TL;DR: Anthropic just signed a blockbuster deal with SpaceX to access all the compute at its Colossus 1 data center. Backed by this and a slew of other compute deals, Anthropic also raised limits for paying Claude users today. It’s a reminder that in the AI race, no beef survives a serious compute shortage.

What happened: The AI lab’s revenue and usage jumped 80-fold in Q1—eight times what it planned for, with CEO Dario Amodei saying "that is the reason we have had difficulties with compute." With Colossus 1, which is billed as the world’s largest supercomputer, Anthropic will have 300-plus megawatts of extra compute powered by over 220,000 Nvidia GPUs. This capacity comes online within the month.

The SpaceX deal is just the latest in Anthropic’s compute spree: It’s nabbed up to 5 gigawatts with Amazon, 5GW with Google and Broadcom, up to 1GW with Microsoft and Nvidia, and a $50 billion Fluidstack deal to build compute infrastructure in the US.

All this compute has allowed Anthropic to double Claude Code's five-hour rate limits for Pro, Max, Team, and Enterprise plans starting today. Peak-hour caps are now history for Pro and Max accounts, and API rate limits for Opus—Anthropic’s most powerful publicly available model—jump considerably for developers.

The demand problem: At Anthropic’s Code with Claude conference Wednesday, Amodei called the Q1 revenue and usage growth “just crazy” and “too hard to handle.” Users have definitely felt the compute crunch as a result, with Claude outages being more frequent and the company rolling out stricter usage quotas earlier this spring.

The unlikely landlord: At first glance, an Anthropic-SpaceX collab might look like it came out of left field. In February, Elon Musk declared on X that Anthropic “hates Western Civilization.” It was also, according to Musk, “Misanthropic.” Cut to last week, when he praised Anthropic’s senior team and noted that “no one set off my evil detector.” (It probably doesn’t hurt that Musk and Anthropic have a rival in common.)

But beyond any personal feelings, the deal may make financial sense: As Axios notes, Musk has historically over-built compute infrastructure relative to xAI’s demand, leaving capacity to lease out.

Bottom line: When you grow eight times faster than you expected, you buy compute from whoever has it—including the guy who called you civilization-hating just months ago. And if this growth rate keeps up, Anthropic may not be done yet.

What this means for you: Claude Pro, Max, Team, and Opus API users get fewer throttles and no peak-hour caps starting today. (See the full rate-limit table here.) —WK

Also at Anthropic…

• Per Bloomberg, Anthropic’s trying to improve Claude for everyday personal questions—whether around health, travel, or cooking—as it works to become more attractive to consumers, not just enterprise customers.
• Claude can dream now—the AI can autonomously review past sessions and update memory files to improve on its own.

Instagram's encrypted DMs are going away tomorrow—here's what to know

Starting tomorrow, May 8, Instagram is removing end-to-end encryption from direct messages. The feature has been opt-in since 2023 (and only available in certain regions), but it prevented even Meta from reading your conversations.

If you had E2E encryption turned on

You likely already know if you do, and Meta said it was prompting these users to download their encrypted message history (although it hasn't explained what happens to these chats after May 8, downloading now is the safest move).

• Look for a lock icon on any of your conversations. If you see one, check for an in-app prompt.
• You can also download your chats manually: Accounts Center → Your information and permissions → Export your information → log in → Create export → choose Instagram → Export to device → choose which information to export and a date range (“all time” is safest, but takes longer).

For everyone else

If you never had E2E encryption on, Meta has always been able to access your DMs—meaning it could potentially use them for things like content moderation, complying with law enforcement requests, or ad targeting. In December 2025, Meta started using interactions with its AI tools inside private DMs for ad targeting and personalizing content, so it's possible regular DMs could be used for this as well.

If you'd rather keep certain conversations or information off Meta's servers entirely, Signal is the most private option. WhatsApp keeps message content encrypted by default, but Meta still collects metadata from it—who you message, how often, device info—so it's not a complete alternative. —SM

If you have a tech tip or life hack you just can’t live without, fill out this form and you may see it featured in a future edition.

The following sentence isn’t the result of a Mad Libs prompt: A robot in South Korea has officially taken vows to become a Buddhist monk. The 4-foot-3-inch-tall Gabi—a Buddhist name meaning “mercy”—is the first android in the nation to be accepted into a Buddhist sect (surprisingly, it’s not the first in the world). Gabi isn’t a fully ordained monk yet, but the bot promised to “devote” itself to the religion. It even took a group photo with the other (human) monks, striking a heart pose with its metal arms.

Gabi’s just one example in a growing number of artificial religious figures—from a Christian robo-priest in Germany that looks like an angry ATM to GitaGPT, a chatbot that offers advice based on Hindu scripture. Gabi might even have a leg up on humans in following one of Buddhism’s central tenets: to let go of all worldly desires. And it could be a sign that AGE (artificial general enlightenment) is right around the corner. —WK

Chaos Brewing Meter: /5