Anthropic Dispatch: SpaceX Gets Into AI Landlording and Your Rate Limits Doubled

The Part That Actually Matters To You Right Now

On May sixth, twenty twenty-six, Anthropic announced three changes that hit immediately. They doubled Claude Code's five-hour rate limits for Pro, Max, Team, and seat-based Enterprise. They removed the peak hours limit reduction for Pro and Max. And they raised API rate limits considerably for Opus models.

Let that sink in for a second. If you have been running Claude Code and hitting the wall mid-afternoon because everyone in San Francisco just woke up and started prompting, that wall is gone. Peak hours throttling is dead for Pro and Max. The five-hour cap doubled. If you are on the API side, your Opus rate limits went up across every tier.

This is the kind of announcement that changes your daily workflow more than any model upgrade. You can plan longer sessions. You can run more background agents. You can stop doing that weird dance where you switch to Sonnet because you burned through your Opus allocation by two PM. The practical impact of not being interrupted mid-task is enormous, especially if you have ADHD and getting back into context after a forced break costs you thirty minutes of momentum.

The SpaceX Deal

Now for the part that reads like science fiction but is apparently just a Tuesday in twenty twenty-six. Anthropic signed a deal to use all of the compute capacity at SpaceX's Colossus one data center. That is over three hundred megawatts of power and more than two hundred twenty thousand NVIDIA GPUs, coming online within the month.

Three hundred megawatts. To put that in perspective, a medium-sized Swedish town like Östersund uses roughly that much power total. Anthropic just added an entire town's worth of electricity dedicated to running AI inference and training. And this is just one of their compute deals. They also have up to five gigawatts coming from Amazon, another five gigawatts from Google and Broadcom starting twenty twenty-seven, thirty billion dollars of Azure capacity through Microsoft and NVIDIA, and a fifty billion dollar infrastructure investment with Fluidstack.

These numbers have gotten so large that they have stopped sounding like business announcements and started sounding like civilization-scale infrastructure projects. Which, to be fair, they might actually be.

The Orbital Compute Footnote

Buried at the very end of the announcement, almost as a throwaway line, Anthropic mentioned that they have expressed interest in partnering with SpaceX to develop multiple gigawatts of orbital AI compute capacity. Orbital. As in, in space. Data centers in orbit.

Now, "expressed interest" is corporate language for "we talked about it and neither side laughed the other out of the room." This is not a signed deal. This is not a timeline. This is two companies whose leaders probably had dinner and one of them said what if we put GPUs in space, and the other one said let me think about that, and somehow it ended up in a press release.

But the fact that it is in the press release at all tells you something about where these companies think compute is going. The constraint on AI is not algorithms anymore. It is not data. It is electricity and cooling. And if you are looking for places with unlimited cooling and no neighbors to complain about your power draw, low Earth orbit starts looking surprisingly practical. Absurd, but practical.

The Competitive Context

This announcement did not happen in a vacuum. Three minutes after Anthropic posted the rate limit increases, the developer community started doing the math against OpenAI's Codex. The phrase "on par with Codex" spread immediately. A week later, Anthropic bumped limits again with a fifty percent increase through July, and OpenAI responded the same day by shipping Codex on ChatGPT mobile.

This is an arms race, but instead of building bigger models, they are racing to remove the friction that makes developers switch tools. Rate limits, peak hour throttling, context windows. The actual day-to-day annoyances. For users, this is great. Competition for your daily workflow means the experience keeps improving and the limits keep going up.

The International Angle

One more detail worth catching. Anthropic mentioned that some of their capacity expansion will be international, specifically calling out regulated industries like financial services, healthcare, and government that need in-region infrastructure. They said they are partnering with democratic countries whose legal and regulatory frameworks support investments of this scale.

That is a pointed statement. It is Anthropic drawing a line about where they will and will not build infrastructure, and framing it as a democracy question rather than a market opportunity question. They also committed to covering consumer electricity price increases caused by their data centers in the US, and are exploring extending that to new countries.

For anyone running AI infrastructure in Europe or wondering when Claude gets proper EU-region inference, this is the signal that it is coming. Not tomorrow, but it is in the pipeline alongside the gigawatts.