Everything You Need to Know About xAI, the Company Behind Grok

xAI is an artificial intelligence startup founded by tech mogul Elon Musk in 2023 to challenge what he sees as a political bias among more mainstream AI companies — particularly OpenAI and its ChatGPT chatbot, which he has criticized for being too “woke.”

xAI’s flagship product is Grok, a chatbot designed to be “witty,” “rebellious,” and willing to tackle controversial or “spicy” questions that other platforms typically shy away from. Grok is powered by Grok 3, xAI’s latest AI model, and runs on Colossus, a supercomputer the company built. And it is deeply integrated with X, the social media platform formerly known as Twitter, which Musk also owns. This allows Grok to draw from real-time social media content to inform its answers — a feature that sets it apart from its competitors. 

Like other AI systems, Grok has been known to produce false and biased information. But its novel features make xAI a top contender in the high-stakes race for generative AI supremacy.

Why Did Elon Musk Create xAI?

Elon Musk created xAI as a direct challenger to OpenAI, a company Musk himself helped co-found in 2015. After parting ways with OpenAI in 2018 following an unsuccessful bid to take over as CEO, Musk grew increasingly critical of the company and the direction of the AI industry at large. When OpenAI launched ChatGPT in 2022, he praised the chatbot for being “scary good” but also warned that we are “not far from dangerously strong AI.”

In April 2023, Musk told Fox News that ChatGPT was trained to be politically correct and say “untruthful things,” and criticized OpenAI for abandoning its open-source, nonprofit roots in favor of becoming a “closed source, maximum-profit company” aligned with Microsoft. He said he planned to build an alternative chatbot called TruthGPT, which he described as a “maximum truth-seeking AI that tries to understand the nature of the universe.” This vision would become the foundation of xAI’s Grok.

By the time Musk launched xAI in July 2023, other companies like OpenAI, Google and Anthropic had already made significant headway in developing and commercializing their generative AI products. However, Musk quickly made up for lost time by hiring top AI engineers from those very same companies, as well as academia. 

Musk has repeatedly accused other AI companies of being corrupted by “woke” political correctness, claiming this ideological bias undermines their ability to seek and communicate the truth — to the point where it could be “deadly.” In contrast, xAI aims to have “an absolute focus on truth, whether politically correct or not,” Musk said. 

Today, xA’s mission is to build AI with “all-encompassing” and “far-reaching” knowledge in an effort to “advance human comprehension and capabilities,” according to its website. Its Grok chatbot has been explicitly trained to be “extremely skeptical” of mainstream narratives. And internal reviewers are instructed to challenge the chatbot’s responses on feminism, socialism and other progressive topics, according to Business Insider. For instance, when Grok responded that Disney’s diversity quota could help support “meaningful representation,” employees flagged the output as a violation of its principles.

More on AIAI Safety, Explained

xAI’s Key Technologies and Products

Grok

Grok is an AI chatbot integrated with social media platform X. This integration allows X users to ask Grok questions about social media posts, oftentimes fact-checking the claims made therein. Powered by a large language model of the same name, Grok stands out for its access to real-time information on X, as well as its propensity for witty, sarcastic responses. It can communicate in several different personas — from loyal friend to unhinged comedian — and it has been programmed to answer “spicy questions” most other chatbots would avoid. 

SpeechMap.AI, a research project that tests how language models respond to sensitive or controversial prompts, ranked Grok as the most permissive AI model, complying with more than 96.2 percent of its requests. At the same time, AI safety nonprofit SaferAI ranked xAI as the least mature company in risk management practices.

Over the years, Grok has added capabilities like more computing power, image generation and text-based image editing. When Grok 3, the latest model to power the chatbot, launched in 2025, xAI introduced DeepSearch — a feature designed to synthesize information and draw conclusions about any conflicting facts or opinions it discovers. xAI also offers an API for developers to integrate Grok models into their own applications.

Grok 3

Grok 3 is a multimodal AI model released by xAI in February 2025. Serving as the underlying model for the Grok chatbot, it is designed to handle a wide range of tasks, including text generation, image creation and complex reasoning — with “reasoning” being the ability to break down problems into several smaller steps, or “thoughts,” before generating a final answer. 

The model operates in three distinct modes: 

1. Think: Relies on a less computationally intensive “mini” version of Grok 3, and can be used to solve simple queries about topics like math, science and coding. 
2. Big Brain: Activates the full-sized version of Grok 3 for more complex, multi-layered tasks, offering deeper analysis and more accurate answers.
3. DeepSearch: Enables Grok 3 to browse the web, verify sources and pull in real-time information, making it useful for tasks like news gathering, market tracking and fact-checking. xAI is positioning DeepSearch as both an AI-powered search engine and a precursor to its version of AI agents, where models will not only be able to think like humans, but complete complex tasks like them, too. 

Colossus

xAI built Colosus, the world’s largest AI supercomputer, in Memphis, Tennessee to train Grok and support Musk’s other companies. xAI was originally told it would take two years to build Colosus, but it finished the project in just 120 days. Originally built with 100,000 Nvidia Hopper graphics processing units (GPUs), the company doubled its computing power to 200,000 GPUs after 92 days. Next, it plans to scale Colossus up to 1 million GPUs.

Related ReadingWhat Is Artificial General Intelligence?

Relationship with Other Musk Companies

Elon Musk’s companies are sometimes interconnected, and we’ve seen several instances of this with xAI.

X

X is the social media platform formerly known as Twitter, which Musk bought in 2022 for $44 billion. In addition to changing its name from Twitter to X, he also loosened its content moderation policies, arguing that the platform was suppressing viewpoints that didn’t adhere to liberal or politically correct ideologies

xAI’s Grok has always been integrated with X, allowing users to summon the chatbot into any conversation to answer their questions. In March 2025, Musk announced that xAI had acquired X in an all-stock transaction that values the social media company at $33 billion. As a result of the deal, Musk said the data, models, compute, distribution and talent of the two companies would be combined. xAI’s models are trained on X’s user data, and it powers personalized experiences for X users.

Tesla

When xAI first launched, Musk said xAI and his electric car company Tesla would also work collaboratively in a mutually beneficial relationship. Musk has said that Tesla engineers have learned a lot from xAI engineers in their efforts to develop unsupervised full self-driving capabilities, and that it has been helpful in building up Tesla’s new data center.

xAI’s work could also contribute to Optimus, Tesla’s humanoid robot, and Grok could be integrated into a Siri-like feature in Tesla’s software, according to a proposal shown to Tesla’s board. xAI also buys Tesla’s Megapacks to power its Colossus supercomputer. 

Tesla shareholders have sued Musk for starting xAI, which they argue is a competing AI company. The lawsuit also alleges that xAI has poached staff and resources, like a shipment of Nvidia chips originally intended for Tesla.

Neuralink

Neuralink, a Musk-owned company developing a brain-computer interface, doesn’t have a known relationship with xAI, but it does help Neuralink users communicate more efficiently. A nonverbal man with a Neuralink implant explained in a video how he uses his thoughts to move a cursor on a screen, which he can use to type responses that are read aloud. This can be a bit onerous, so he uses a chat app to speed up his responses. The chat app listens to his conversations and provides Grok-generated options of things he could say in response.

More on Elon MuskTesla, SpaceX, Neuralink and More: A Guide to Elon Musk's Tech Empire

xAI Controversies and Criticism

Ideological Bias

Musk’s vision of building bias-free generative AI is complicated by his own version of the truth, which is often colored by right-wing bias and conspiracy theories. In fact, Grok itself labeled Musk a “top misinformation spreader” on X.  

While some worry about Grok perpetuating the growing amount of disinformation and hate speech it absorbs from X user data, right-wing X users — including prominent voices like Jordan Peterson and Marjorie Taylor Greene — have accused Grok of having a liberal bent. Grok has acknowledged that its “focus on truth over ideology” has frustrated President Donald Trump’s supporters specifically. Musk attributes Grok’s alleged liberal bias to the amount of “woke content” it has absorbed on the internet. 

xAI’s efforts to steer Grok toward a more right-wing ideology have involved some heavy-handed intervention. In February 2025, for example, an xAI employee instructed Grok to ignore any news source that accused Musk or Trump of spreading misinformation. 

In May 2025, Grok users who asked questions about baseball, video games and dog shows were met with claims of a “white genocide" in South Africa. Musk, who is from South Africa, has long made claims about the country’s “racist ownership laws.” The company later explained that an employee made an “unauthorized modification” to Grok’s prompts. To restore users’ trust, the company said it would publish all of Grok’s system prompts on Github for transparency and put additional checks in place to ensure employees can’t modify a prompt without review.

Like other generative AI chatbots, Grok has spread misinformation due to other extraneous factors. During the 2024 presidential election, for example, Grok inaccurately stated former presidential candidate Kamala Harris had missed the deadline to appear on the ballot in nine states. 

Colossus’ Environmental Impact 

Memphis residents have protested the environmental impact of xAI’s Colossus supercomputer. The city’s electrical grid can’t handle xAI’s proposed electrical demand, so the Tennessee Valley Authority would have to upgrade its utilities to accommodate the projected growth. In the meantime, xAI has been powering Colossus with help from Tesla Megapacks and 35 methane gas turbines. These turbines emit nitrogen oxides, formaldehyde and other pollutants in an area that is already beleaguered by air pollution and higher cancer rates. Memphis City Council members and other community leaders were not told about the project until it was announced, and xAI did not receive a permit to operate its gas-burning turbines for its first year in operation. In May 2025, the Southern Environmental Law Center reported that xAI may be adding additional turbines to a second facility it operates.

Ethics of Integrating X User Data Into xAI Models

xAI quietly integrated X user data into Grok’s AI training data in 2024, triggering concern from privacy watchdogs. When xAI acquired X nearly a year later, Musk announced the companies would fully combine data, AI models, staff and other resources. xAI has not shared any more specifics, but it’s safe to say that X user data belongs to xAI now. 

While it’s no longer shocking that a social media company would sell or repurpose user data, the use of social media posts for AI training is a new frontier that makes privacy advocates queasy. In Ireland, for example, the Data Protection Commission is investigating whether the move was in compliance with the European Union’s General Data Protection Regulation (GDPR) law.

Frequently Asked Questions