Anthropic Files for IPO; Most Revenue Is Pay-Per-Use

Anthropic filed for a Nasdaq IPO on Oct. 26, and its prospectus says most revenue comes from metered usage fees, not fixed contracts.

If your store uses Claude for support replies, product copy, ad analysis or buying agents, the filing describes where your bill is heading: you pay for what the agents consume. That turns AI from a flat software subscription into a variable cost line, and it makes your choice of model closer to a supplier decision than a tool decision.

The numbers in the filing are enterprise scale, so your invoice is not the point. The direction of travel is, along with three risks the document spells out: capacity, vendor concentration and who is liable when an agent gets something wrong. At Ecommerce Paradise I track how store owners use these tools. My guide to building a leaner high-ticket business with AI covers putting Claude to work in a store. This post covers what the S-1 adds to that picture.

Agents that quote prices and promise delivery dates create exposure, and the cleanest way to fence it off is a properly formed LLC. Form your LLC with Northwest Registered Agent, the best LLC formation service for ecommerce owners. It has operated for 25 years, which is why I trust it over formation sites that change hands and pile on upsells. Form your LLC with Northwest →

Anthropic S-1: $16.3B H1 Revenue, 80% Pay-Per-Use

Anthropic, PBC filed a Form S-1 with the SEC on Oct. 26 for Class A stock to be listed on the Nasdaq Global Select Market under the ticker ANT, according to PPC Land. The price range and share count are blank. Goldman Sachs and Morgan Stanley lead a 22-bank syndicate, per BetaNews.

Revenue for the first half of 2026 was $16.3 billion, up from $1.2 billion in the first half of 2025, according to TechCrunch. For scale, PPC Land reports full-year revenue of $4.585 billion in 2025 and $386 million in 2024. BetaNews puts consumption-based fees at about 80% of first-half revenue, with much of it not tied to multiyear contracts.

The filing’s own risk factors, as quoted by PPC Land, say revenue is “primarily consumption-based, and much of it is not subject to long-term contractual commitments.” PPC Land adds that the filing says consumption fees rise with task volume and that agentic workloads use more compute per interaction.

The losses look enormous and mostly are not cash. Net loss for the half was $80.7 billion, of which about $72.4 billion came from non-cash fair-value changes on convertible notes, derivatives and SAFEs, per PPC Land. The operating loss was $693 million against $2.1 billion a year earlier, per TechCrunch. Yahoo Finance’s live coverage reports a $974 million operating profit in the second quarter after a $1.7 billion operating loss in the first.

On customers, TechCrunch reports that as of Sept. 30 more than 11,900 customers spend at least $100,000 a year, more than 1,900 spend $1 million or more, and about 250 spend $10 million or more. Net revenue retention was about 900% at the end of the third quarter. More than half of the $100,000 customers now use both the Claude Platform and a first-party app such as Claude Code or Cowork, up from about 30% at the end of last year.

BetaNews flags the risks that matter to buyers. Anthropic has requested more cloud capacity than providers were willing to supply, and its contracts do not guarantee delivery dates. The filing also says it is unclear when an agent’s conduct could bind the organization that deployed it, and whether contract liability caps would hold. A civil investigative demand from the FTC arrived in October, covering consumer protection, data protection, privacy, product safety and security.

PPC Land’s summary adds that concern about customer disruption “may lead us to limit or delay the deprecation of older models.” Reports cited by HNGN put the roadshow start at Nov. 9 and the listing in the week of Nov. 16. HNGN notes the filing itself does not confirm either date.

From Flat Seats to Metered Usage: The Billing Shift Behind the S-1

The S-1 did not introduce metered pricing. It documents it. Implicator.ai reported on April 14 that Anthropic restructured its enterprise plan so the seat fee covers platform access only, with usage billed separately at standard API rates. Per Implicator, Anthropic’s help documentation says the seat fee “doesn’t include any usage,” and organizations on older seat plans must migrate by their next renewal.

Tracy Woo, Principal Analyst at Forrester, laid out the before and after on Aug. 12. Premium seats had cost $200 per user per month and standard seats $40, with API discounts of roughly 10% to 15%. Under the new structure Claude.ai for nontechnical users is $10 per user per month and Claude Code is $20, with no API discounts.

Woo’s conclusion was blunt. Anthropic did not lower prices, but it also did not break from market norms, and Forrester sees OpenAI, Google and Cohere converging on the same consumption model. In her read, forecasting and governance of usage now fall mostly on the customer. Forrester’s clients reacted negatively, citing tighter usage limits and unclear communication.

There is a counterpoint. The seat prices Woo cites are cheap, and a business that uses AI lightly can come out ahead. None of the sources I reviewed says what small-store plans cost under the new structure, so check your own plan page before assuming either direction. I broke down one agent product aimed at small shops in my post on Claude Duet buying supplies via link and Shop Pay.

What Usage-Based AI Pricing Does to a High-Ticket Store

My read is that the cost itself is the smallest part of this. Here is hypothetical math, not reported figures. Take a store doing 400 orders a month at a $2,200 average order and a 22% gross margin. Gross profit is about $193,600 a month. An AI line of $1,500 a month is under 1% of that.

That is the nature of high-ticket dropshipping. Each order carries enough margin that a few cents per support ticket never shows up. What does show up is a runaway agent. A script that loops overnight, or a Q4 week when ticket volume triples, can turn a predictable line into a surprise, and consumption billing passes that straight to you. This is the exact scenario Forrester’s Woo warns about when she says successful adoption becomes a budgeting problem.

The second issue is capacity. BetaNews says Anthropic asked its cloud providers for more compute than they would supply. In practice that means rate limits and slow periods during peak demand, which for a store means Black Friday week. Implicator.ai’s April analysis put Claude API uptime at 98.95% over the 90 days ending April 8, against the 99.99% standard it attributes to established cloud providers. If an agent handles live chat, you need a human fallback for the days the model is slow.

The third issue is liability, and it deserves more attention than the first two. If the filing is unclear on when an agent binds the business that deployed it, assume the vendor will not make you whole after a bad quote, a refund the agent should not have issued or a delivery date it invented. Walmart is already drawing this line on its side of the counter. Its new agent terms require customer approval for agent purchases by default, as I covered in my post on Walmart opening to AI agents. Banks are doing the same, as I reported when Citi and BofA restricted AI agents.

Identity is the weak spot on every side. Arkose found that 97.7% of legitimate AI agent sessions lacked verified identity, which I wrote up in that study breakdown. A store that lets an agent act on customer accounts is trusting a system that cannot yet prove who it is.

Advertising is a smaller thread. PPC Land reports Anthropic publicly pledged in February 2026 that Claude would stay ad-free on every subscription tier, and that the filing summary it reviewed does not mention advertising. Google, by contrast, is putting AI review summaries inside ads. A pledge in a press release is not a contract term, so treat it as a current position and nothing more.

One more caution on the numbers. Sources disagree on how much compute Anthropic has committed to buy. PPC Land lists estimates ranging from about $245 billion in future minimum payments (Reuters, via Investing.com) to about $430.6 billion across four non-cancelable lines (Yahoo Finance), and the filing’s own total is left blank in the version PPC Land read. I am not building any conclusion on a specific figure. The relevant point is that the vendor carries very large fixed obligations while most of its revenue can leave without a long-term contract. Your lever is switching cost, so keep yours low.

I am not a financial advisor, and nothing here is a view on whether to buy ANT. Founders hold 50.1% of the votes through a single share class, per TechCrunch, which is a governance fact for investors to weigh. For a store owner the useful question is operational: how exposed is your business to one model, one invoice and one set of rate limits? If you would rather not wire the fallback logic yourself, that is part of what my team handles in the turnkey high-ticket store build.

Not sure which parts of your store should run on metered AI and which should stay human? Work through the cost and risk math with me one-on-one. Get the coaching details →

5 Moves for Your AI Budget Before the Reported Nov. 16 Listing Week

None of these depend on the IPO happening on schedule. They are the same moves I would make on any day a vendor shows you a metered bill.

Pull 90 days of AI spend and split it by workflow: support, copy, ads, operations. Put it on its own line in your books with Finaloop so it stops hiding inside card statements.
Cap every key and every agent that can loop. Set monthly spend limits and alerts, then run all AI tools through one business account such as Mercury so a single statement shows the total.
Test a second model on your top three prompts. Run them in ChatGPT, record quality and cost, and keep the prompts in your own document instead of a vendor’s interface. Put store facts in Shopify’s admin, where the new built-in knowledge base holds what your agents read.
Rewrite agent permissions. No refunds, price quotes or delivery promises without human approval, and a clean handoff to a person in Gorgias. Then run the nine checks in my Shopify AI-agent readiness audit to see what outside agents can do in your store.
Keep customer data in systems you own, such as Omnisend for email. Use HubSpot for your CRM. Then swapping a model never means re-exporting a customer list.

If you want a second set of eyes on the stack, book a discovery call with my team.

Automation tools meter work too. My breakdown of Zapier pricing and task limits shows how fast a busy workflow chews through a plan. My list of 34 dropshipping tools by job is a quick way to spot overlap before you pay for it twice.

Disclosure: I link to Claude through an affiliate link, which means I have a stake in how it is priced and run. That is why I am telling you to build a second option, not a dependency. You can compare Claude’s plans against your own usage and keep the alternatives in view.

Frequently Asked Questions

When will Anthropic list?
Reports cited by HNGN put the roadshow start at Nov. 9 and listing in the week of Nov. 16, but the filing leaves the price range and share count blank and does not confirm the dates.

Does the S-1 change what I pay for Claude today?
None of the coverage I reviewed reports a price change in the filing. Your plan page and invoice are what govern your cost, and Forrester’s Aug. 12 analysis explains how the metered structure works.

Is metered billing worse than a flat seat fee?
It depends on usage. Light users can pay less, while agents and scripts that run constantly can pay far more, which is the budgeting risk Forrester’s Tracy Woo describes.

Should I buy ANT stock?
I am not a financial or legal advisor and this post is not investment advice. TechCrunch reports the founders will hold 50.1% of the votes, so read the governance section of the S-1 yourself before forming a view.

Does AI cost matter much for a high-ticket store?
Usually less than margin, supplier terms and ad spend. If you are still deciding what to sell, start with the free niches list. Then use my launch guide for 2026 and add AI where it removes labor.

What if my AI vendor retires the model I rely on?
PPC Land’s summary says Anthropic may limit or delay deprecating older models over customer-disruption concerns, but that is a risk factor, not a promise. Keep prompts portable and test a replacement before you have to.

Who is liable when my agent makes a mistake?
BetaNews says the filing itself calls this unclear. Talk to a business attorney, keep agents away from refunds and binding promises, and operate through a properly formed entity. My comparison of the best virtual business addresses for an LLC covers the filing-address side.

Metered AI costs make your margin math matter more. Want my team to scale the store you already have? See the scaling service →

Pricing models for AI will keep moving, and the filing gives us the clearest look yet at how one major vendor earns its money. I will keep tracking what it means for store owners as the listing gets closer. Subscribe to the YouTube channel for daily breakdowns. More breaking news coming through the day.

If this was useful, these go deeper:

Trevor Fenner

Trevor Fenner is an ecommerce entrepreneur and world traveler from Seattle, Washington, and the founder of Ecommerce Paradise, a platform focused on helping entrepreneurs build and scale profitable high-ticket ecommerce and dropshipping businesses. With over a decade of hands-on experience, Trevor specializes in high-ticket dropshipping strategy, niche and product selection, supplier recruiting and onboarding, Google & Bing Shopping ads, ecommerce SEO, and systems-driven automation and scaling. Through Ecommerce Paradise, he provides free education via in-depth guides like How to Start High-Ticket Dropshipping, advanced training through the High-Ticket Dropshipping Masterclass, and fully done-for-you turnkey ecommerce services for entrepreneurs who want a faster, more hands-off path to growth. Trevor is known for emphasizing sustainable, real-world ecommerce models over hype-driven tactics, helping store owners build scalable, sellable, and location-independent brands.

He maintains homes in Seattle and Bali, spends most of the year overseas, and usually returns to Seattle for about a month each summer.