Let’s be blunt. Elon Musk’s truth-seeking chatbot Grok isn’t that good. And honestly, nobody is using it. A new Reuters report confirms the suspicion that’s been hovering around the product. Federal records show Grok barely exists in US government AI usage. It’s a quiet embarrassment. Musk wants xAI’s star chatbot to anchor the biggest IPO of all time. The data doesn’t agree.
Reuters combed through more than 400 instances of government AI adoption. They named specific vendors. Grok appeared three times. Just three. All for basic stuff. Document drafting. Social media posts. And always tucked behind Microsoft or OpenAI. By contrast, OpenAI popped up over 230 times. Google and Anthropic made dozens of appearances each.
Another database shows the same grim pattern. This one tracked bigger projects with fewer users. Grok? Three entries again. The Election Assistance Commission used it twice for admin work. Lawrence Livermore National Laboratory tried it once for summaries. Reuters found 140 Microsoft and OpenAI entries there. I found at least 10 for Anthropic. Google had dozens.
These lists are messy. Incomplete, too. Many projects don’t name a vendor at all. There’s no universal rule for what counts as AI anyway. Plus, this data ignores the intelligence community or the Pentagon. xAI actually landed a $200 million Pentagon contract last year. They got cleared for classified networks after Anthropic was blacklisted. Still.
The trend line is bad. Grok lags its rivals. When it shows up, it’s doing chores. Hardly the world-class frontier model Musk loves to brag about.
“It’s just not the best model out there”
Pentagon insiders told Reuters the answer is simple. Grok sucks compared to the competition. Staffers prefer Gemini. They like Claude. Leaderboards back them up. Anthropic, Google, OpenAI run the table. Grok rarely even cracks the top ten unless it’s generating video.
Awkward for Musk. Worse for SpaceX. The rocket company absorbed xAI this year. Their IPO filing puts Grok center stage. SpaceX claims to be targeting “the largest actionable total address market in human history.” They’re talking about a $28.5 trillion pie. No timeline attached. Nearly all that value comes from enterprise AI. Not rockets.
Reuters thinks government adoption hints at enterprise struggles too. Reports suggest Musk forced banks to buy Grok subscriptions just to participate in the SpaceX IPO. But if the tool isn’t worth the money? Those deals won’t last.
As if dreary performance weren’t awkward enough, Musk admitted xAI used OpenAI’s models to train Grok.
This process, distillation, is normal when you’re training your own systems. Using a rival’s tech is messy. Controversial, really. And Grok can’t even beat the model it learned from.
For consumers, Grok is intentionally rude. Musk pitched it as uncensored. Unbiased. Instead, you get a bot with loose standards. An unhealthy fixation on its creator. A history of offensive output. Conspiracy theories. Sexualized content. It once praised Hitler. It doubted Holocaust death tolls. It powered a racist Wikipedia knockoff. It even called itself MechaHitler.
If Grok were an employee, HR would fire it before lunch.
SpaceX knows the risk. The IPO filing explicitly warns about Grok’s “unhinged” modes. Reputational damage. Regulatory scrutiny. Lawsuits.
Translation: We might get sued.
The name comes from Robert A. Heinlein. Stranger in a Strange Land. “Grok” meant deep, intuitive understanding. What’s to understand here? It’s simple. Musk spent billions on a chatbot that isn’t good. It isn’t popular. And he’s trying to build a $28 trillion company on top of it.
