They're doing about a billion per month in revenue by running proprietary models on GPUs like these. Unless they're selling inference with zero/negative margin, it seems like a business model that could be made profitable very easily.

▲

mgh95 4 days ago | parent | next [-]

Revenue != profit, and you don't need to become net negative margin to be net unprofitable. Expensive researchers, expensive engineers, expensive capex, etc.

Inference has extremely different unit economics from a typical SaaS like Salesforce or adtech like google or facebook.

▲

humanizersequel 4 days ago | parent | next [-]

All of those expenses could be trimmed in a scenario where OpenAI or other big labs pivot to focus primarily on profitability via selling inference.

▲

mgh95 4 days ago | parent [-]

Currently, selling LLM inference is a red queen race: the moment you release a model, others begin distilling and attempting to sell your model cheaper, avoiding the expensive capitalized costs associated with R&D. This can occur because the LLM market is fundamentally -- at best -- minimally differentiated; consumers are willing to switch between vendors ("big labs", as you call them, but they aren't really research labs) to whomever offers the best model at the lowest price. This is emphasized by the distributors of many LLMs, developer tools, offering ways to switch the LLM at runtime (see https://www.jetbrains.com/help/ai-assistant/use-custom-model... or https://code.visualstudio.com/docs/copilot/customization/lan... for an example of this). The distributors of LLMs actively working against LLM providers margin provides an exceptionally strong headwind.

This market dynamic begets a low margin race to the bottom, where no party appears able to secure the highly attractive (think the >70% service margin we see in typical tech) unit economics typical of tech.

Inference is a very tough business. It is my opinion (and likely the opinion of many others) that the margins will not sustain a typical "tech" business without continual investment to attempt to develop increasingly complex and expensive models, which itself is unprofitable.

▲

humanizersequel 4 days ago | parent [-]

I don't disagree but you're moving the goalposts. I never said that they could achieve the profits of a typical tech business, just that they could be profitable. Also, the whole distilling problem doesn't happen if the model is proprietary.

	▲	mgh95 4 days ago \| parent [-]
		> I don't disagree but you're moving the goalposts. I never said that they could achieve the profits of a typical tech business, just that they could be profitable. Also, the whole distilling problem doesn't happen if the model is proprietary. In the absence of typical software margins, they will be eroded by providers of "good enough" margins (AWS, Azure, GCP, etc.) who gain more profit from the bundled services than OpenAI does from the primary services. This has happened multiple times in history, either resulting in smaller businesses below IPO price (such as Elastic, Hashicorp, etc.) or outright bankruptcy. Second, the distilling happens on the outputs of the model. Model distillation refers to the usage of a models outputs to train a secondary smaller model. Do not mistake distillation for training (or retraining) to sparse models. You can absolutely distill proprietary models. In fact, that is how DeekSeek-R1-Distill-Qwen and the DeepSeek-R1-Distill-Llama are trained. This also happens with Chinese startups distilling OpenAI models to resell [2]. The worst part is OpenAI is already having to provide APIs to do this [1]. This is not ideal, as OpenAI wants to lock people into (as much as possible) a single platform. I really don't like OpenAIs market position here. I don't think it's long term profitable. [1] https://openai.com/index/api-model-distillation/ [2] https://www.theguardian.com/technology/2025/jan/29/openai-ch...

▲

mrandish 4 days ago | parent | prev [-]

> Revenue != profit

Indeed. And even if that revenue is net profitable right now (and analysts differ sharply on whether it really is), is there a sustainable moat that'll keep fast-followers from replicating most of OpenAI's product value at lower cost? History is littered with first-movers who planted the crop only to see new competitors feast on the fruit.

▲

AlexandrB 4 days ago | parent | prev [-]

And even if they are selling inference at negative margin, they'll make it up in scale!

	▲	kapone 4 days ago \| parent [-]
		These kinds of phrases are...eerily similar to the phrases heard right before...the .com bust. If you were old enough at the time, that's exactly what the mindset was back then. The classic story of the shoeshine boy giving out stock tips...and all that. We all know how that turned out.