I’ll be upfront: I use their models and I’m not a hater. But “I use it” and “I’d put my money in it” are two different conversations. I’m passing on giving them my money. Here are my points, talk me out of them.

1. The IPO is a funding story, not a growth story.

Anthropic is not profitable. It is, in fact, very far from profitable (despite the news and claims you see online: they did not use GAAP when reporting their ‘claimed profits’ in the last two quarters). The IPO is to fund day-to-day operations and service existing debt. When a company goes public to keep the lights on and pay down what it already owes, this is a red flag. Read carefully their use-of-proceeds section twice before you fall in love with the roadmap.

2. The capex treadmill.

The entire capital-expenditure thesis requires two things to stay true, indefinitely: inference keeps getting more expensive, and cloud dependency keeps getting deeper. Every dollar of data center is a bet that the moat holds. But the moat is exactly the part I don’t believe in (see below). You keep building, you keep borrowing, you keep needing the next round and now you need the public markets to keep the belt tight. That’s a treadmill with a countdown on it.

3. The moat is no longer the model.

This is the biggest one. Their frontier models are impressive, sure. But they are facing aggressive competition from all angles:

  • open weights are commoditizing fast, and the gap that matters for day-to-day engineering work is closing
  • closed weights from other big labs
  • Yann LeCun in the making of world models

4. One letter away from the dark.

On June 12th, a single export-control letter switched Fable 5 off for every customer worldwide for 17 days. One letter, and your model is gone. If you’re building a business on models (or an investment thesis) that can be switched off by a piece of paper, you’re not diversified.

5. The irony.

They have frontier models under their fingertips, and those models cannot solve their business model or their financing problems. The so-claimed best models in the world are a wonderful thing to have when your problem is compute. It is a much less useful thing to have when your problem is P&L.

So, after doing my research: not a single cent. I’ll happily keep paying per-token for the models I use from them. But an equity stake in a company that is far from profitable, riding a capex treadmill, whose competition is extremely fierce, and whose flagship model can be switched off by a single letter? Convince me otherwise.

  • irotsoma@piefed.blahaj.zone
    link
    fedilink
    English
    arrow-up
    1
    ·
    1 day ago

    Most tech IPOs in modern times are to create bubbles for short term profit, pass debt off to the public market, etc.

    LLMs I’m their current form are especially unsustainable considering the extremely high cost of training them without researching and filtering the training material first, which is mostly driving income from LLM investors to data centers and extracting profit from there while the LLM company goes into serious debt. Like most IPOs, but especially fad tech, it’s highly unlikely the stock price will not crash, eventually. Though it’s hard for me to say how far or how long it will take since I’m not a pro and not interested in analyzing details.

    Typically, those who have access to get IPO shares with no selling restrictions know how much financial and/or technical debt there is and will likely sell immediately. Those with time restrictions will then sell in 6 months or whenever their share restrictions expire. The majority of other activity during this time will be from short selling, retirement and other funds that will ultimately be used to keep the price from falling too quickly and trading getting halted, companies looking to acquire a majority, split the company’s valuable assets (usually IP) from their debt through bankruptcy or other processes, and let shareholders or bankruptcy absorb the debt, and very short term investments betting on the volatility of IPOs to drive small peaks and troughs before funds pick it up and smooth it out.

    I’m not saying that will happen here, but it is a common pattern with these things. I’m not an expert in Anthropic specifically or a professional financial advisor or anything. Just my experience from being involved in these kinds of things in the last decade or two. Used to be different when company names were valued for brand loyalty and such. Now everything is shell companies and product branding is separate and transferable, e.g. Meta, Alphabet, X, Anthropic, etc., not being valuable only their products’ brands like Facebook, Google, Twitter (finally they forced Musk to learn that, lol), Claude, etc., respectively.

    Anyway don’t take any of this as specific advice, just my opinion based on past observations and experiences.