Comment by bunderbunder
5 days ago
OpenAI's leaked financials suggest about a $38bn loss last year. They recently claimed their ad business had an annualized run rate of $1bn. Annualized run rate is a bullshit, non-GAAP, obviously fudged number designed to impress people who've never heard of regression to the mean, but still, let's trust it for the sake of argument. And let's assume that "yesterday's weather" is a good enough estimate for how their overall finances are looking in 2026. That implies that all they need for profitability is to quickly grow their ad revenue by, say, 3,500 - 4,000%.
Easy peasy. Shouldn't be too hard at all.
Ads in an LLM, such as nudging users to pick a certain solution or brand, could easily be orders of magnitude more effective than traditional banner ads. Certainly such numbers are possible if you design the product to fit. We have hardly even seen the beginning of that yet.
I can't think of a greater incentive to only use open weight models than ads.
Disclaimer: I already use open weight models only since ~Jan this year and very happy about it.
Sam Altman having direct access to my chat messages is a close second.
Combined with the terrible system prompts that brick usefulness, increased restrictions from the US model makers, and Chinese equivalents that are actually decent
It’s simply rent seeking, and there is no moat
Imo the US model makers were doomed by the US ecosystem itself
[dead]