# I don't think anyone would deny that LLMs are hugely useful for a ton of tasks (coding, customer service, translation & editing services).
Looking back historically towards similar industries, no one would deny that air travel has similarly changed the world for the better. But I can't help but wonder if LLM companies could end up like Airlines.
Note, I'm ***not talking about Nvidia or other hardware companies***. Airbus itself has been consistently profitable for the last 25 years.
I think there are the following similarities:
1. Massive capital costs (GPUs versus Planes)
2. Relatively fixed per user costs (No more SAAS cost reduction)
3. Basically commodity products (you got a small percentage of people paying for first class & high intelligence models, while most people fly economy class on whatever is cheapest)
4. Tight security (manual identity verification versus TSA pat down)
5. Capital equipment leasing companies (Hyperscalers leasing GPUs versus Aircraft Leasing Companies)
6. Lots of competition (LLMs could become commodities, where open source models are good enough & significantly cheaper)
***GLM 5.2 costs $4.4 per million*** output, while ***Opus 4.8 costs $25 per million*** output, nearly 85% lower cost, and GLM 5.2 (open source) is only less than 5 percentage points away from Opus 4.8 Medium effort on the independent DeepSWE benchmark (software coding tasks). See [https://deepswe.datacurve.ai/](https://deepswe.datacurve.ai/)
Obviously everyone wants to use Fable/ChatGPT-5.6, but if you can't due to govt restrictions, and you're stuck at Opus 4.8, why shouldn't you just cut costs by 85%? And if Fable/Mythos/ChatGPT-5.6 get unbanned? Open source LLM labs could just distill them again.
Now we have OpenAI is delaying their IPO, until maybe 2027 to clean up their balance sheet, I somehow don't think any of these frontier labs will ever make a single cent.
For reference, airlines have lost money for roughly 26 of the 48 years since deregulation.
