Want to Invest in Anthropic? Here’s What to Know Before the IPO
Plus, what Anthropic’s amended filing still leaves out, and what its ownership structure means for public shareholders.
Ivanna Hampton: Welcome to Investing Insights. I’m your host, Ivanna Hampton. Anthropic’s days as a private company are ticking down. The artificial intelligence frontier lab was founded in 2021 and is reportedly expected to price in mid-November. Many market watchers are predicting the company’s IPO will be one of the biggest of the year. Anthropic recently updated its S-1 to disclose how much revenue it generated in Q3 2026. A key figure left blank in the Claude chatbot maker’s original filing with the SEC. What would the private company’s arrival to the public market mean for investors and what should they watch for? Harrison Rolfes is a senior research analyst for Morningstar’s PitchBook, and he’s been tracking Anthropic’s rise. Welcome to Investing Insights, Harrison.
Harrison Rolfes: Well, thanks for having me. I’m excited to talk about everything Anthropic.
Hampton: Well, of course. Let’s start with an explainer on Anthropic’s business. What does it build, who are its customers, and what differentiates it from its rivals?
Rolfes: Sure. Anthropic builds a Claude family of AI models and sells access to them by the token. So, about 80% of revenue is metered usage through its API. Then AWS, Google Cloud, and Azure marketplaces, which equated to about 48% of first-half revenue. Then, you have agents like Claude Code, and then the other 20% is from subscriptions. In terms of their customers, their customer base is enterprise, not consumer like OpenAI. More than 11,900 customers spend over $100,000 a year as of Sept. 30, which is up from about 950 a year earlier. More than 1,900 of those customers actually spend over $1 million a year, and more than half are actually Fortune 500 companies, which is wild.
The difference from OpenAI is that OpenAI is more consumer-facing, and the rest of the differences are actually structural. Anthropic sells to businesses and developers first, and they buy compute across five chip sources. Think of Nvidia NVDA, Broadcom AVGO, Google GOOGL, Amazon.com AMZN, Advanced Micro Devices AMD, et cetera, rather than just one. They run their business with about 5,300 employees, which equates to about $6.6 million of trailing revenue per employee. On top of all that, they’re a public benefit corporation whose board majority is chosen by a trust, which shapes both the growth and the governance policies.
Hampton: Now, PitchBook gives AI companies an AI Business Quality score that measures how durable the business is and how well it treats outside investors. After Anthropic released its S-1, you lowered its AIBQ score. Can you talk about the areas, like revenue quality, that declined according to your team’s analysis? We have a graphic from your report that we can put up on the screen.
Rolfes: Yeah. We lowered Anthropic’s AIBQ score to 7.45 from 8.17 after the filing because the business is stronger than outsiders could see, but the terms to investors are actually weaker. Revenue quality fell because 80% of revenue is pay-as-you-go, which means that people can turn or switch models anytime they want without any switching costs. And then they also have contracted, but unrecognized, revenue. Think of a backlog, which is where customers can turn down at will. Also, the backlog itself reaches about $13.8 billion.
As of Sept. 30, that’s two and a half months of sales, and about, I would say, roughly 75% of it is expected to convert within a year. That contract backlog hasn’t quite hit yet. Nearly half of revenue flows through cloud partners whose fees sit inside a selling line that’s equal to about 19% of first-half revenue. The filing itself attributes 5% to 7% of nine-month revenue to customers who violated their usage policies, which, according to my math, was about $2 billion. On top of all that, you have the founder LLC, which is what the seven co-founders own shares under, which equates to 50.1% of the vote. The corporate governance, the revenue quality, and everything in between are actually not as favorable to investors, which is why we lowered the score.
Hampton: Anthropic’s revenue has grown significantly since early 2025. It recently revealed to the Securities and Exchange Commission that it had about $17 billion in revenue in the third quarter. What do you make of that, Harrison?
Rolfes: Well, the third-quarter revenue is about 12 times the $1.38 billion a year earlier and 47% above the second quarter. The second quarter is $11.56 billion, which annualizes to roughly $68 billion. The growth rate slowed sharply from about 144% in Q2 and 132% in Q1. And then the filing actually gave us the reasons because of this. One, compute capacity ran short of demand; summer seasonality in enterprise buying also lowered, and then they actually had to cut off their usage policies for the ones who were violating it. And then about $600 million of price reductions and contract discounts on top of all that. Demand wasn’t really the constraint. It was the compute, which is a major topic right now. They had to buy $5 billion of short-term capacity to fill the gap, which pushed compute to about 58% of revenue. And then, they turned Q2’s $974 million operating profit into a $567 million Q3 operating loss.
Hampton: As a follow-up, why focus on dollars added rather than the growth rate?
Rolfes: Because percentage growth actually decays as the company gets bigger. Forty-seven percent of an $11.6 billion base is about $5.4 billion of new quarterly revenue, which is more new money than tripling would have produced a year earlier. Dollars added tells you how much new demand the company actually captured and whether new compute is being converted into sales. By that measure, Q2 added, I want to say, $6.8 billion and Q3 added $5.4 billion. The first quarter in the filing to add fewer new dollars than the one before it is something that we should take into consideration. On top of all that, what matters for Q4 is whether new revenue comes in above the roughly $5.4 billion, which supports our case, or if they go below $4 billion, it’ll actually break that case.
Hampton: Harrison, help us understand how Anthropic spends what comes in and what profitability looks like. What should investors know?
Rolfes: Compute and infrastructure are definitely the largest costs, which is why they had to front-load as much of this compute from data centers, et cetera, way out into the future. So, that $9.86 billion in Q3 equated to about 58% of their revenue just to get that infrastructure in place so that they have the capability of supporting the demand as they continue to grow. On top of all that, on a GAAP basis, the company lost $567 million from operations in Q3 after its first operating profit in Q2. That’s $1.3 billion of operating losses for the nine months in between. The reported $23.3 billion net loss, which is about 94% of which is a noncash charge on their convertible notes that essentially rises as Anthropic’s valuation goes up, is another thing to take into account because it adjusts their operating income and it also adjusts the amount of losses that are booked on the records, which if you take all that together with the revenue as well as their costs, you’re looking at about a 15% margin, which was down from 19%.
From all that, investors should actually take two things from this. One, the economics only work when compute is bought on a multiyear contract, and it stops working when it’s bought in the short term. Once roughly $108.6 billion of convertible notes convert at the IPO, the losses mostly disappear, while the stock compensation of roughly $11 billion a year will remain the real cost.
Hampton: Harrison, thank you for buttoning that up for us. We appreciate that.
Rolfes: Yes.
Hampton: Talk about the firm’s debt and long-term commitments. I mean, this chart from your report shows the biggest suppliers. How much annual revenue would Anthropic need to generate to cover the payments, and what are the risks?
Rolfes: Sure. As of Sept. 30, Anthropic had $895.4 billion of purchase and lease commitments. That’s $69.8 billion due in the next 12 months and then $447.5 billion before the end of 2030. The rest actually runs into 2045, with the largest counterparties being Broadcom, SpaceX SPCX, et cetera, which means that the near-term payments alone equal roughly today’s revenue, which is $69.8 billion due against the $68 billion annualized. What we need to see from the filing is the payment schedule, which puts the 2028-30 obligations near $115 billion a year, which at Q2’s 48% compute ratio would require roughly $240 billion of annual revenue, which is about 3.5 times their current level.
Hampton: Can you explain Anthropic’s ownership structure and what it means for public shareholders?
Rolfes: Absolutely. Anthropic has five share classes. The first is the public Class A share, which gives one vote per share, but it’s capped at 49.9% of the vote in aggregate. Then you have a single Class F share, which is held by an LLC, founder LLC, with the seven co-founders sitting inside it, and they own the 50.1% on the other side, but the vote has no economic rights, so that’s pretty important to take into consideration. There’s a single Class T share that’s held by the long-term benefit trust, which elects four of seven directors.
And then, there’s a nonvoting Class N share, which is held by Amazon, that’s capped at 12% ownership, as well as Google. There’s a public benefit fund that holds about 5% of fully diluted shares. For a public shareholder, this means that economic exposure has no path to change the board or its capital allocation. What’s important for investors is that public investors aren’t really going to have a choice on board removal, et cetera, if they feel that Anthropic is not going in the right direction.
Hampton: What are three key areas you would want Anthropic to disclose in its amended filing?
Rolfes: There was an amendment on Monday, which actually filled in a good deal from the original S-1, which is actually the third-quarter results, which is what we’re looking for. That put us at the $895.4 billion in commitments, with $69.8 billion due in 12 months and then $447.5 billion before 2031. First, a year-by-year payment schedule is what I’m really trying to dig into so we can understand exactly how much they owe to each person when it’s due. That way, we can compare it to their revenue to see whether or not they’re on track to actually pay off their commitments.
The second is revenue durability. The share of revenue under minimum spend contracts versus pure consumption. We want to see cloud marketplaces as well, which, what I mean by cloud marketplaces is the partner fee breakdown because right now, the way they do accounting is hard to splice out what that fee is per partner. And then third is actually the offering terms themselves. We want to know the price range, the share count, the split between company and selling shareholder shares, the run rate revenue figures, et cetera, because we want to know exactly what Anthropic wants to be valued at and we want to know how much we’re going to be paying per share.
Hampton: I know you’ll be watching for that. A couple of quick questions. SpaceX made a big splash with its IPO in June. What does its debut tell investors about how a deal of Anthropic’s size can trade?
Rolfes: Well, the lesson is that a trillion-dollar listing with a small float can actually produce a large first-day pop that the market gives back once that scarcity fades. All the people who had FOMO or who wanted to really support Anthropic, it’s going to reflect an Anthropic share price almost immediately, but as we continue to watch it over the next, let’s say month, we might see a similar dip the way that SpaceX did. On top of all of that, some employees might sell off about 10% of the locked-up shares on day one and about 35% after the 2026 results. We need to judge the deal against the price range, not against the first day of trading.
Hampton: Does the fact that Anthropic is on track to go public before OpenAI suggest that it will be the leader in the space?
Rolfes: Not necessarily. Anthropic has just taken control of this short window right now. We just had midterms, and now we can go into a pocket of the market that allows people to essentially distract themselves from everything else. The difference between Anthropic and OpenAI is that Anthropic focuses on enterprises, and in order to grow at the rate that they’ve told investors, they really need to focus on capturing the market in this moment. On top of that, OpenAI already told a bunch of people that they don’t want to IPO so that they can clean up their financials and be more prepared for the future. At the end of the day, it’s a sequencing event where Anthropic has just decided to take this opportunity to fulfill not only the obligations, but to move forward with the power and scaling laws that they identified in their S-1.
Hampton: What’s the takeaway ahead of Anthropic’s blockbuster IPO?
Rolfes: The takeaway is that the S-1 proves that Anthropic is a real business. They have about $17 billion of quarterly revenue, more than 11,900 customers paying over $100,000, roughly 900% net revenue retention, and a 15% adjusted operating margin, which proves that the price has to carry three things that growth doesn’t necessarily settle. First, $895 billion of compute obligations, revenue that’s 80% pay-as-you-go, and no shareholder vote. At the end of the day, there’s still no price range, but the roadshow can begin as soon as Nov. 10, and some reports are pointing to trading the week of Nov. 16 or just before Thanksgiving. In my overall opinion, the key takeaway is Anthropic has a lot more to work on, but they have established that they are a real business that continues to move forward.
Hampton: Harrison, thank you so much for joining me on Investing Insights.
Rolfes: Yes, thank you for having me.
Hampton: Before we go, let me tell you who’s the next guest on The Long View podcast. It’s Cambria Investment Management CIO, Meb Faber. Co-host Amy Arnott and Ben Johnson will ask Faber about why the US market has been so resilient over time and about key lessons he’s learned from market history. Check out the new episode on Tuesday, Nov. 10. Click the link in the show notes now, right now, to follow the podcast. Thanks to senior video producer Jake Vankersen, associate multimedia editor Jess Bebel, and the PitchBook team. I’m Ivanna Hampton, editorial multimedia manager, Morningstar. Take care.
The author or authors do not own shares in any securities mentioned in this article. Find out about Morningstar’s editorial policies.

