Now, after the “shocking” news of a sudden xAI’s availability of over 300 megawatts in GPUs, equal to about 220K Nvidia GPUs, let’s point to the main suspect.

Meta.

Meta has been the second biggest client of Nvidia since 2023. Meta has been stockpiling millions of Nvidia GPUs. Some are deployed, and some are collecting dust. Meta “builds” on average 2.5 JPMorgan Chase Tower worth of value every quarter.

Meta is not a hyperscaler. Meta does not have an LLM with demand like Anthropic, OpenAI, or Google. In some ways, Meta is very similar to xAI. While xAI has its own model, demand is still relatively small. Now Meta has its “lab,” and lets its teams use available GPUs freely. Meta has been striking compute deals with many neoclouds. Meta also encourages its engineers to use AI for programming because it has abundant compute availability.

Meta has a problem of oversupply, a big problem. But Zuck is a friend of Jensen, and he cannot disappoint him by reducing GPU purchases, even if Meta already has enough. Zuck has FOMO. He sees “shortages” and wants to stockpile even more, because he can. Meta has the cash, and it is easy for the company to access debt financing, so why not?

What could happen? Nothing, right?