Anthropic IPO – Game On!

The Good, the Bad and the Disingenuous.

• Anthropic’s S-1 reveals that its situation is much better than we had been led to believe, but there are still plenty of disingenuities that continue to support my position that one can not really trust anything that this company says.
  • The Good: which to be fair, is almost all of the financial statements.
  • In H1 26, the company grew revenues nearly 16x YoY while compute costs grew by a much smaller 3.8x YoY, allowing the company to come close to break-even and generate $3.9bn in cash from operations.
  • This trend was also evident in Q2 26 (page 157) where revenue increased 2.4x QoQ while compute costs increased just 1.6x.
  • Furthermore, the customer concentration has also decreased materially, with no single customer now accounting for more than 11% of revenues, which greatly reduces the risk of a major upset.
  • The company also has over $130bn in cash and short-term investments that were raised through the issuance of $108bn of convertible debt that will convert at the time of the IPO.
  • This will result in an unencumbered balance sheet, giving Anthropic plenty of options going forward.
  • Furthermore, the massive loss recorded in Q2 26 is almost entirely due to an accounting adjustment due to the increase in the value of the shares it has issued to early investors that will convert at the IPO.
  • Hence, as a going concern, and looking at this company in this snapshot of time, I am pretty happy and nicely surprised.
  • The Bad:
  • Looking at the numbers leads me to think that they have been presented to paint the best possible picture of the company rather than to give investors a true and fair picture of the company’s current position.
  • For example, by presenting H1 2026 in the main financial disclosures (unusual) and tucking Q2 26 away on page 157, it hides the slowing growth picture.
  • Another example is how the company has hidden its compute commitments, as the front page data is redacted, meaning that one has to go digging on page 170 to find it.
  • Here, Anthropic has massive, non-cancellable contracts with Microsoft ($44.9bn), Google ($114.5bn), Amazon ($110bn), SpaceX ($120.5bn (90-day notice)), AMD ($20bn), Nscale ($44.6bn), Lamda ($41.3bn) and Broadcom ($161.2bn).
  • This makes a total of $535bn (excluding SpaceX) of non-cancellable commitments that the company has to meet between now and 2033.
  • That amounts to an average of $76bn per year, giving an idea of how massive the cost base is now committed to becoming.
  • This means that if demand does not materialise as expected or there is a huge price war with OpenAI or someone else, it has to spend the money regardless which could bankrupt the company.
  • I doubt that it would come to this and that contracts would end up being renegotiated, but for equity investors this would be a virtual Armageddon.
  • The Disingenous:
    • IPO: if what this company says is true, it has no business being listed on the stock exchange.
    • Anthropic states at the outset: “We are first and foremost a research lab-our mission is the responsible development and maintenance of advanced AI for the long-term benefit of humanity”.
    • In my opinion, Anthropic is none of these things.
    • Instead, it is a for-profit enterprise that intends to maximise the economic opportunity that its products have created.
    • There is nothing wrong with this, but trying to extract a premium by pretending to be a virtuous company is disingenuous and misleading in my opinion.
    • A research lab exists to further the cause of science and does not seek profit, meaning that it has no business being listed.
    • So, either Anthropic is being less than truthful about who it is, or it should not be listed.
    • It cannot have it both ways.  
    • Regulatory capture: It is clear from the ever-changing names at the top of the league tables that there is no sustainable moat in large language models.
    • This means that those investing billions in every generation of LLMs need to find another way to keep competition at bay.
    • This is what the “catastrophic or existential risks to humanity” and the “we all need to slow down” comments are all about.
    • Any federal regulator will look to Anthropic, OpenAI, et al for guidance on how to regulate them, meaning that they will be able to set the rules and squeeze the smaller companies and especially open source out of the market.
    • IPO: If Anthropic was really concerned about the well-being of the human race, it would not expose itself to the commercial pressures that could force it to potentially cause the very harms of which it warns.
    • Instead, it would delay its IPO until such time that it could be sure that its products are safe and under control, but this is not what is happening.
    • This is yet another example of the ethical conflicts of interest that riddle this company once again, underscoring that its statements outside of its financial disclosures are not to be trusted.
• The net result is that Anthropic should be viewed as a for-profit company that will maximise its opportunity, and it is offering to take investors along for the ride.
• Investors will have no say in how the company is governed or managed and so will have to put up with management decisions, be they good or bad.
• Taking all this into context, with the fact that RFM Research sees OpenAI as having reaccelerated its growth in Q3 26 and having a better cost base going into 2027 makes me think that the risks outweigh the opportunity at a valuation of $2tn.
• If it has a heavy fall right after the IPO, it may be worth looking at but for now, I am sitting this one out.