Updated October 24, 2026. This post first ran on April 2, when both OpenAI and Anthropic looked set to list in the fourth quarter. Most of that changed, so the post has been rewritten around what is now known.
In April, this post told enterprise buyers two things: both frontier labs were headed for a Q4 2026 listing, and prices would rise once public shareholders started asking for margin, so lock in multi-year contracts now. Six months later, one of those labs has stepped off the track, and the price call was wrong in the direction that matters most to a budget. The more useful advice from April has turned out to be the line we put last: read the prospectus. One of those documents is now close enough to read, at least in draft, and it says more about your AI vendor than any benchmark does.
What follows is the AI IPO 2026 story as it actually unfolded, organized around the questions a buyer should be asking rather than the ones an investor asks.
Who is still going public this year?
Only Anthropic. The sequence went like this.
Anthropic raised a $65 billion Series H at a $965 billion post-money valuation on May 28, then announced on June 1 that it had confidentially submitted a draft S-1 to the SEC. OpenAI filed its own confidential paperwork around the same time, which this site covered when its numbers showed a company losing $1.22 for every dollar of revenue. By late June the New York Times was reporting that OpenAI was leaning toward waiting until 2027, with Sam Altman reportedly treating any cut to a trillion-dollar valuation as a nonstarter.
Altman made it official in a Fortune interview published September 12. Asked about timing, he said “I would say not 2026, yeah,” and framed the delay around AI safety. He did not commit to 2027. We pulled that decision apart at the time in our look at why one lab stopped and the other sprinted, since both companies used the word “safety” to justify opposite choices.
OpenAI has not gone quiet on capital. Bloomberg reported this month that it is seeking at least $30 billion at a valuation of about $1.4 trillion, a private round that, according to Dealroom’s summary of the reporting, is timed to land before Anthropic lists. The private market is doing the job the public market was supposed to do.
Anthropic, meanwhile, held an investor day on October 14, and Bloomberg reported that formal marketing could start the week of November 9 with trading before US Thanksgiving, per FinanceFeeds’ timeline. Anthropic has confirmed neither date, and IPO valuation talk ranges from roughly $1.8 trillion to $3 trillion. The prospectus has to be public at least 15 days before a roadshow starts, so if the November 9 date holds, the public S-1 had to land by about October 23. Reports pointed to that date, but as of this update we have not confirmed a public filing. Every figure below comes from reporting on the leaked draft, not the final filing, and some of it will move.
Did prices go up, the way we said they would?
No. They fell, and fast.
On September 22, Anthropic and OpenAI cut flagship prices within hours of each other. Claude Opus 5.5 launched at $4 per million input tokens and $20 per million output, down from Opus 5’s $5 and $25. OpenAI’s GPT-6 Sol came in at $2 and $10, half the prior Sol rate. Ramp’s lead economist, Ara Kharazian, told Fortune the labs were “engaged in a price war that is driving down the price of AI.” Two weeks later Haiku 5.5 shipped with a 90% cut on both input and output, landing exactly on GPT-6 Luna’s $0.10 and $0.50. We covered both moves in detail: the flagship repricing and what the Haiku cut actually buys.
The April argument went wrong on timing, not on logic. It assumed margin pressure starts when a company files. In practice, the months before a listing reward revenue growth above almost everything else, and the cheapest way to show growth is to cut price and win volume. Anthropic’s investor communications put third-quarter revenue at $16.8 billion, up 45% from the second quarter. That is the number a roadshow is built around, and price cuts helped produce it.
Our view, and it is a view rather than a forecast with a source behind it: the margin conversation arrives after the listing, once quarterly comparisons get harder and insiders can sell. That is the window to plan for, not this one.
What does the draft actually show?
Three numbers set the shape. The Wall Street Journal, which reviewed the draft, reported $16.3 billion of revenue in the first half of 2026, against about $1.1 billion a year earlier, and a first-half net loss of $80.7 billion. About $72 billion of that loss came from non-cash accounting adjustments tied to convertible notes and financing instruments, so the headline loss says more about how Anthropic raised money than about how it runs. For 2025, Fortune’s read of the leaked draft puts revenue at $4.6 billion and the net loss at $42 billion. Compute and infrastructure spending in the first six months of this year was nearly $9 billion, more than the $7.3 billion spent in all of 2025.
The figure from April that aged worst is ours. We cited $19 billion of annualized revenue as a sign of momentum. A single quarter now comes close to that whole annual figure.
How exposed is your vendor to a few big customers?
This is the disclosure a buyer should read first. According to the WSJ, about 20% of first-half revenue came from just two customers, and Anthropic expects a limited client base to keep producing a large share of revenue. Fortune’s account of the same draft says a quarter of revenue comes from two clients. The public filing should settle the gap; either way, the concentration is high.
There are two ways to read that if you are not one of those two accounts. On the upside, the people underwriting your vendor are counting on revenue that does not depend on you. On the downside, roadmap priorities, capacity during shortages and the attention of the solutions team go to whoever produces a fifth of the revenue. If your contract has no capacity guarantee, that is the clause the disclosure should push you to ask for.
What is the compute bill, and why should a buyer care?
Because your vendor’s supply chain is your supply chain. The WSJ reports that Anthropic is committed to roughly $450 billion in cloud compute agreements over the next decade, including $111 billion with Google Cloud, $110 billion with AWS and $31 billion with Microsoft, and that it could spend as much as $750 billion on capacity in the coming years once deals still under negotiation are counted. Dealroom notes that much of the total is payable regardless of usage.
The risk section names the problem directly. According to the reporting, the draft warns that hyperscalers may come to treat Anthropic as a competitor rather than a customer. Investor Chamath Palihapitiya, in remarks Dealroom added on October 23, summed up the warning as compute from Amazon, Google and Microsoft being “curtailed, repriced, or terminated.” That is his phrasing, not a quote from the filing. In April we listed Anthropic’s availability on all three major clouds as a strength. It still is, for distribution. The filing shows the other side: the three companies that host Claude for you are also three of its largest suppliers, owed most of that money whether or not Claude usage grows into it, and each one sells a competing model.
There is a financing layer on top. Broadcom agreed to lend Anthropic up to $42 billion to lease the chips Broadcom helps build, and according to Finimize’s read of the draft, certain payment or performance defaults could make the lease obligations due early and limit Anthropic’s ability to draw on that same financing. We traced how that loop works in our breakdown of the Broadcom loan. For a buyer, the practical question is not whether Anthropic will fail; it is whether rate limits and regional capacity hold steady during a year when the vendor is paying for capacity it has not yet sold.
What should a buyer do differently now?
The April checklist said: lock in pricing, diversify providers, read the S-1, and build internal capability. Here is the revised version, in light of what happened.
- Lock ceilings, not rates. A three-year agreement signed in April at April prices would now be above market. Negotiate a price ceiling with a step-down clause, or a most-favored-customer term tied to the public rate card, so cuts reach you automatically and increases stop at the cap.
- Spend your negotiating leverage on capacity and data terms. Pre-IPO vendors are generous on price because price drives the growth story. The scarcer things are guaranteed throughput, regional availability and retention terms. Ask for those while the vendor wants your logo in the roadshow deck.
- Read four sections of the public S-1 when it appears. Customer concentration, purchase obligations, revenue recognition and the compute risk factors. The draft also warns that damages from a severe AI security event could exceed available insurance coverage limits, the same gap we found when insurers started writing AI out of corporate policies. If your vendor cannot fully insure a catastrophic incident, your contract’s liability cap matters more than it did.
- Keep a second model qualified, for a different reason than before. In April the argument was price protection. Now it is supply: a vendor whose capacity depends on three competitors’ clouds can hit rate limits that have nothing to do with you. A tested fallback on another provider is cheap insurance.
- Put a date on the calendar for after the listing. If pricing discipline arrives, it arrives once the stock trades and quarterly comparisons get harder. Schedule your contract review for then, not for the week of the IPO.
One April question is still open. We flagged the possibility that the SEC would push Anthropic to change how it reports revenue from cloud-credit arrangements. Nothing in the reporting on the draft settles that either way, which makes the revenue-recognition note the single most useful page in the public filing for anyone trying to judge how durable that $16.3 billion is.
For OpenAI customers, the situation is simpler and less transparent. There is no public prospectus coming this year, so the audited view of your vendor’s economics that April promised is at least a year away. The company is raising privately at a higher valuation than the one it walked away from listing at. Until there is a filing, the loss-per-dollar figures from its confidential paperwork and its pricing behavior are the best evidence a buyer has.
The original version of this post ended by telling readers to buy contracts now and read the S-1 later. The order has flipped. Prices are falling on their own; the prospectus is the scarce thing. Read it first, and negotiate with what it tells you. Our earlier coverage of Anthropic’s confidential filing and its first profitable quarter has the background on how the company got here.
