Date: November 8th, 2026 10:49 PM Author: The Penis
Yes there is a current shortage, it isnt just demand forecasts. Interestingly enough there still might be an investment bubble because part of whats driving building more is based on demand forecasts that could potentially still be too optimistic.
Date: November 8th, 2026 11:05 PM Author: The Penis
Frontier ai labs operating at Scale who need to find thousands of accelerators. Open ai is operating across like 5 cloud providers. Their main contract is with Azure, but they rely on AWS and oracle too and some other specialized providers for certain things. For smaller companies it can be an issue too, not as big because they don’t need as much but you have to look across a bunch of different data centers for availability for things like h100, h200 and have to request access from customer service.
Date: November 8th, 2026 11:15 PM Author: The Penis
Idk exactly what they use for what I think the breakdown is complicated. It’s something like this: Azure was their original cloud partner and they get first releases and apart from open AIs platform running on azure they have a separate platform on azure where Microsoft hosts open ai models. Bedrock on AWS does this too now. But open ai and chat gpt themselves, through the open ai console and chat app are hosted across these providers using kubernetes. Azure, AWS are the main ones but they use oracle and google cloud too. For training rhey have some special facility called star gate through oracle, but that’s not all they use, I’m sure it’s distributed. They use a couple smaller companies for specialized inference deployment and training too, one of them is called cerebras.
It seems extremely unlikely that "Frontier ai labs operating at Scale" are sacrificing their R&D capacity to ensure that consumers have sufficient capacity to make AI girlfriends and "employed cats" videos.
Date: November 8th, 2026 11:19 PM Author: The Penis
Why would the existence of consumer inference workloads show that labs have sufficient compute for their desired research programs? You can’t make a valid inference connecting those. You have to consider that inference is also what generates revenue to finance reaearch.
Date: November 8th, 2026 11:34 PM Author: The Penis
lol their revenue is all from consumer subscriptions, enterprise access and api tokens. What do you think investors are all given them money purely to Seek Knowledge?
The data centers they want to build are for something other than connecting a bunch of Nvidia chips to produce some kind of output? What are they going to put in them if you know?
Date: November 8th, 2026 10:54 PM Author: The Penis
Btw the current shortage only really applies to massive ai companies. But even for small companies you start on AWS and azure with a GPU capacity of 0, compared to cpu compute you can just deploy. You have to put in a request and talk to customer service just to deploy a tiny vm of GPU.