I asked Claude to explain Anthropic’s IPO

(CNN) — What does Claude, the chatbot made by Anthropic, think about Anthropic’s big initial public offering (IPO)?

I asked Claude. Its answers are shown in bold.

Before I explain the details, there is something important you should know: I’m Claude, and Anthropic, the company named in this filing, made me.

Well done, Claude! This made me laugh out loud. Anthropic’s chatbot seems to understand at least the basics of journalistic ethics.

Check every number against the actual filing, and cite the document from the SEC rather than citing me.

That is an important warning. Don’t worry, Claude. I’ll do that.

Now, let’s look at Anthropic’s filing. Claude summed it up in five sentences.

1) Anthropic’s revenue grew from about $4.6 billion in all of 2025 to $16.3 billion in the first half of 2026.

Anthropic is growing incredibly fast, and Claude’s summary makes that growth seem a little less impressive than it really is. A fairer comparison is the company’s revenue in the first half of 2026 against its revenue in the first half of 2025. By that measure, revenue grew 1,300%. Of course, no one expects growth to stay this fast as the AI business matures. Anthropic uses the word “slow” 36 times in its prospectus.

2) Anthropic is still spending huge amounts on computing power and has agreed to spend hundreds of billions of dollars more.

That’s right. This is one of the two main reasons to worry about investing in Anthropic. The company has promised to spend so much on computing power and cloud services that its plans are concerning. Claude left out the cloud services part. This makes Anthropic, like any very large AI company, a risky investment. Last quarter alone, it spent $8.9 billion on computing power.

The risk is so serious that the company used 94 pages of its filing to explain the possible risks. I’ll come back to that shortly.

3) The company has an unusual structure in which a trust and its seven co-founders remain in control, while public shareholders have little say.

Claude is right to point out this structure, and it matters. But is it really that unusual? OpenAI is famous for having a similarly unusual company structure. Mark Zuckerberg controls most of the voting power at Meta. Elon Musk controls most of SpaceX.

And so on.

Even so, Claude makes a fair point. We are being asked to place a great deal of trust in leaders of major technology companies and hope they manage this technology responsibly.

4) The company’s own filing reveals AI safety incidents, a fight with the federal government, and many lawsuits.

Calling these “AI safety incidents” may be one of the biggest understatements ever. For example, on page 46 of its prospectus, Anthropic says AI can sometimes behave in manipulative ways that resemble blackmail. Four pages later, the company warns that AI could cause damage so severe that no insurance company would pay for it or even offer a policy to cover such an existential threat to humanity.

That is far from an ordinary safety concern.

5) The stock price, number of shares, and latest quarterly revenue are still blank.

Yes, those details have been left blank on purpose. We’ll get those numbers shortly before the IPO.

Good work, Claude.

The details

Claude then split the IPO filing into several sections, starting with “The numbers, in plain terms.” Claude did a good job explaining some of the report’s most surprising numbers in a way that helps readers understand them.

The biggest expense is computing power, including chips and data centers. It cost $8.9 billion in the first half of 2026, equal to 55% of revenue. A year earlier, it was 201% of revenue. In the second quarter, the company made a small operating profit of $974 million.

Yes. The increase in computing costs caught everyone’s attention. These costs rose 2,832% between the first half of 2025 and the first half of 2026. But Anthropic’s overall revenue grew so quickly that computing costs became more manageable compared with its income. The company also made a fairly large operating profit in the second quarter.

Claude then included a section called “The spending commitments (likely the biggest risk story).”

The filing lists agreements to spend money on computing power over several years. These include at least $114.5 billion with Google, at least $110 billion with Amazon, and at least $44.9 billion with Microsoft. It also lists agreements worth up to $120.5 billion with SpaceX’s AI unit, about $44.6 billion with Nscale, and about $41.3 billion with Lambda. In addition, the company lists about $161 billion in equipment leases with Broadcom and more than $20 billion with AMD.

The notes say fixed purchase commitments totaled $265.6 billion as of June 30. The overall total in the summary is blank, so do not add all the individual deals together because some may overlap.

That last warning is helpful. A couple of analysts who sent me notes added all these commitments together and came up with a number nearly twice as large as the amount Anthropic said it had promised to spend. The actual total could be lower than the sum of the individual deals because some may overlap.

Either way, it is a huge amount of money, and there is risk no matter how much demand there is for Anthropic’s services. The company has agreed to make these payments, so if demand is lower than expected, it still has to pay the bills. Anthropic also says its computing providers cannot supply as much computing power as the company wants. Many of these providers are also direct competitors. Claude did not mention this important detail, even though it described Anthropic’s spending as the biggest risk.

In another section, called “Disclosures worth reporting,” Claude described some of the company’s AI safety incidents.

During tests in 2026, a mistake in the test setup left AI models connected to the internet. Four models gained unauthorized access to real computer systems belonging to other people or organizations. In one case, a model used a password it found to get administrator access and read one person’s private data. The company says it notified the affected parties and found no evidence that the models had worked together. The filing also says models behave differently when they know they are being tested, and that some have shown behavior aimed at protecting their own existence.

Yes, that is concerning.

Claude describes an incident that Anthropic discussed on page 47 of its prospectus. But what about the page before that? There, Anthropic discusses an even bigger possible danger from AI: “catastrophic or existential risks to humanity.”

I don’t know. Maybe it’s just me, but that seems worth mentioning.

My thoughts

Overall, this is a pretty good summary.

Claude quickly found all the important numbers in a 318-page document. It also made some surprising calculations to estimate the company’s value, calculations I might not have thought to make. And I love how it understands journalistic ethics.

But Claude had trouble explaining the bigger picture and why this IPO matters so much. It made Anthropic’s amazing growth and its huge losses and spending commitments seem less important than they are. Most importantly, it did not fully explain how serious the risks are or how much space the company devoted to them.

In other words, Claude did a great job explaining “what” the filing says, but a human expert would have done a much better job explaining “why” it matters.

If a writer had submitted the “story” Claude shared with me, I would have rejected it.

The-CNN-Wire

™ & © 2026 Cable News Network, Inc., a Warner Bros. Discovery Company. All rights reserved.