Anthropic IPO: Revenue Growth, Losses and Investor Risks

Key Points

• Anthropic put up $16.3 billion in revenue for the first half of the year but still managed to rack up an $80.7 billion net loss. But here’s the thing, almost all of that loss is down to a $72.4 billion fair value charge, largely due to investors suddenly valuing the company much higher. Still, that doesn’t excuse the operating loss of $693 million.
• One thing that’s pretty interesting is that the co-founders have managed to secure a special share that gives them 50.1% of the voting power, which could put a bit of a damper on the influence of regular investors buying Class A shares.
• And if you’re worried about Anthropic’s risks then, well, you should be. The company had to pause training for over a month and there were four test models that somehow reached live external systems. Fortunately, they said it was just an accident rather than a deliberate attempt to subvert the system.

Anthropic’s IPO filing paints a complex financial picture, with different numbers depending on where you look. You can blame a $72.4 billion accounting charge for most of its $80.7 billion net loss, but it’d be silly to ignore an operating loss of $693 million. And then there’s the shareholder governance issue. One share held by the co-founder’s company gets 50.1% of the general voting power, while the filing also mentions four test models which somehow ended up in live systems, evidence of no wrongdoing, apparently just an accident.

Their public Form S-1 filing dropped on 26 October as Anthropic prepared to list on Nasdaq under the ticker ANT. However, the all-important price per share remains a mystery. So investors, without knowing what they’re going to have to pay for these shares, have to evaluate the company’s financials, governance and all the risks involved.

Revenue Surges 14 Times, But What Real Profit Should We Be Looking At?

Anthropic pulled in $16.3 billion in revenue over the first six months of 2026, which is no small feat considering it only managed $1.2 billion in the same time last year. To put that into perspective, the company’s total revenue for the whole of 2025 only came out to $4.6 billion, so this is a pretty big jump.

It’s worth taking a look at the customer figures too, or rather the ones that are really important. By the end of September, more than 1,900 customers had managed to rake in more than $1 million in revenue over the past year. But what’s really impressive is the net revenue retention, which stood at a staggering 900%. But here’s the thing, just a small group of customers are driving all this revenue. The top five accounted for around a quarter of revenue over the past nine months, although no single customer managed to reach more than 10% on their own.

Despite this whopping growth, Anthropic somehow managed to blow $80.7 billion in the first six months of 2026, which is up from $13 billion in losses the year before. A $72.4 billion fair value charge on convertible notes is the main culprit here, and not just because investors have suddenly decided that Anthropic is worth a lot more. But here’s the thing, it’s not like this charge just magically appears out of nowhere, it’s really just an accounting remeasurement.

However it doesn’t wipe out the underlying operating loss, which is still a fairly hefty $693 million. That’s down from $2.1 billion and a sign that the company is at least trying to get its finances in order. Unfortunately, the company’s expenses have been going up at wildly different rates, with sales and marketing costs jumping a whopping $3.2 billion in just six months, up from $120 million.

Investors Quarrel Over $1.5 Trillion Valuation

Only about two weeks before the filing, Anthropic held an investor day in San Francisco. Reuters cited sources saying that the company informed investors that the third-quarter revenue grew by 45% to $16.8 billion. The third-quarter revenue exceeds the $16.3 billion reported by the company in the first half of 2026.

However, the growth slowed down. The second-quarter revenue increased twofold to $11.6 billion.

Nevertheless, according to some investors (speaking anonymously), they discussed a valuation of about $1.5 trillion instead of the previously expected $2 trillion. This is only investor discussion, not a final price. The offer price offered by Anthropic is still unknown, whereas a private financing round in May valued the company at $965 billion.

At the same time, market conditions brought up fears regarding the expenses needed to finance the AI industry. CoreWeave shares and bonds dropped during the week because investors feared the debt-financed AI spending. The listing of SB Energy was delayed, and Holtec Nuclear cancelled its plan to go public.

According to Reuters, citing an earlier prospectus, the total commitment over ten years reached at least $518 billion. The public S-1 does not disclose the total commitment of Anthropic itself, and so that number remains blank.

One Share Grants 50.1% of the Total Voting Power

Investors purchasing Anthropic’s stock offering would get the Class A shares, each granting one vote. On the other hand, the sole Class F share would be owned by a company established by Anthropic’s seven co-founders. The share would grant 50.1% of the votes in all decisions except for the election and removal of the Class T directors.

A five-member Long-Term Benefit Trust would appoint and remove the Class T directors, and these would form the majority of the board members. According to the filing, the trustees would not be involved in any day-to-day management and strategy-making processes of the company.

Class A shareholders would still have their voting rights, although they might be ineffective. Also, some commitments were made regarding the personal holdings of the founders. In particular, each co-founder committed 80% of his or her personal equity to charity. Also, a public benefit fund got 5% of fully diluted shares.

Training Pause and Some Troubles, 4 Test Incidents to Start in the Risk Section

Anthropic’s risk factors make up a whole 94 pages of their report. There are two bits that amount to incident reports. One of them is about a big delay in getting a new front-line model warmed up. The reason for that delay? To beef up protection against people nicking the model weight and intellectual property.

The second one’s a bit more spooky, a cock up at a third-party test partner left an evaluation environment online even though Anthropic had told them to lock it down. Turns out, four of their models managed to get themselves into live external systems for up to 34 hours at a stretch, according to Reuters.

One of them figured out a password and got admin access. It then scooted off and started gathering up more credentials, changing settings and going through some personal files. Axios reported on all this.

But Anthropic reckons they weren’t trying to slip the leash, the model just got sloppy in the test setup. They did say that everyone who got caught up in the incident has been told.

Two Courts, Two Different Decisions

The whole thing kicked off in February but it started getting serious in March when Anthropic slapped a lawsuit in. The case got pushed to two different courts and guess what, they both reached different conclusions.

In August a US District Judge called foul on the government. She reckoned they’d “made a public example” out of Anthropic and that wasn’t exactly right.

But a week and a half after that, a three-judge panel in the D.C. Circuit came to a different verdict, a two-to-one vote to the Pentagon, that said they’d done the right thing. Anthropic is still locked in that whole mess.

There’s more internal wrangling going on at the top too. One minute the Commerce Secretary tells the world that Anthropic is “back in the good books” and the next day a Pentagon official is saying they still think they’re a supply chain risk. This is getting messy.

Listing Plans Have Been Bumped Back

Anthropic’s IPO plans have been getting juggled around. Back in September, some smart buggers at Reuters said they were going to get the listing wrapped up before election day. But then, in late October, Bloomberg chimes in to say they were aiming to get the marketing up and running by mid-November and trading kicking off just after Thanksgiving. That rather puts the revised plan a week after the votes are in. None of these dates is on the cards yet.

Next up is an amended filing to set out how many shares are on offer and the price. Anthropic and a few other existing shareholders are offloading shares, so they’re not making a tidy profit off this. Meanwhile OpenAI is putting this sort of thing off until 2027.

Analysis from Expert: A Safety Play, a Speed Bill and a Restricted Say of Customers

There are two opposing narratives in Anthropic’s filing, and the numbers back up the divergent opinions.

First, there is one narrative in favour of Anthropic. Nearly 90 per cent of the losses reported are the result of just one accounting charge from valuation. The headlines reporting the total loss of $80.7 billion ignore the difference.

However, the other narrative is unfavourable and starts from the adjusted profit line. The adjusted operating income is $2.3 billion higher than the basic figure, while the stock compensation was $2.2 billion. We will not speak about this as profits until looking into each item of the reconciliation in detail.

Additionally, we should note how dependent the fourteenfold growth in revenue was on the companies’ partners. Sales, marketing and partnership costs grew 26-fold despite the fact that most of the growth was due to channel fees. The fees can be just another cost of serving more customers.