SAN FRANCISCO, May 7, 2026, 04:03 PDT
Anthropic struck a deal to tap every bit of available capacity at SpaceX’s Colossus 1 data center, looking to ease a crunch in computing power. CEO Dario Amodei recently acknowledged demand for Claude has blown past what the company expected. Under the new agreement, Anthropic expects to bring on board over 300 megawatts of capacity and more than 220,000 Nvidia GPUs within the next month.
The timing is critical here: “compute”—the actual chips, electricity, and data-center capacity that power and train AI models—has set a ceiling for the biggest AI labs. At a San Francisco developer conference, Amodei said Anthropic had braced for usage and revenue to jump by 10 times, maybe a bit more. Instead, it shot up 80 times, year over year, in the first quarter. He quipped the breakneck acceleration was “too hard to handle,” and that he’d have gladly settled for a “mere 10x.” Business Insider
Customers are already feeling the impact. Anthropic has bumped up Claude Code’s five-hour rate limits, scrapped peak-hour restrictions for Claude Code Pro and Max tiers, and lifted API caps for Claude Opus models. Rate limits—essentially, how much you’re allowed to use in a set period—aren’t as tight as before.
According to an xAI blog post, SpaceXAI’s Colossus 1 is built out with over 220,000 Nvidia GPUs—H100, H200, and GB200 models all in the mix. Anthropic plans to take advantage of the new muscle to boost Claude Pro and Claude Max. The post also mentions Anthropic’s interest in orbital AI compute—that’s data centers launched into space—but for now, that’s still just a concept, not a functioning system.
The arrangement hands SpaceX a high-profile AI client, while giving Anthropic an opening to ease the bottleneck around Claude Code—the coding assistant fueling demand among developers. “The fact that serious companies are even discussing compute capacity in space tells you how aggressively the market is searching for power and scale,” said Ryan Mallory, CEO of Flexential, the data-center operator, in comments to Reuters. Reuters
The rivalry with OpenAI is intensifying. Bloomberg Opinion’s Dave Lee says Anthropic’s capacity squeeze “cracks the door” for OpenAI, pointing out that critics have wondered if the restricted launch of Mythos—Anthropic’s new model—signals not just safety concerns but also weaker demand than advertised. Bloomberg
The SpaceX deal marks an unexpected shift in the rivalry. According to Axios, Musk’s xAI plans to stick with Colossus 2. PitchBook’s Harrison Rolfes pointed out that Colossus 1 was built with capacity Grok users never actually filled, leaving Musk with unused assets he can now monetize.
Investor pressure has entered the picture. Last week, Bloomberg said Anthropic was considering funding proposals that might lift its valuation past $900 billion—possibly overtaking OpenAI. The negotiations are still in early stages and nothing’s finalized. Reuters, for its part, noted Anthropic previously turned down overtures at around $800 billion or higher.
Anthropic moved further into agent software during the event, rolling out a “dreaming” research preview meant to help Claude systems revisit their work between sessions and refresh context files. Reuters previously reported on the feature, which highlights how coding is shaping up as a key battleground in the AI product race. Reuters
The SpaceX agreement hasn’t eliminated uncertainty. Money details aren’t public, and the future for orbital data centers is still a question mark. Residents near the Colossus site in Memphis have lodged complaints about gas turbine emissions, and there were protests close to a SpaceX investor gathering, WIRED noted.
Claude users get an immediate reality check: will those higher limits actually last when weekday demand spikes? Anthropic faces its own hurdle—can it keep scaling up chips and energy fast enough to prevent a capacity crunch that OpenAI might seize on as a selling point?
