Musk prepares X.AI to compete with OpenAI

On March 9, 2023, X.AI Corp was registered in Nevada; the state’s official entity search allows readers to check entity E28915122023-8 and separate the corporate fact from plans that had not yet been announced.

Elon Musk has taken a formal step toward returning to the artificial intelligence business with X.AI Corp, a company registered in Nevada on March 9. Corporate filings list Musk as director and Jared Birchall, head of his family office, as secretary.

The initiative puts the entrepreneur up against OpenAI, the organization he co-founded in 2015 and left the board of in 2018. The new company has not yet announced a product or technical details, but the registration comes amid reports that Musk has been in contact with researchers and investors to launch his own AI project.

A new front in the rivalry with OpenAI

The timing is no coincidence. OpenAI has gone from being a lab known mainly in technology circles to occupying the center of the industry following the launch of ChatGPT in November 2022 and GPT-4 this past March. Microsoft, its main partner, has integrated the technology into services such as Bing and strengthened an alliance that makes OpenAI one of the toughest rivals to catch.

Musk has openly challenged that evolution. His main criticism is that OpenAI was founded with a promise to develop artificial intelligence for the benefit of humanity, under a nonprofit structure, and that its commercial relationship with Microsoft has transformed that original approach. OpenAI maintains a nonprofit entity that controls its structure, alongside a limited-profit subsidiary that finances the costly training of its models.

The disagreement is not merely about business. In late March, Musk signed an open letter alongside researchers and business leaders calling for a pause of at least six months in the training of systems more powerful than GPT-4. The letter called for shared safety protocols, independent oversight and rules to prevent the commercial race from driving developments that are difficult to control.

Starting a company of his own while calling for a pause may seem contradictory, but it reflects a central debate in the industry: who sets the rules for the most advanced models and what incentives shape them. Musk is not stepping away from AI; he is seeking the ability to influence its direction.

Musk already has experience, data and computing power

X.AI is not starting from scratch within Musk’s orbit. Tesla has spent years developing computer vision systems and neural networks for its assisted-driving features. The company is also working on Dojo, a computer designed to train models on the huge volumes of video collected by its vehicles.

Twitter, acquired by Musk in October 2022, is another potentially important piece. The platform hosts conversations, news and public content at massive scale—material that can be valuable for training language models. In recent weeks, reports have linked purchases of Nvidia graphics processors to AI projects at Twitter, although their specific use has not been clarified.

Training models comparable to GPT-4, however, requires far more than access to data. It takes specialized chips, engineers capable of designing and fine-tuning the systems, power infrastructure and considerable investment. The shortage of graphics processors and the concentration of talent at a handful of companies make the competition difficult even for a billionaire.

More competition, but also more questions

X.AI’s arrival would add pressure to a market where Google is accelerating Bard, Anthropic is developing Claude and companies such as Meta maintain a significant position in research. For users and businesses, greater competition could mean more options, better performance and lower prices for access to generative models.

But the decisive question will be its approach. An AI company is defined not only by the model it trains, but also by what data it uses, how it evaluates its errors, what limits it places on its responses and who is accountable when it fails. Musk has made those issues part of his criticism of OpenAI. X.AI will have to show through its actions that it offers a different technical and safety alternative—not merely another competitor in the race to build ever-larger models.

A corporate filing proves existence, not strategy

A public record can establish a name, jurisdiction, filing date and declared officers. It does not demonstrate a model, a complete team, closed funding or authorised access to data owned by other companies. Read a newly formed corporation through two columns: facts in the document and claims requiring another source. “X.AI Corp exists” belongs in the first; “it will compete with a particular product” was still a journalistic hypothesis.

Nevada’s search does not expose a stable link for each result, so the reproducible reference is the entity name and number. Preserving that reference, the consultation date and a capture of the record is better than linking to an aggregator. If an officer changes later, a new search shows current state but should not rewrite what was visible in April.

OpenAI’s history also requires dated sources. The 2015 founding announcement named Musk as co-chair and described a non-profit laboratory. It supports his founding role and the original promise; it does not document every later corporate change or turn Musk’s criticism into a neutral conclusion.

How to separate an apparent contradiction from a proven one

The letter published on March 22 called for a pause of at least six months in training systems more powerful than GPT-4 and listed oversight conditions. Signing it while forming a company can appear inconsistent, but a contradiction would require evidence that the new entity was training a covered system during that period. A registration alone does not show that.

The same discipline applies to shared resources. One person controlling companies with data, chips or talent does not automatically give a new company the right or ability to use them. A transfer requires a contract, announcement or operating evidence. Corporate proximity is a lead to investigate, not permission to add every asset together.

The transferable skill is to read a company's birth in layers: legal existence, team, capital, infrastructure, data, product and evaluation. Each layer needs its own dated evidence. A report that jumps from incorporation to a future product without the intermediate steps is presenting possibility as capability.

The article date creates another boundary: xAI’s public launch came later and cannot be used to fill retrospectively what was known at incorporation. The name suggested a direction; it did not reveal mission, staff or product. An honest chronology leaves those fields blank until a dateable announcement appears.

Following the case requires only a milestone table with a link and evidence level: filing, hiring, funding, infrastructure, demo, technical documentation and public access. Every new row may confirm or reject an earlier hypothesis without altering previous rows. This avoids two symmetrical errors: declaring a winner after incorporation alone and dismissing a company because its first document did not yet contain a model.

When a product arrives, the same table continues with version, access terms, declared sources, evaluation and limits. The original filing keeps its value as the chronology’s beginning rather than advance proof of success. Knowing when to leave a field blank until evidence appears is central to the method.