From Skeptic to Supporter
Elon Musk, who previously expressed strong reservations about Anthropic, labeling them 'misanthropic' and questioning their motives, has undergone a notable
change of heart. Following a recent meeting with senior executives from the AI company, Musk conveyed a surprisingly positive impression. He stated on his social media platform that none of the individuals he met triggered his 'evil detector.' This apparent reversal coincides with a significant announcement: Anthropic will be utilizing the entirety of the Colossus 1 supercomputer, a substantial data center facility developed by Musk's xAI before its integration with SpaceX. This new arrangement grants Anthropic access to an impressive 300 megawatts of power and over 220,000 Nvidia GPUs, including advanced H100, H200, and GB200 chips. The integration of inference workloads onto this powerful cluster is slated to commence shortly, marking a pivotal moment in the relationship between these two prominent technology entities.
Massive Compute Power Deal
This substantial partnership involves Anthropic leasing the entire Colossus 1 supercomputer, a facility initially built by Musk's xAI. The agreement provides Anthropic with an immense capacity of over 300 megawatts, along with access to more than 220,000 Nvidia GPUs, encompassing the cutting-edge H100, H200, and GB200 chip architectures. This influx of computational resources is expected to significantly bolster Anthropic's AI model development and deployment capabilities. Musk's decision to lease this infrastructure is partly explained by xAI's own strategic shift to its newer Colossus 2 facility for its training operations. Furthermore, Musk has retained the prerogative to reclaim the compute resources if Anthropic's AI models were ever to exhibit harmful behavior, underscoring a cautious approach despite his newfound endorsement. The announcement was made by Anthropic's Chief Product Officer, Ami Vora, at their developer conference, highlighting the collaborative nature of this AI compute expansion.
Addressing Compute Demands
Anthropic has been navigating considerable challenges in meeting the escalating demand for its AI services, particularly for its Claude Code product. Users have frequently experienced rate limits and service interruptions, especially during peak usage periods. While Anthropic has secured other major compute deals with giants like Amazon (5 GW agreement), Google and Broadcom (5 GW partnership), and Microsoft ($30 billion in Azure capacity), these resources are not expected to become fully operational until late 2026 at the earliest. In stark contrast, the Colossus 1 supercomputer, now leased from SpaceX, offers immediate availability, with Claude inference capabilities set to go live within days. This swift deployment capability was a critical factor in Anthropic's decision-making process, making the SpaceX cluster an attractive and timely solution to their pressing compute needs.
Future in Orbit Compute
Beyond the immediate compute needs, the partnership between Anthropic and SpaceX extends to even more ambitious future endeavors, including discussions about establishing multi-gigawatt AI compute facilities in orbit. This forward-looking collaboration underscores the companies' shared vision for the future of AI infrastructure. Notably, SpaceX's absorption of xAI earlier this year has been officially cemented, with the combined entity now operating under the name 'SpaceXAI.' This merger signifies a consolidation of AI efforts within SpaceX, with xAI's distinct products being integrated into the parent company. The consolidated SpaceXAI is reportedly preparing for an initial public offering (IPO) in the near future, potentially as early as next month, which is anticipated to be one of the largest corporate offerings in history. This strategic alignment positions SpaceXAI as a dominant force in both space exploration and advanced AI computing.
