Anthropic's October Nasdaq Plan: The Quieter IPO That Could Reset AI Pricing

Anthropic filed confidentially for a listing on 1 June 2026 and has been scheduling investor meetings for an October debut on Nasdaq. Its reported run rate reached about $47 billion in late May, up from $9 billion at the end of 2025. The number that will decide how it prices is not that one. It is the roughly 85% of revenue that comes from enterprise and developer customers.

What has been filed and what is scheduled

Anthropic submitted a confidential S-1 on 1 June 2026 and was subsequently reported to be scheduling investor meetings from July for an October Nasdaq listing. As with any confidential filing, the financials remain private until the company elects to proceed publicly.

The sequencing matters. Anthropic filed a week before OpenAI, which means both companies entered the review process within days of each other and will be priced against one another whether or not they list in the same window.

Why October, specifically

Autumn windows are conventional for listings because they follow a completed half year of audited figures and precede the year-end slowdown. A company targeting October is working to a normal calendar rather than reacting to conditions.

That is itself a signal. A business rushing to list ahead of deteriorating conditions does not schedule investor meetings three months out. It moves when the window opens.

What a confidential filing does not tell you

It carries no disclosure obligation. Nothing in the financials becomes public until the company elects to proceed, which means every figure discussed publicly today is a company-supplied number or a report of one.

That is not a criticism of the company. It is a limitation on what anyone outside it can currently know, and it is worth holding in mind against the confidence with which these numbers are usually quoted.

The revenue ramp, plotted honestly

The growth curve here is the reason the listing is being watched. Each monthly snapshot has been stale within weeks of publication.

A run rate annualises a recent period rather than reporting a completed year. It is a legitimate metric and it is not revenue. Treat every point on that chart as a snapshot with a date attached.

The customer data moves in the same direction. At the February 2026 Series G, more than 500 business customers were each spending over $1 million annually. By April that figure exceeded 1,000, roughly doubling in under two months.

Why the revenue mix matters more than the run rate

Public markets do not price all revenue equally. Enterprise contracts carry longer commitments, higher switching costs and more predictable renewal behaviour than consumer subscriptions, so the same dollar of revenue supports a higher multiple.

This is also the clearest structural difference between the two large labs. One built a consumer product that became an enterprise business. The other built an enterprise business that happens to have a consumer product.

Why enterprise revenue prices higher

Three properties do the work. Contract length, which makes revenue predictable across quarters. Switching cost, which makes churn low once a model is embedded in a workflow. And expansion behaviour, where an existing customer grows spend without new acquisition cost.

Consumer subscription revenue has none of those reliably. It churns monthly, it is sensitive to price, and it requires continuous marketing spend to replace losses. The same revenue dollar therefore supports a materially different multiple.

The customer concentration data reinforces the point. More than 1,000 accounts each spending over $1 million annually is a very different revenue base from millions of individual subscriptions, and it will be disclosed as a risk factor as well as a strength.

The gross basis problem

Anthropic reports revenue on a gross basis. That means total end-customer spend flowing through cloud resellers such as AWS, Google Cloud and Microsoft Azure is counted as revenue, with partner payouts booked as an expense.

Reporting closer to net, as peers do, would produce a lower number for the same underlying business. Neither approach is wrong. They are simply not comparable, and almost every published comparison ignores this.

When the public S-1 appears, the accounting policy note will be the most valuable page in it. Read that before the growth chart.

What an enterprise-heavy listing means for buyers

Two consequences, one immediate and one slower.

The immediate one is scrutiny of concentration. A public company discloses customer concentration risk. If a small number of large accounts drive a meaningful share of revenue, that becomes visible, and it changes your negotiating position if you are one of them.

The slower one is margin pressure. Public markets reward gross margin improvement, and in AI the largest cost line is compute. Expect more aggressive tiering, more usage-based structures and fewer unlimited allowances after any listing. That dynamic sits inside the wider pricing shift covered in the analysis of where AI returns are actually measured.

The reseller question you should answer now

Because a large share of revenue flows through cloud resellers, many customers do not contract with the model provider directly. They contract with a hyperscaler and consume the model through it.

That distinction decides who your counterparty is, whose terms govern your data handling, and who you negotiate with if pricing changes. It also decides whether you are visible to the provider at all as an account worth retaining.

Find out which you are before a listing, not after. The answer is usually in the contract rather than in anyone's memory of how the relationship started.

What changes for a direct customer

Direct customers negotiate terms with the provider and appear in its revenue as a named account. That gives you visibility and it gives the provider a reason to retain you.

It also means you carry the provider's credit and continuity risk directly. If capacity is constrained, allocation decisions are made by the party you contracted with, and large direct accounts generally fare better than small ones.

What changes for a reseller customer

Buying through a hyperscaler means your commercial relationship, your data processing agreement and your support path all run through that intermediary. Pricing changes reach you filtered through their margin rather than directly.

The advantage is that you inherit an existing enterprise agreement with terms already negotiated at scale. The disadvantage is that you have limited influence over model access, deprecation timelines or feature availability, because none of those decisions is made by your counterparty.

Neither route is universally better. What matters is knowing which one you are on before pricing moves, because the mitigation available to you is completely different in each case.

Where this reading is weak

Every figure above is a reported private number. None has been audited in public. A confidential filing produces no disclosure obligation until the company chooses to proceed.

The customer count is also a company-supplied metric with a company-supplied definition. A customer spending over $1 million annually could be one large deployment or forty small ones inside a single group.

And the growth rate itself invites a fair objection. Curves this steep have historically been followed by a period of normalisation, and no public market has yet had the chance to price what normalisation looks like here.

What a listing changes that a funding round does not

Private companies choose what to disclose. Public companies disclose on a schedule, in a defined format, with legal liability attached to accuracy. That difference produces three things this market has not had.

The first is a comparable revenue figure. Once accounting policy is stated in a filing, the gross-versus-net question stops being a matter of interpretation. Every comparison written since will need revisiting, including the ones in this post.

The second is a cost structure. Compute is the dominant expense in this business and nobody outside these companies knows the real gross margin. A filing shows whether margin improves with scale or whether each additional customer costs roughly what the last one did. That single disclosure would settle more arguments than any product launch.

The third is a risk factors section, which is the most useful and least read part of any filing. It is where a company states, under legal obligation, what could go wrong. Concentration, model competition, regulatory exposure and supplier dependence all appear there in the company's own words.

For anyone buying enterprise AI, that section is worth more than the growth chart. It tells you what your vendor believes its own vulnerabilities are, which is information you cannot obtain in a sales conversation.

What to watch as the window approaches

The comparison case, including why the other large lab has been weighing a delay, is covered in the piece on OpenAI's listing timeline.

Frequently asked questions

Where to start this week

Two things worth doing before any listing window opens.

First, find out whether your organisation is a direct customer or buying through a cloud reseller. The answer changes who your contractual counterparty is, and it changes what happens to your terms if pricing is restructured.

Second, write down what your AI spend would need to look like at a 30% higher unit cost. Not because that is a forecast, but because listing pressure moves in one direction and you want the answer before the conversation, not during it.

Related reading