Grok has been effectively unusable since Tuesday evening. It is now Thursday. That is three days of a flagship AI chatbot throwing "high demand" errors at nearly every user who tries to send a message. Free users, paid users, SuperGrok subscribers paying $30 a month for AI companion characters. All of them locked out.

The most insulting part? xAI's official status page at status.x.ai still shows every service as fully operational. Green across the board. Web, iOS, Android, API. All good. Nothing to see here.

Meanwhile, Downdetector is logging hundreds of reports per hour. PiunikaWeb, IBTimes, StatusGator, and multiple Reddit threads have documented the rolling outage in detail. More than 800 user reports hit just in the last 24 hours according to IBTimes Australia. The error has been confirmed across the US, UK, and Germany.

What Users Are Actually Seeing

The error comes in two flavors. The first is the classic "high demand" blocker that tells users the system is under heavy usage and suggests upgrading to SuperGrok for priority access. The second is vaguer: "Sorry about that, something didn't go as planned. Please try again." That one shows up inside the Companions tab when Ani, Rudi, or Valentine fail to respond.

The irony of the first message is obvious. It tells free users to pay for priority access. Paid users are seeing the same message. SuperGrok costs $30 a month and explicitly promises priority access during high-demand periods. That promise has been broken for 72 hours straight.

Some users have discovered workarounds. Accessing Grok through X directly (x.com/i/grok) instead of the standalone app occasionally gets through. The voice chat feature using the Expert model reportedly still works for some accounts. But standard text prompts? Practically dead.

Why This Is Happening (The Real Reason)

xAI has not confirmed a cause. But the timing is hard to ignore.

Three things landed in the same week. First, a major feature rollout: custom templates and smarter video extensions went live earlier this week. Big feature drops always cause a usage surge. Second, SpaceX announced its landmark deal with Cursor, the AI coding startup, giving SpaceX the option to acquire the company for $60 billion. Part of that deal involves Cursor tapping directly into xAI's Colossus supercomputer, pulling tens of thousands of GPUs to train its next-generation coding models. Third, the quiet rollout of the Grok 4.3 beta to select SuperGrok users.

If xAI is suddenly routing massive compute to fuel Cursor's model training while simultaneously pushing a new Grok version and handling a feature-launch traffic spike, it is not hard to see why regular users are getting squeezed. Compute is finite. Something has to give. And right now, the thing giving is everyone who actually uses Grok to chat.

This Is a Bigger Problem Than a Bad Week

The outage itself is bad. But the real story is what it reveals about xAI's priorities.

xAI is telling users their $30 SuperGrok plan buys priority access at the exact moment priority access does not work. The message on screen says pay to skip the queue. The queue is not skippable because the infrastructure is saturated. That is not a throttling decision. That is a capacity failure landing on the people who paid to avoid it.

Then there is the status page. Three days of widespread outage, hundreds of user reports per hour, multiple news outlets covering it, Reddit threads piling up, and the official status page shows green. No incident declared. No acknowledgment. No estimated time to resolution. If you cannot trust the platform to tell you when it is broken, the premium subscription loses its meaning entirely.

This is not Grok's first reliability crisis. January, March, and early April all had documented outages lasting anywhere from 30 minutes to several hours. But this one is the longest and most widespread. And it is landing during the same week that OpenAI shipped GPT-5.5 and Anthropic crossed $1 trillion in secondary market valuation. The contrast is brutal.

The Cursor Question

Here is the part nobody at xAI wants to talk about. Business Insider reported earlier this month that Cursor was already tapping into xAI's Colossus supercomputer for model training. Then SpaceX formalized the relationship with the $60 billion acquisition option on Monday. That deal explicitly involves joint development work on "next-generation coding and knowledge work AI."

Training cutting-edge coding models requires enormous compute. If Colossus GPUs that were serving Grok inference are now being allocated to Cursor training runs, the math is simple: same hardware, more demand, something breaks. And the something that broke is the consumer product that millions of people actually use.

Speculation? Partly. xAI has not confirmed anything. But the timing of the Cursor deal and the onset of the worst Grok outage in the platform's history is hard to call a coincidence.

What Happens Next

If you are a SuperGrok subscriber, your options are limited. Try off-peak hours (2 AM to 6 AM US Pacific). Try accessing through X directly instead of the app. Try voice chat. Or wait.

The broader question is whether xAI can serve two masters at once. Building a consumer chatbot and running a massive compute-for-hire business requires either enough hardware for both or a willingness to deprioritize one. Right now, the consumer product is losing that fight.

While Grok goes dark, OpenAI just launched its most ambitious model. Anthropic just became the most valuable private AI company on Earth. Google just unveiled eighth-generation TPUs. Every competitor shipped this week. xAI could not keep the lights on.

Three days of silence from the company that claims to be building the future of intelligence. The status page is green. The chatbot is dead. Those two facts cannot both be true.