Anthropic Just Filed for Its IPO—How Big Is the Opportunity?

The loss headline versus the real risk

0:00 Anthropic's IPO filing shows an $80.7 billion loss in the first half of 2026, and that isn't the number I'd worry about most. Most of that loss came from revaluing financial instruments rather than cash leaving the business. Operating cash flow was actually positive. By the end, you'll know which numbers actually matter, and why the same stock can look more or less expensive depending on the revenue period you choose. The bigger risk is committing to years of bills when your customers could leave.

0:25 Imagine signing years of kitchen leases because your restaurant is packed. While your customers can decide tomorrow that they'd rather eat somewhere else. And if you're buying the IPO, the question is whether future growth can cover those commitments and justify the price you pay. A great AI product can still be an expensive stock.

What Anthropic investors would own

0:40 First, you need to know what investors would actually own. Anthropic makes Claude, including the chat bot, the tools that help people write software and services other companies can build into their own products. So this business reaches beyond somebody paying for help, drafting an email. The company has applied to list on the NASDAQ exchange under A-N-T, ANT. But as of October 26th, the offering size and price range are still undetermined. And filing the paperwork does not mean the shares are already trading.

1:05 You still need the price tag.

Revenue growth and enterprise demand

1:07 Start with the strongest part of the business. Revenue reached 16.2 billion in the first six months of 2026, compared with 1.1 billion in the same period of 2025, roughly 14 times as much. And the acceleration happened inside that period because quarterly revenue increased from 4.7 billion in the first quarter to 11.5 billion in the second quarter, which is why the latest quarter deserves its own look. That growth deserves your attention. Even if you don't really like the hype, the customer details help explain it.

1:34 By September 30th, Anthropic counted more than 11,900 customers generating at least $100,000 of revenue over the preceding 12 months, including about 250 generating at least 10 million. Those are spending thresholds rather than a claim that every customer signed a long-term contract. The larger customers are included within the broader group. you cannot add all the categories together. Dario explained that enterprise focus to CNBC in January. Because enterprises really, they value the reliability, they value the trust, they value the long-term relationship with customers.

2:04 That focus helps explain why a customer might start using Claude in a small team, then put it into more workflows across the company. Anthropic benefits when that existing customer uses more, even without winning another account.

What 900% net revenue retention means

2:14 The filing reports approximately 900% net revenue retention at September 30th, meaning the qualifying existing customer group generated roughly nine times its earlier quarterly revenue after expansion, contraction, and customers leaving were accounted for. That is remarkable, but it is not a customer retention percentage. The calculation compares the latest quarter with the corresponding quarter a year earlier. Excludes pro and max subscribers and starts with customers that generated at least $1 in the earlier period.

2:39 So very small starting accounts can grow dramatically. That helps explain why the bull case has substance. Companies are spending real money and expanding their usage. Although 900% should not quietly become the growth rate you assume forever in a spreadsheet. Wojak’s spreadsheet can make anything look affordable.

Compute costs and operating results

2:55 Now put that growth beside the cost. Compute and infrastructure expense was 8.9 billion in the first six months, which is about 55% of revenue before personnel, marketing, partnerships, and other operating expenses. But that 55% is not enough to calculate a normal software gross margin because Anthropic groups, computing costs for research, model development, customer service, and internal use together in that expense category. The label matters as much as the numbers.

3:18 Now, there is another wrinkle in the revenue. For certain cloud marketplace arrangements, Anthropic records the customer sale at its gross amount and puts the channel partner's fee in expenses, so the reported sale is not money it keeps. Even with those costs, second quarter operating income reached $974 million on $11.5 billion of revenue, an operating margin of about 8.4%, while the full first six months still showed a $693 million operating loss.

3:43 That is progress, not proof that the hard part is over.

Accounting losses, cash flow and dilution

3:46 So how do you get from that operating loss to an 80.7 billion net loss? The largest bridge is a 72.3 billion charge from revaluing convertible notes, derivative liabilities and agreements for future equity. Think of financing that can turn into ownership. As the company becomes more valuable, the accounting value of those obligations can rise, producing a large expense without an equivalent cash payment during that period. An accounting charge is not the same thing as a cash bill.

4:11 Another 8.7 billion of other expense was primarily driven by initial recognition and changes in the fair value of assets and liabilities under note purchase agreements. It also needs context before you label it operating spending. But non-cash does not mean irrelevant because financing the converts and the shares can affect how much of the company existing investors own. And future investors still need to examine the conversion terms and the total share count.

4:31 You can avoid a cash payment and still give something up. The cash flow statement shows $3.9 billion provided by operating activities in the first six months. But Anthropic says the improvement also reflects the timing of customer receipts and vendor payments alongside better results after non-cash adjustments. That is why I would not describe this as a permanently self-funding business. Operating cash flow is only part of the picture. Investment needs continue and the filing itself warns that additional financing could be needed sooner than expected.

4:57 The timing of the bills is the next thing to check. At June 30th, Anthropic had 112 billion in cash and cash equivalents plus 19 billion in short-term investments and it reported 110.7 billion of financing cash inflows during the first six months. That is a substantial cushion, but a large bank balance does not tell you how comfortable the business is until you know what it has already promised to spend. The restaurant still has those kitchen leases.

Dario Amodei on compute risk

5:28 In February, Dario used a hypothetical $1 trillion of compute spending to illustrate this. Or if the growth rate is 5x a year instead of 10x a year, then you go bankrupt. He wasn't saying Anthropic was bankrupt, he was explaining why demand forecasts matter.

Long-term infrastructure commitments

5:44 Take Amazon, Anthropic committed to at least $110 billion between May 2026 and April 2036, regardless of actual usage, subject to the agreements, adjustments, and exceptions, with a payment to make up a shortfall. Google has a separate commitment of at least approximately $114.5 billion through July 2033, and Microsoft has at least $44.9 billion from November 2026 through May 2033. those are multi-year amounts, not annual bills. The calendar is doing a lot of work here.

6:13 The SpaceX AI arrangements are different. They describe up to 120.5 billion through 2029, but allow cancellation on 90 days notice after initial non-cancelable periods ending in January and March, 2027, under the respective agreements. Broadcom equipment leases add another major disclosure, approximately 161.2 billion of rent over their lives excluding interest and Anthropic also describes Nscale, lambda, AMD, and additional TPU capacity arrangements with their own conditions.

6:39 You cannot treat every agreement as the same type of debt. The filing leaves the aggregate contractual obligations table and several totals blank. So adding every headline amount together would produce our own calculation with different periods and conditions, not a verified total supplied by the company. The cleaner risk is already explicit. Much of Anthropic's revenue depends on usage without long-term customer commitments. While computing obligations can extend for years, so a slowdown could leave expensive capacity without enough demand to pay for it.

7:04 But there's another side of this, if demand keeps growing, securing capacity early could become an advantage. Imagine customers arriving faster than you can serve them. Having the computing power ready could help Anthropic meet that demand instead of turning business away. That's the bull case, not a forecast. The commitment itself doesn't tell us whether management made a good decision. What matters is how much that capacity customers ultimately pay to use.

Customer concentration and competition

7:32 The restaurant can lose customers without losing its rent bill and demand is concentrated. The top five customers generated approximately 25% of revenue over the first nine months of 2026, even though no single customer exceeded 10% during that period. That does not mean one customer controls the business, but it does mean changes in a few large accounts can matter, especially when some customers can reduce spending and the cloud platforms also influence distribution.

7:55 The biggest customers do not have to disappear to create a problem. Competition makes that more complicated. Google is both a major infrastructure partner and an AI competitor, and the filing identifies OpenAI and Google among its principal rivals. So better models alone do not guarantee pricing power. The bull case is that Claude becomes deeply embedded in valuable workflows and gets cheaper to operate. The bear case is that competitors force prices down while the research and infrastructure bills keep growing.

8:21 Both stories can contain real growth.

Legal and AI safety risks

8:23 Legal and safety risks also belong in the investment case. The 2025 results include a 1.5 billion Bartz settlement expense, and the filing describes continuing intellectual property, security, regulatory, and AI safety exposure. Those warnings are not predictions that a particular disaster will happen, but they can translate into litigation costs, restrictions on products, reputational damage, or slower deployment, all of which can affect the business that you are buying.

Voting control and shareholder dilution

8:45 Now look at what owning a share actually gives you. Public investors would buy Class A shares with one vote each, but a single Class F share held by the founder's company is designed to carry 50.1% of the voting power on matters put to stockholders generally. Separately, the long-term benefit trust generally elects the majority of the board through Class T shares. So this is not a simple arrangement where public shareholders can take control by buying most of the economic ownership.

9:09 You are buying into that governance structure, not choosing a new one. The public benefit fund also matters. The planned initial issuance represents 5% of fully diluted shares as of June 30th on the filing’s stated assumptions, and the board has a discretionary framework for additional issuances at valuation milestones. Specifically, it may issue up to an additional 2% of the defined share base whenever the share price rises 50% above the previous milestone, starting from the IPO price.

9:39 That is discretionary dilution, not an automatic cash dividend to you. The mission and the shareholder economics both belong in your analysis.

What an IPO valuation would need to justify

9:48 So what would the IPO price need to justify? NBC reports that the offering could value Anthropic above $2 trillion, but the actual valuation remains unknown. So let's use $2 trillion only as a hypothetical example. Multiply second quarter revenue by four, and you get a $46.22 billion annualized pace. A $2 trillion equity valuation would be about 43 times that figure, which is a sales comparison not a price to earnings ratio. An annualized quarter is not a full year of earned revenue.

10:14 Now keep the company and that assumed valuation exactly the same. What happens to the sales multiple. If we use the first half instead using the first six months and doubling them instead gives 32.5 billion, making the same hypothetical valuation, about 61 times annualized sales, the business did not change between those calculations. Only the period we chose growth can eventually make today's price reasonable, but only if enough growth and profit actually arrive in a generous valuation leaves less room for slower adoption, price cuts, higher costs, or additional dilution.

10:43 Liking Claude, it's not a valuation method. Before committing money, compare the final price with the fully diluted share count, the latest financial results, and the infrastructure payment schedule.

What to check before buying

10:52 Then check how much of the offering funds the company versus selling shareholders. So what would make you more comfortable? I'd want to see customers keep spending after the initial excitement, cash generation hold up even when payment timing is less favorable, and a price that leaves room for growth to disappoint. A strong launch alone wouldn't answer those questions. Think of it as a test the business has to keep passing rather than a story you only need to believe once.

11:16 Revisit customer concentration, the durability of cash generation, and the final lockup terms as the filing changes. The point is to decide what evidence would justify the price rather than letting the excitement decide your assumptions.

Price, commitments or control?

11:27 Which would change your view the most? The price, the commitments, or the control? Anthropic has shown extraordinary demand and meaningful operating improvement, but the return available to a new shareholder depends on what they pay and what growth must happen afterward. You do not have to confuse an impressive business with a good deal. YouTube picked this next one for you, and the admission price is easier to evaluate.