Elon Musk bought 10,000 GPUs for Twitter's secret AI project

Elon Musk signed an open letter asking everyone to pause giant AI experiments for six months. Meanwhile, he was busy buying thousands of high-end graphics cards behind the scenes.

According to Business Insider, Musk recently purchased roughly 10,000 GPUs for an internal AI initiative at Twitter. The hardware is set to run in one of Twitter's two remaining data centers. As one source pointed out, there is basically no financial reason for Twitter to drop that kind of cash on data-center-grade processors unless it plans to train massive AI models.

DeepMind talent and Twitter's proprietary data

The GPU shopping spree follows Twitter's recent hiring of two former researchers from DeepMind just over a month ago. With hardware and research talent now in place, Twitter appears ready to build its own generative AI.

Insiders report that the project involves training large generative models on Twitter's massive proprietary data archive. While exact product plans remain under wraps, the tech could be deployed to rebuild Twitter's struggling ad platform or radically upgrade its native search capabilities.

Musk has spent recent months attacking OpenAI, an organization he helped co-found back in 2015. He publicly complained on Twitter about how an organization he backed with roughly $100 million as a non-profit somehow transformed into a $30 billion for-profit entity.

The roots of the OpenAI feud

Reporting from Semafor indicates the tension between Musk and OpenAI is deeply personal. Back in 2018, Musk reportedly told Sam Altman that OpenAI was lagging dangerously behind Google, proposing that he should take over and run the lab himself.

Altman and the other co-founders pushed back and rejected the takeover. While Musk officially exited OpenAI citing potential conflicts of interest with Tesla's self-driving efforts, he reportedly cut off his promised $1 billion funding commitment after leaving.

That sudden cash crunch forced OpenAI to create a for-profit arm in 2019 to secure capital, which paved the way for Microsoft's initial $1 billion investment. When ChatGPT debuted in November and took over the tech world, Musk was reportedly furious. Within weeks, he severed OpenAI's access to Twitter's data firehose.

Why it matters

Musk's moves look intensely pragmatic, if a bit hypocritical. Calling for an industry-wide pause while quietly snapping up scarce compute resources looks a lot like buying time to catch up with Microsoft and Google.

For Twitter, this is about leveraging its single biggest untapped asset: a massive, continuously updating stream of human dialogue. By locking OpenAI out and training internal models, Twitter can attempt to revive its ad business and search without relying on third-party vendors.

However, this is an expensive bet for a company that has been cutting costs, firing staff, and shutting down server infrastructure. Deploying and maintaining 10,000 top-tier GPUs will strain Twitter's already tight operational budget. In the broader AI landscape, Musk's re-entry guarantees that data moats will become far more guarded, as platform owners realize their archives are too valuable to give away to competitors.