Anthropic confidentially submitted a draft S-1 to the SEC on June 1, 2026. The filing later leaked, and the picture that emerged was not pretty. The company posted nearly $42 billion in net losses in 2025 and is looking at $518 billion for cloud compute through 2030. Last year’s revenue was only $4.6 billion. Dario wants a public-market debut that could value the company above $2 trillion after a May mark of supposedly around $965 billion. Sam’s shop confidentially filed in June as well. Same furnace, same landlord pricing, same compute bill. These businesses are not profitable. The compute they need to keep going is insane, and the prospectus is the receipt.
The Leaked S-1 and the Headline Numbers
Anthropic confidentially submitted the draft S-1 on June 1, 2026. It posted that fact on its site, so it was not much of a secret, even if the filing itself was confidential. Late September is when Reuters and The Straits Times started walking through the actual numbers from the prospectus. Revenue was up 12 times to nearly $4.6 billion. Operating losses were over $8 billion. Compute spend was almost $8 billion. The obligation needed to stay on this trajectory is $518 billion. Nearly a quarter of the revenue came from two customers who are not locked into long-term contracts. The debut will likely come after the November midterms, and there will be a lot of chatter about a $2 trillion valuation. Translated plainly, there is no clean, profitable way to talk about this. These businesses are not profitable and are nowhere close to profitable.
2025 Results Look Like a Train Wreck
Reuters could see a valuation above $2 trillion. The 2025 financials look like an absolute train wreck. Revenue was $4.5 billion, which the company is saying is up a thousand percent year over year from $368 million. A thousand percent year over year sounds impressive. The operating loss was $8 billion, which had been $3 billion. The earlier ratio was $3.6 million of revenue against $3 billion of losses. The percentage of losses has been cut as revenue moved to $4.6 billion against operating costs of $8 billion, but GAAP accounting is part of the story. Fancy GAAP accounting covers for the net loss by carrying those losses over.The headline remains a $42 billion net loss that needs context. Roughly $34 billion came from an accounting charge tied largely to the rising of a financial instrument. Anthropic spent $7 billion on compute in 2025, up 190 percent year over year. To keep training these models, the company will need more and more processing. Compute need more than doubled year over year and was nearly three times the year previous. The spend has continued to skyrocket. It is not leveling off. That compute figure alone represents 58 percent of $12 billion.
Better Models, Not 200 Percent Better
The models are getting better, but they are not getting 200 percent better. This is the point of diminishing returns. A new model came out and barely even made a headline. Differences now look like 1 and 2 percent. On benchmarks the company will say it is 10 percent. It is not. There are definitely not 200 percent gains.
Two Customers Carry the Revenue
Amazon and Google are major strategic partners and Anthropic’s two biggest customers. Nearly 25 percent of revenue came from those two customers. Neither is locked in. Either one can walk away. Anthropic is competing in an increasingly expensive race with OpenAI, Google, Meta, and xAI. This month the CEO wrote that “we must slow down the pace.” This week the company shipped a new flagship model. The irony is obvious. The CEO also asked Washington for an antitrust waiver so that rivals can coordinate how fast they go. Now the company wants $2 trillion from the public.Seven founders who own about 14 percent will control 50.01 percent of the vote. That is what makes the structure striking. You carry the losses. Two customers carry the revenue. Seven founders carry the vote. Humanity carries the risk. Either this thing is as dangerous as they say and nobody should be selling it, or it is not, and an 80 percent page of apocalypse language is sales pressure. The filing also highlights risks around increasing autonomous AI systems, including unexpected behavior, security concerns, and potential misuse. An IPO after the November midterms would give public-market investors one of their first pure-play ways to value a frontier AI lab.
The Same Numbers, Scaled to a Household Budget
One breakdown divides Anthropic’s actual numbers by 100,000 so they become relatable. Someone comes to you and says you make $45,000 a year, you spend $126,000, and somehow you are worth $20 million. You also just promised you would spend $5 million. Another version of the same scaled picture: you make $48,000, you spend $126,000, you are worth $5.1 million, and you just promised you would spend another $5 million. That is how the loop works. Rich people keep investing so the company can keep buying their product.
Why This Math Does Not Work Forever
If this kind of math is expected to work forever, it will not. The losses are being passed outward so the company can pocket the upside on the way through. That will not work forever. There is not going to be $518 billion worth of value of profit from these companies between now and 2030. Competition and the open-source model change the picture. A team can build harnesses that do every single thing a cloud-code subscription can do. That is why putting money into this company does not make sense to the speaker.The speaker is personally more in the crazy camp. The valuation looks hugely inflated. This is the bubble, and they are trying to pass the bubble onto the market. The speaker will not put a dollar into OpenAI or Anthropic. That is not financial advice. It is an opinion. The value does not look like it is there. The numbers do not make sense.
An Open-Source Alternative and Custom Software
That is the reason Openmonoagent exists: a way for people to run their own AI. Hardware can start as simple as a small brick. Each unit lets you run your own hardware and your own AI stack. Work built on Openmonoagent includes products such as Swiftcase Legal.ai and startup security. The same framework is being given away for free. Free as in free. Open source. The goal is to democratize this. Openmonoagent.ai is a terminal-native AI coding agent running entirely on local LLMs, with zero API costs, zero telemetry, and full ownership. It is growing fast because serious engineers recognize real infrastructure when they see it.StartupHakk builds custom software solutions for companies. Spencer Thomason is a fractional CTO and founder of StartupHakk, with a decade of executive leadership as a fractional CTO and 25 years in software development. That work includes transforming tech teams and products and building custom AI solutions. Executive leadership experience includes organizations such as GoDaddy, SRP, and Wells Fargo. Most companies do not have a technology problem. They have a leadership problem. They pay for it in missed deadlines, failed integrations, and AI investments that delivered nothing. Bad technology decisions do not just waste money.
They kill companies.The businesses winning right now are not the ones chasing the latest AI trends. They are the ones built on solid engineering who treat AI as infrastructure they own, control, and integrate into software that actually works. StartupHakk is a custom software development team doing database architecture, API design, system integration, and scalable infrastructure built the way it has been done for over 25 years. When AI belongs in a solution, it is built in. Not bolted on. Not a wrapper around someone else’s API. It is designed into the architecture and runs in your environment, not a vendor’s cloud and not on someone else’s pricing schedule. As a fractional CTO, the same standard of leadership is available without full-time executive costs: strategic architecture, hands-on delivery, and clear accountability. No slide decks that gather dust. If an organization wants custom software built to last, with AI integrated where it actually makes sense, the sites given are startuphakk.com and openmonoagent.ai.
Conclusion
Anthropic confidentially filed an S-1. The leaked numbers show $4.5 billion to $4.6 billion in revenue, operating losses over $8 billion, about $7 billion to $8 billion in compute spend, a $42 billion net loss with a large accounting charge inside it, and a $518 billion compute obligation through 2030. Roughly a quarter of revenue sits with two unlocked customers. Seven founders who own about 14 percent would control 50.01 percent of the vote. The company is not profitable. The compute bill is enormous. Whether public markets will pay for a valuation talked about above $2 trillion is the question the leaked prospectus now puts in front of everyone.
