Elon Musk launches xAI to take on OpenAI

Elon Musk spent months blasting OpenAI for morphing into a "closed-source, maximum-profit demon." Now, he is officially stepping into the ring himself with a startup whose stated mission sounds even wilder than colonizing Mars: to "understand the true nature of the universe."

Musk formally unveiled xAI, an artificial intelligence startup created to go toe-to-toe with OpenAI, Google, and Microsoft. While corporate filings show the venture is legally separate from his X Corp umbrella, Musk made it immediately clear that xAI will lean heavily on the Twitter firehose and Tesla's engineering machinery.

A 12-person frontier all-star team

xAI's initial bench is tight but stacked. The startup's launch roster features just 12 people, almost entirely poached from the cutting-edge divisions of DeepMind, OpenAI, Google Research, and Microsoft Research.

The founding group includes prominent AI minds like Igor Babuschkin, Manuel Kroiss, Christian Szegedy, University of Toronto professor Jimmy Ba, and co-founder Greg Yang. Advising the crew on safety is Dan Hendrycks, director of the Center for AI Safety—the very outfit that recently mobilized over 33,000 signatures calling for a pause on giant AI experiments.

Yang teased that the team plans to tackle the core mathematics of deep learning, aiming to develop a "theory of everything for large neural networks" that pushes generative AI well beyond today's scaling limits.

Feeding on tweets and driving on Dojo

During a Twitter Spaces session shortly after the reveal, Musk laid out the playbook. He openly complained that nearly every AI organization on Earth had illegally scraped Twitter data, blaming those data raids for the platform's recent rate-limiting debacles. His takeaway? xAI will tap public tweets for text, image, and video training just like everyone else did.

However, Musk pointed out that human text alone is not enough. He pointed to DeepMind's AlphaZero self-play approach as an inspiration and argued that truly frontier AI must understand the physical world, not just internet text. That is where Tesla's vast fleet of camera-equipped vehicles comes in.

On compute, Musk has already stockpiled thousands of Nvidia GPUs, but he also hinted at tying xAI into Tesla's proprietary Dojo supercomputer and vehicle silicon, highlighting Tesla's Hardware 4 deployment and upcoming Hardware 5 pipeline.

Pund-IT analyst Charles King remained skeptical, noting that xAI's early footprint pales against the billions poured in by Google and Microsoft. Constellation Research's Holger Mueller disagreed, arguing that Musk understands AI will reshape the global economy and has the track record to make this lean crew punch above its weight.

Why it matters

We think Musk's smartest card here is his unfair proprietary data pipeline. Most foundation model startups are running out of clean web data and have zero bridge to the physical world. xAI, by contrast, sits on Twitter's real-time public discourse and Tesla's unmatched real-world driving data.

The operational risks, however, are massive. Routing public company resources from Tesla and leveraged assets from Twitter into a private Musk startup is a recipe for shareholder friction and regulatory heat. Furthermore, Musk's promise to build a "maximum truth-seeking" TruthGPT that skips standard guardrails will quickly run into enterprise and brand safety realities.

For the broader industry, OpenAI's co-founder has become its most well-capitalized critic. The race for AGI is no longer confined to cloud APIs—it is now directly tied to social platforms, autonomous vehicles, and real-world robotics.