The deal was announced the same morning as Code with Claude 2026, and the two announcements are connected. On stage that afternoon, Dario Amodei explained why the compute was needed: Anthropic had planned for 10x growth per year, but Q1 2026 came in at 80x annualised. “That is the reason we have had difficulties with compute,” he told the conference audience. Pro and Max subscribers had been hitting throttles. Claude Code users were running into peak-hour rate reductions that capped their sessions. The product was outrunning the infrastructure.
Colossus 1 is the data centre that gave xAI its competitive opening. Built in 122 days in 2024, it hit 100,000 Nvidia H100 GPUs in record time and was Musk’s standing proof that xAI could compete with OpenAI and Anthropic on raw scale. By early 2026, expansions had pushed it past 220,000 accelerators with next-generation GB200 systems coming online. It also became Musk’s biggest political liability in Memphis. xAI’s subsidiary installed dozens of natural-gas-burning turbines to power the facility without federal permits, arguing they were temporary. Local residents in predominantly Black neighbourhoods documented worsening air quality and organised sustained protests. The Information reported that xAI was achieving only 11% GPU utilisation at Colossus 1. With xAI’s training workloads migrating to Colossus 2, the economics of leasing the first facility to a paying tenant were straightforward.
The reversal in tone from Musk was striking. In February, he had called Claude “misanthropic and evil.” By May, after what he described as a week of meetings with Anthropic’s senior leadership, he wrote on X that he “was impressed” and that “no one set off my evil detector.” He added that SpaceX reserved the right to reclaim capacity if Anthropic’s AI “engages in actions that harm humanity.” It is possible to read this as a genuine shift. It is also possible to read it as a founder who needed a tenant for a data centre burning cash in a politically hostile environment.
Anthropic passed the benefit to users immediately. Claude Code’s five-hour rate limits doubled for all paid tiers. Peak-hour limit reductions for Pro and Max were removed entirely. API rate limits for Opus models increased sharply. These were the most tangible user-facing improvements from any Anthropic infrastructure deal to date, precisely because the GPUs were available now rather than in 2027.
The orbital compute detail is easy to dismiss as a press release flourish, but both companies have reasons to take it seriously. SpaceX is preparing for an IPO and needs to demonstrate revenue lines beyond launch services and Starlink. Anthropic has said future infrastructure deployments will target politically stable, democratic countries with secure AI supply chains while exploring ways to offset electricity consumption. Whether space-based compute is practical at scale remains an open question. That both companies were willing to put it in the announcement suggests the conversations are more than hypothetical. Colossus 1 solves a problem that exists today. The orbital ambitions are about the problem Anthropic expects to have in three years, when 80x growth rates and multi-gigawatt terrestrial facilities start running into the limits of where you can put them on the ground.