and 7.81 billion in R&D from last year. I don't know how long it took to build the weights for the current model, or exactly how much that costs, but it's certainly more than zero days and zero dollars.
I also doubt that OpenAI could set that R&D expense to zero and survive without an agreement from Anthropic that they'll do the same... so that R&D expense can't be ignored when figuring up the total cost of the current model.
You're missing the point. There was a lot of debate around if inference was subsidized or not. And that's a huge point to confirm in the public discourse.
Can I ask what is your opinion about their core CapEx, i.e. model training?
The general trend I observe is that the "shelf lives" of large language models are really short. It costs $1-10 billion to train cutting-edge models at the moment, and they only really last 6 months at best.
There seems to be very little brand loyalty too. Whenever a shiny new thing comes out, people just switch over, which implies that they constantly need to fight the time decay.
It's high, really high. But, that isn't bad. In fact... they are better of with it being extremely high. Then scale matters. They need enough revenue at high enough margins to earn a decent return on that spend, but higher is, from a competitive perspective, better.
I understand your logic ("the high CapEx is the moat"), but on the other hand, isn't it be a bit like multiple high speed railway systems trying to connect San Francisco to Los Angeles?
And there are three internal players chasing the same goal at the moment (OpenAI, Anthropic and Google), and two others (Deepseek and Alibaba/Qwen). What will prevent them from cutting the price floor each other?
Looking from a different angle: Microsoft has been able to maintain its monopoly because it was/is a huge pain for companies to switch the operating system, but do LLMs have that stickiness?
> You're missing the point. There was a lot of debate around if inference was subsidized or not.
To answer that question you have to take into account the cost to produce the thing that inference uses. If you don't, then that's like claiming that the total cost of a car is the cost to keep it on a dealer's lot until it's sold.
"Figuring out how much R&D adds to the total cost of a thing" absolutely isn't a new problem. And given that models seem to get supplanted every year, it's not like you're gonna be able to spread those R&D costs out very much.
Ed Zitron claims to not know what terms mean, but that doesn't mean his ignorance is wisdom.
You can just run his content through AI to get a more balanced flash take. Example:
> Zitron repeatedly describes OpenAI as having "removed" costs — $3.74B in 2024, $17.87B and $3.95B in 2025 — via "net loss attributable to noncontrolling members capital," and says "it's unclear what this means." This is standard consolidated-statement mechanics, not a maneuver. When a parent consolidates entities it doesn't wholly own, the slice of losses belonging to other equity holders is split out as "noncontrolling interests." Nothing is removed or hidden; the total loss is unchanged, it's just allocated. Framing it as OpenAI "lowering" its loss "by removing costs" implies something sketchy where there's only routine GAAP. Saying "I will not speculate further" while leaving that insinuation hanging is the rhetorically convenient version of restraint.
For what it's worth, I think AI is a "bubble" and am not convinced at the long-term sustainability or viability of many of these companies but that doesn't mean that every armchair critic actually has the financial expertise to make accurate arguments.
I mean, his whole sensationalized 8X headline is based on a non-cash conversion charge, which is literally the biggest straw man you can find in the financials. He chose it because he's editorializing even as he leads his post with "I am not going to do very much editorializing". Hilarious.
I feel like the labs and their army of bots are responsible for spreading the "inference is subsidized" narrative. It plays right into their hand and justifies high prices and price increases. Anthropic in particular loves milking people.
It is subsidized by gov contracts. Everyone who has common sense immediately said the real money is Altman getting into the governments pants which is why he and Brockman lobby so hard. You take away those contracts and OpenAI is dead in the water.
They are different things. Government money is very predictable and consistent, and based on different calculations that typical consumer-oriented sales. Profits are usually easier.
Government contract is not a subsidy. It's a payment for a product or service provided to the government. Examples of subsidies are section 8 housing or USDA PLC. SpaceX providing launch services to the government is nothing like that.
Distinction without a difference as it pertains to the conversation. Most of the money SpaceX received from the government was under the COTS program where NASA gave SpaceX money to develop a product and then NASA would become a customer of that product. It is as close as you can get to a technical subsidy without it being technically a subsidy.
