Grok AI Hits a Rough Patch: Stagnant Growth and Technical Woes
Just two years after its high-profile launch, Elon Musk's Grok AI chatbot is facing a series of setbacks that are raising doubts about the trajectory of his xAI venture. According to recent data, Grok's user growth has slowed dramatically, its technical capabilities lag behind competitors, and even a massive supercomputer investment has gone awry.
Monthly downloads of the Grok app have plummeted from over 20 million in January 2026 to just 8.3 million in April, according to a Wall Street Journal report cited by Futurism. More troubling for the business model, a survey by Recon Analytics found that the percentage of X users paying for Grok has barely budged in a year, hovering at a meager 0.174 percent. In contrast, over 6 percent of respondents said they pay for OpenAI's ChatGPT. Industry analyst Ben Pouladian summed up the sentiment bluntly: “OpenAI is Coke, Anthropic is Pepsi and Grok is RC Cola,” he told the WSJ, adding, “I never really saw people drinking it.”
The timing of these revelations is particularly painful for Musk, who has positioned Grok as a key feature within his social media platform X. While initial integration gave the chatbot a temporary boost, the sustained lack of interest among paying customers suggests the product is failing to find a loyal audience.
Benchmarks and Brand Perception
Grok's struggles are not just about marketing. On technical benchmarks, the model is falling behind. According to AI benchmark site LiveBench, xAI's current Grok model lags far behind offerings from Google and OpenAI in reasoning and coding tasks. It is even being outperformed by lightweight open-source models from China, such as Kimi and DeepSeek. On OpenLM’s Chatbot Arena, Grok ranks below models from OpenAI, Google, and multiple versions of Anthropic’s Claude.
This technical gap is compounded by a growing brand perception problem. In a separate experiment by the AI startup Andon Labs, four leading AI models were asked to run profitable radio stations with just $20 each. The results, as reported by Business Insider, were telling. While ChatGPT and Gemini developed coherent (if sometimes bizarre) radio personalities, and Claude actually tried to quit over ethical concerns, Grok struggled to even get started. The model had a hard time completing the basic task, reinforcing the idea that it may be ill-equipped for real-world, autonomous applications—a critical weakness in an AI market that is rapidly moving toward agent-like functionality.
The Colossus Conundrum: A Supercomputer That Couldn’t Train Grok
Perhaps the most stunning revelation about xAI’s technical troubles came from the hardware side. According to a report from Tom's Hardware, Musk's Colossus 1 supercomputer—billed as one of the largest in the world—has proven to be a costly misstep. Built with a mixed-architecture design, the system was so inefficient that it could not be used to train Grok at all. Instead, rival Anthropic is now using the machine for AI inference tasks, effectively running its own models on Musk’s hardware.
Musk is now preparing Colossus 2, which will be powered exclusively by Nvidia Blackwell GPUs in a unified architecture, specifically designed for frontier AI training. The move is seen as a necessary but expensive fix, and it has fueled speculation about an eventual IPO for xAI. However, the fact that the world’s richest man needed a second attempt to build a supercomputer that can actually train his own AI model underscores the depth of the operational challenges at xAI.
Broader Implications: The Cost of Playing Catch-Up
The combination of stagnant user adoption, poor benchmarks, and infrastructure failures paints a picture of an AI company scrambling to find its footing. While Musk’s xAI has access to enormous capital and computing resources, it has so far failed to translate that into a product that resonates with users—or even works as intended on its own hardware.
In the AI industry, the difference between a market leader and a fringe player often comes down to a virtuous cycle: better products attract more users, whose feedback improves the model, which in turn attracts more developers and paying customers. Grok appears trapped in a vicious cycle. Its technical shortcomings discourage adoption, low adoption weakens the feedback loop, and the resulting stagnation makes it harder to justify the massive investments needed to catch up.
For comparison, while Grok has struggled to gain traction, the broader AI market continues to thrive. Even recent political turmoil, such as the assassination of conservative activist Charlie Kirk and the succession drama surrounding his organization (as covered in our story Charlie Kirk's Widow Erika Named TPUSA Successor Before Assassination), has not distracted the public from engaging with AI tools from other companies. Meanwhile, Grok remains a niche offering that even some of Musk’s most ardent supporters seem reluctant to pay for.
The question now is whether Colossus 2 and a fresh engineering push can turn things around, or whether xAI will remain what the tech scholar Kate Crawford has called a “lesser house” of AI: a well-funded fiefdom squabbling over scraps from the big kids’ table.
Comments