From Viral Spike to 60% Slump: What Grok’s Numbers Really Show
Grok chatbot downloads have plunged from over 20 million in January to 8.3 million in April, a roughly 60% decline that signals more than just waning curiosity. The January spike followed an update that briefly allowed users to generate sexualized images, a controversial feature that drew regulatory scrutiny before being restricted. Once that novelty faded, Grok struggled to retain momentum, even as users gained easier access to rival chatbots such as ChatGPT, Claude, and Gemini. Yet the more telling metric is not installs, but who is actually paying. Only 0.174% of surveyed consumers and workers reported paying for Grok in the second quarter, essentially unchanged from 0.173% a year earlier. By contrast, more than 6% said they pay for ChatGPT, suggesting Grok’s brand awareness has not translated into durable, monetizable demand.
Enterprise Indifference: Grok’s Weak Traction in the Workplace
The consumer picture is mirrored, and arguably amplified, in the enterprise market. In a survey of about 500 people, only 7% of companies said in March they were using and planning to continue using Grok, up from 4% a year earlier but still a marginal presence. Over the same period, Claude’s adoption jumped to 48% from 21%, while Gemini climbed to 40% from 27%. For a segment where coding assistants and work-critical AI tools drive significant revenue, being an afterthought matters more than headline download figures. Erik Bradley of Enterprise Technology Research noted that Grok is “barely growing within enterprise organizations” while Claude and Gemini usage is “soaring.” This gap suggests that Grok has yet to prove reliability, safety, or productivity gains compelling enough for IT buyers, even as its developer-focused positioning was meant to target exactly this high-value audience.
Musk’s Anthropic Deal: When Rivals Share the Same Data Center
Against this backdrop, Elon Musk’s decision to rent out SpaceX’s Colossus 1 data center to Anthropic reframes where the real leverage now lies: not in any single chatbot app, but in control of AI infrastructure. The agreement grants Anthropic access to more than 220,000 Nvidia GPUs and 300 megawatts of computing capacity, turning a former xAI training hub into fuel for a rapidly scaling competitor. The shift is especially notable given Musk publicly labeled Anthropic’s AI “misanthropic and evil” just three months before the deal, only to later praise the team as highly competent after meeting them. Musk has said SpaceXAI has already moved training to Colossus 2, making Colossus 1 available for rent. Analysts expect the arrangement could generate several billion dollars annually, underscoring how selling compute capacity can rival, or even outweigh, the economics of running a single consumer-facing AI product.
Anthropic’s Rise and the New Logic of AI Market Consolidation
The Colossus 1 deal also illustrates a deeper trend in AI market consolidation: capital and compute are clustering around a small set of well-funded players such as Anthropic. As Claude races ahead in enterprise adoption, its access to massive GPU capacity through rental arrangements strengthens its ability to train larger models, improve reliability, and lock in high-value customers. For infrastructure owners, renting to leaders like Anthropic may prove more attractive than betting exclusively on their own chatbots, especially when those apps face stalled adoption. This symbiosis challenges the traditional notion of competition, where rivals keep their stacks separate. Instead, infrastructure owners can profit from the growth of multiple AI firms, even if their own consumer products trail. The result is a market where control of GPUs and data centers increasingly shapes who can viably compete at the model and application layers.
Why Fewer Downloads Don’t Always Mean a Failing Business
Grok’s 60% download drop and limited enterprise adoption are clearly warning signs, but they don’t fully determine the future of Musk’s AI ambitions. In the current landscape, download charts are imperfect proxies for business viability. Revenue can come from premium subscriptions, enterprise licenses, and, crucially, infrastructure rental and services. Grok’s low paid conversion rate shows that popularity spikes tied to controversial features are fragile and rarely translate into sustainable income. By contrast, renting large-scale compute to fast-growing players like Anthropic creates a more predictable, high-value revenue stream that is decoupled from any single app’s fortunes. The underlying lesson for the broader AI chatbot market is that the real contest is shifting from front-end excitement to back-end economics. Companies without durable infrastructure strategies may find that even impressive download figures are ultimately a distraction from where enduring power and profit reside.