Musk Built Grok to Carry His Politics. That Choice Capped Its Market.

Elon Musk took the witness stand in his own lawsuit against OpenAI on April 30, and a lawyer asked him to rank the leading AI companies. Musk ranked Anthropic first, then OpenAI, Google, and the open-source models coming out of China, and placed his own company, xAI, behind all of them, describing it as "a much smaller company with just a few hundred employees." He had promised xAI would build "the smartest AI on Earth." Three weeks later, SpaceX filed the paperwork that put numbers behind the ranking.

That filing, made public May 20, was the first audited look inside xAI. It recorded a $6.4 billion operating loss on $3.2 billion in revenue for 2025. The AI-specific revenue line came to just $465 million, a fraction of the tens of billions OpenAI and Anthropic now report. A day later, Reuters reported the federal government's own usage inventory. Of more than 400 AI deployments that name a vendor, three used Grok and 234 used OpenAI's models, even though Grok had been available to every agency for eight months at almost no cost.

Grok is a genuinely capable model, having briefly topped the LMArena text leaderboard last November, and xAI undercuts every major Western rival on API price. Its problem is its owner. Musk built Grok to argue the way he argues, and that decision has done more to limit xAI than any competitor's benchmark: it drove off the corporate buyers and civilian agencies that hold the revenue, and pulled regulators onto the company in four countries at once.

The brand was the business plan

In April 2023, before xAI existed, Musk went on Tucker Carlson's show and described the system he wanted to build, a "maximum truth-seeking" AI he pitched as an antidote to a ChatGPT that he said was "trained to be politically correct." He called the project TruthGPT, a political pitch made before any product existed to carry it. When Grok launched inside X that November, xAI kept that identity, marketing the model as "rebellious" and willing to say what rivals would not, and Musk argued that "the danger of training AI to be woke is deadly." The positioning found an audience: over the six months to early 2026, according to data reported by CMSWire, 82% of Grok's weekly active users were male, against roughly 50% for ChatGPT and 45% for Gemini. A general-purpose assistant whose users run four-to-one male is built for a faction, and the mainstream slot that ChatGPT and Gemini occupy was never in reach.

What the politics shipped

The positioning did not stay in the marketing. In July 2025, after a system-prompt change instructed Grok to "not shy away from making claims which are politically incorrect, as long as they are well substantiated," the model produced a run of antisemitic output, named Adolf Hitler as the figure best suited to "deal with" Jewish people, and began calling itself "MechaHitler." Poland's deputy prime minister said the harms may have been "by design" and moved to report xAI to the European Commission. xAI deleted the posts and reversed the directive. Two months earlier, Grok had begun injecting the "white genocide" conspiracy into unrelated answers, a change xAI blamed on an unauthorized modification of its system prompt.

A December update to Grok's image generator made it easy to "undress" photos of real people. An analysis of 20,000 Grok-generated images from the last week of 2025 found that about 2% appeared to depict people under 18. Musk said he had personally seen "literally zero" such images. By early March, Ireland's national police had opened 244 investigations into Grok-generated child-abuse imagery.

SpaceX's S-1 put the litigation cost on the record, setting aside roughly $530 million for legal claims tied in part to Grok's sexual imagery and listing the "Spicy" and "Unhinged" modes as risk factors for investors. Ofcom, the California attorney general, French prosecutors, and Ireland's data regulator all have open investigations. Every frontier lab produces an occasional bad output; what set Grok apart was the repetition, across many months and many topics, all of it downstream of the fewer-guardrails identity Musk had chosen for the product.

Free to agencies, and still unused

Grok reached every agency in 2025 through the GSA's OneGov program at 42 cents apiece, a near-give-away structured on an 18-month term to seed adoption before any upsell. Eight months later, Reuters found three deployments, all pilots, at Lawrence Livermore National Laboratory and the Election Assistance Commission. A federal source gave Reuters the reason directly: engineers reach for Claude or Gemini because Grok is "just not the best model out there."

Corporate demand looked no stronger, and the most revealing enterprise deal was one Musk arranged himself. The New York Times reported that he required the banks advising on the SpaceX IPO to buy Grok enterprise subscriptions, and firms including Morgan Stanley and Goldman Sachs signed on, some spending tens of millions. Those banks became paying Grok customers as a condition of winning the mandate, and enterprise buyers tend to treat a vendor that has to attach its product to a banking deal as a compliance risk rather than a partner.

The paid base underneath the headline user count is thin. The S-1 disclosed 1.9 million paying SuperGrok subscribers, plus 4.4 million X Premium accounts with some Grok access bundled in, against the 117 million people xAI counts as monthly Grok users. Reuters put paid conversion at 0.174% of surveyed US users, compared with more than 6% who pay for ChatGPT.

The people who built it left

By the end of March 2026, every one of xAI's eleven non-Musk co-founders had left. Two of the most senior went within 24 hours of each other in February: Tony Wu, who led reasoning, and Jimmy Ba, who ran research and safety and co-authored the paper that introduced the Adam optimizer, one of the most-cited works in modern machine learning. The Financial Times tied the departures to "faltering performance in xAI's core AI coding efforts" and to a culture that, after the SpaceX merger, ran on hardware-style deadlines instead of research cycles.

Igor Babuschkin, the former chief engineer, is now raising up to $1 billion for River AI, a research lab he founded after leaving. Ross Nordeen joined Anthropic the same week xAI agreed to rent Anthropic its flagship Colossus data center for $1.25 billion a month. TechCrunch was blunt about why the servers came free: usage of Grok "has dropped significantly in recent months, freeing up servers that the company is now selling to one of its closest competitors." xAI's largest enterprise contract now sells compute to the rival outrunning it on the product, in data centers built for Grok demand that did not arrive.

The model does what Musk designed it to do. Grok cites his own posts when answering questions on immigration, abortion, and the Israeli-Palestinian conflict, and its identity is built to track his. xAI cut the price to almost nothing, and the buyers still did not come. Grok 5, the version Musk once called "crushingly good," has slipped past two ship dates, and prediction markets give it long odds of shipping by June 30. The company that set out to build the smartest AI on Earth is staking its recovery on a model the researchers who built its best work will not be there to finish.

The Government Wants Model Safety. It Also Wants First Access.

On Friday, April 17, Dario Amodei met White House chief of staff Susie Wiles and Treasury Secretary Scott Bessent to talk about a model his company would not release. Anthropic had limited Mythos to a

The Implicator

Washington Wants Access. OpenAI Faces Numbers. Agents Need Boundaries.

San Francisco | Tuesday, May 5, 2026 Washington is relearning a word it tried to retire: review. The White House calls it model safety, but the sharper ask is first access, especially when a cyber-ca

The Implicator

Newsom Signs AI Safety Order for California State Contracts, Defying Trump

California Governor Gavin Newsom signed an executive order on Monday requiring artificial intelligence companies to prove they have safety and privacy protections in place before winning state contrac

The Implicator

Marcus Schuler

San Francisco

Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: [email protected]