Bubble's gonna burst!

OpenAI generated US$4.3 billion in revenue in the first half of 2025, according to financial disclosures to shareholders.

The artificial intelligence firm reported a net loss of US$13.5 billion during the same period, with more than half attributed to the remeasurement of convertible interest rights.

Research and development expenses were its largest cost, totaling US$6.7 billion in the first half.

(OpenAI current valuation: $500Bn.)

https://www.techinasia.com/news/openais-revenue-rises-16-to-4-3b-in-h1-2025

@cstross that's a lot more revenue than I would have expected, frankly.
@laird A lot of it is probably churn by big corporate customers like MSFT, AAPL, and META who want to convince their shareholders they're all in on AI. Once the bubble bursts they'll have no reason to keep throwing money at OpenAI.

@cstross @laird

For those who haven't been watching:

A load of AI companies are actually just OpenAI resellers. OpenAI is among the biggest investors in these companies, but NVIDIA is also big. The way it works is something a bit like this:

  • NVIDIA invests $10M in some AI company.
  • That company takes $2M in revenue for effectively reselling OpenAI models at well below cost.
  • That company buys $12M of OpenAI credits (from their revenue plus investor money).
  • OpenAI reports $12M of revenue and uses this to justify raising $100M more investor money.
  • OpenAI buys $100M of GPUs from NVIDIA.
  • NVIDIA reports $100M of revenue and an asset worth at least $10M.
  • Everyone wins! Money from thin air!

    There's a more fun variant that is, roughly:

  • Investment firm has a $100M stake in NVIDIA.
  • Investment firm borrows $50 against this stake and uses it to invest in a load of AI startups (maybe even OpenAI).
  • Demand for GPUs goes up!
  • NVIDIA valuation increases significantly and now the firm's NVIDIA portfolio is worth $1B
  • Company slowly sells 20% of their stake (slowly, so it doesn't spook the market, but it's a fairly small fraction of market cap, so that's fine).
  • Company writes off the $50M investment in startups.
  • Company now has $150M in cash and an NVIDIA stake valued at $800M that they can borrow against.
  • If anyone except Trump were in the White House, I'd expect the end result of a bunch of SEC investigations to result in a lot of folks going to prison.

    @david_chisnall @cstross @laird I've been thinking for a while that NVIDIA's operating model has become:

    1. Identify compute-intensive scam (crypto, LLM)
    2. Tweak GPU product pipeline to enable scam
    3. Push scam incessantly to get customers to buy tweaked products
    4. Identify next scam before previous one burns out.

    Also, there is a special place in Hell reserved for whomever decided to label smallish matrices as "tensors" in LLM-speak. They aren't *wrong*, just wildly misleading.

    @oddhack @david_chisnall @cstross @laird not sure Nvidia look for scams as such. They just look for anything that eats vast amounts of compute that fits their product line, it just happens to be scams and gaming. No different to oil companies pushing AI because it drives consumption of their product.
    @etchedpixels @david_chisnall @cstross @laird I was running the OpenGL Architecture Review Board from 1997-2006 so got to watch as the traditional workstation vendors were eaten from beneath by NVIDIA and ATI. I occasionally comforted myself by thinking that, although NVIDIA won the market, at least SGI's customer base was mostly intelligent adults doing science & engineering, not teenage boys blowing shit up.
    @oddhack @david_chisnall @cstross @laird What drives markets is always about volume and price. We did so much Linux network optimization for the porn industry !
    @etchedpixels @david_chisnall @cstross @laird when I joined SGI in the late 90s they had a commanding lead in the porn server market. Unfortunately the marketing people were reluctant to capitalize on this clear evidence of technical superiority (and really, Origins were pretty sweet systems).
    @oddhack @etchedpixels @david_chisnall @cstross @laird too many companies and projects won’t use porn company references. It’s sad as they’re often cutting edge.