But it's not that unusual, right? If I build a house this year and sell it next year, the house might still be profitable even if next year I'm building 3 more houses, so the company as a whole is still in the red on an annual basis.
I mean, I'm not a financial expert but that doesn't seem all that unusual to me.
The second prong of the argument is basically that, when you invest in Anthropic, you can't just invest in one model and then collect the profits from that model. You're investing in a whole company in the hopes that they can be profitable overall; at some point they'll need to stop spending so much money on training and give it back to the investors instead. Zitron argues that this isn't going to happen because training is actually something that companies need to do to retain customers at all. An analogy here might be the fact that Microsoft has to spend a certain amount of "R&D" budget fixing security vulnerabilities in Windows Server just to retain their current customer base; if attackers found out about a serious security hole but Microsoft didn't fix it, everyone would need to stop using Windows Server. LLM companies do the same kind of thing to fix "jailbreaks" and other unexpected model behaviors.
The third prong of the argument is that, in general, there's a long history of companies using creative accounting to try and make themselves look profitable and then collapsing because they're not actually profitable. For example, WeWork's "community-adjusted EBIDTA" figured claimed the company was profitable using very similar arguments to Dario, and then the company went bankrupt. If you're already cooking the numbers, you have almost arbitrary flexibility to report whatever "margins" you want by excluding some of your costs from the calculation.
Construction companies capitalize and depreciate over many years so they can answer "yes" they are profitable even when they are very cashflow negative. This is exactly Dario's point: model training costs are treated as expenses but in practice are much closer to construction costs. Model training effectively produces an asset, the model weights, which will generate revenue for many years into the future.
> Zitron argues that this isn't going to happen because training is actually something that companies need to do to retain customers at all.
This is exactly why Dario's point about each training run being profitable is so important. It suggest that this is not true. Customers are happy to use old models long enough to fully pay off their costs.
> there's a long history of companies using creative accounting
Zitron seems to know very little about accounting evidenced by him using terms like "gross margin" wrong in this article. He's pattern matching against his limited exposure to company financials to find superficial similarities between the AI labs and famous frauds. Find me a company that doesn't report non-GAAP measures. Google search claims 96% of SP 500 companies do it. Are they all frauds too? Sometimes non-GAAP adjustments are eye roll inducing but they are tolerated because they can be genuinely useful to get a fuller picture of the business.
I guess I don't blame you or anybody for having a deal of cynicism, but these arguments just don't seem very concrete. Like, if Dario was lying or not, he probably wouldn't share the actual numbers, and he probably would propose a "model lifecycle" accounting. And if the business model had potential or not, there probably would still be vast investment in the next model. Zitron has had nothing but cynicism towards AI from the start, and it's his whole shtick, and so these arguments don't seem very credible coming from him, even though they seem reasonable coming from you.
