longer runs.
This is why simple model-price comparisons are incomplete. A lower price per token can coexist with higher total compute consumption. Our explanation of how large language models work describes how tokens, context, tools and repeated inference shape the cost of an application.
How the Anthropic IPO filing describes governance
Anthropic is a Delaware public benefit corporation. Its stated purpose includes developing and maintaining advanced AI for humanity’s long-term benefit while balancing stockholder interests and the interests of people affected by the company’s conduct.
The filing also describes the Long-Term Benefit Trust, an independent-purpose trust designed to influence board composition. Under the disclosed structure, the trust generally has power to elect and remove a majority of authorized board members, subject to the detailed terms and transition arrangements in the prospectus.
That structure should not be summarized as either complete external control or ordinary one-share-one-vote governance. Investors and readers need to examine the voting classes, founder arrangements, trust powers and circumstances under which those powers operate.
What the Anthropic IPO filing establishes—and what remains unknown
Established: Anthropic has filed an amended registration statement, launched a roadshow and disclosed extensive revenue, compute and governance information.
Now established: the $83 offering price, the 980 million-share base offering, the planned ANT ticker and the expected November 19 trading debut. Not yet established: the market response, final net proceeds and whether underwriters will exercise their additional-share option. The company’s estimates for future demand, efficiency and infrastructure economics remain forward-looking.
The filing’s most important contribution is therefore not a single valuation headline. It is a detailed view of the physical and financial commitments required to compete at the frontier: years of data-center capacity, accelerator purchases, electricity-dependent infrastructure and capital raised before future demand is fully known.
This article explains public filings and is not investment advice.

