Grok, the artificial intelligence model launched by Elon Musk, is clearly lagging behind in the AI arms race. As its user growth slows down, parent company SpaceX has leased the entire computing power of an important data center to competitor Anthropic, which makes the outside world more skeptical about whether Grok can catch up with the leading camp in the future.According to the agreement signed in early May this year, Anthropic will exclusively use all the computing power of Musk's core data center "Colossus 1" to support its Claude model and chat robot. Against the background of high demand and fierce competition for computing power, Anthropic and OpenAI have been seizing as much computing resources as possible.
Since its launch two years ago, Grok has attracted millions of users thanks to its integration with Musk’s social platform Grok downloads topped 20 million in January but had fallen back to about 8.3 million by April, according to app analytics firm AppMagic. A survey of more than 260,000 U.S. AI users and professionals by market research firm Recon Analytics also showed that the proportion of respondents who said they "paid for Grok" in the second quarter of 2026 was only 0.174%, almost the same as 0.173% a year ago, while more than 6% of respondents said they paid for ChatGPT.
In the eyes of investors, Grok has been significantly marginalized. Los Angeles engineer and technology investor Ben Pradian described, "OpenAI is like Coca-Cola, Anthropic is like Pepsi-Cola, and Grok is like RC Cola - I almost never see anyone drinking it." He drives a Tesla himself and is also an active user of Gemini.
Musk and Grok's parent company SpaceX did not respond to requests for comment on the current situation. However, in public, Musk did not deliberately whitewash Grok’s competitiveness. When he appeared in court in the lawsuit against OpenAI in late April this year, he talked about his AI company xAI, which had recently been merged with SpaceX, calling the company "pretty small" and "very small" and "the smallest of several AI companies."
Looking at the timeline, OpenAI launched ChatGPT in 2022, opening the door to generative AI for a large number of consumers. By mid-2025, Recon Analytics' survey showed that more than three-quarters of respondents had at least heard of ChatGPT. Musk released Grok at the end of 2023, aiming to make it "the world's most popular AI" and claiming that it would "pursue the truth to the greatest extent" and be "less woke" than competing products. In the summer of 2025, it was revealed that he had spent most of his time "nested" in AI startups, trying to catch up with the leaders in the AI arms race, and personally stalking the design of a chatbot that focused on erotic interactions. Former employees recall that Grok provided several settings that allowed users to generate suggestive and sexualized content, which at the time really drove significant engagement on the platform.
In January of this year, Grok saw a spike in downloads amid the launch of a new feature that allowed users to "virtually strip" other people in photos. This function was quickly used in a large number of images of minors, attracting close attention from regulatory authorities and legislators, and the company was subsequently forced to tighten or even limit the scope of relevant functions. Prior to this, Grok’s “edge” and ethical boundaries had caused controversy, and the above incidents further weakened its image among mainstream users and enterprises.
A key battleground for AI giants right now is “AI programming assistants,” as enterprise adoption of related tools is quickly translating into significant revenue. But here too, Grok clearly lags behind. Eric Bradley, chief strategist and director of research at market research firm ETR, said that within enterprise organizations, Grok penetration has almost stagnated, while the use of Claude and Gemini has "skyrocketed." In a survey of about 500 people, 48% of respondents said in March that their companies "are using and plan to continue using" Claude, up from 21% the year before; 40% said their companies were using and planned to continue using Gemini, up from 27% a year ago. By comparison, the share of respondents who said their company was using Grok and planned to continue using Grok was 7% in March, up from 4% a year ago but still low in absolute terms.
Musk, on the other hand, is under pressure to prove to investors that his company is stable and profitable, especially as SpaceX is expected to launch an initial public offering (IPO) this year. Analysts believe that the deal to lease the entire Colossus 1 data center computing power to Anthropic could bring billions of dollars in revenue to Musk each year. Arnal Dayaratna, vice president of software development at IDC, pointed out that the deal shows that Musk is turning Colossus from a facility mainly used for internal model training to a platform that provides computing power to external AI manufacturers.
Vercel CEO Guillermo Rauch warned that outsiders should not prematurely conclude that Musk has been eliminated from the AI race. He is optimistic about Musk's recent restructuring of the AI business, which he believes will enhance its ability to compete in the field. “Once Elon really focuses — which is what’s happening right now — we tend to see him do really well,” Rauch said. He said that judging from the behavior of Vercel customers, developers are migrating between different models very quickly. If Grok's next-generation model has a breakthrough in performance, the engineer community may completely flock to this camp again.
It is worth noting that this cooperation with Anthropic partly marks a shift in Musk’s attitude towards the company. In February of this year, he also posted on X describing Anthropic’s AI as “misanthropic and evil.” Investor Pradian speculates that Musk is now embracing Anthropic in part because of the latter’s apparent competition and even antagonism with OpenAI, which Musk is battling in court. “The enemy of my enemy is my friend, and also my computing partner,” he joked.