Musk’s Compute Deal Raises Anthropic Concerns

Carmen Arroyo / Bloomberg : Sources: Apollo Global, Morgan Stanley, and others are testing Grok internally as part of xAI's finance sector push, but financiers are rarely using it for work — Elon Musk's xAI has recruited multiple Wall Street firms with ties to the billionaire's business empire to test its Grok chatbot …

xAI, Elon Musk’s AI company, has recruited several Wall Street firms, including Apollo Global Management and Morgan Stanley, to pilot its Grok chatbot as part of efforts to boost revenue ahead of parent company SpaceX’s potential IPO. The banks are testing Grok alongside other AI models internally, though reports indicate limited adoption by finance professionals in day-to-day work so far. The move aims to demonstrate commercial traction for xAI and generate pre-IPO revenue opportunities for SpaceX. Partnering with established financial institutions could validate Grok’s utility in enterprise workflows and help xAI scale commercial use cases, but actual usage and impact on revenue remain uncertain.

马斯克的xAI正争分夺秒地争取华尔街公司采用Grok聊天机器人

Elon Musk's Grok Is Losing Ground in AI Race

xAI’s Grok and its community discussions are being examined in a subjective, early-morning essay that probes the company’s stated mission of objectivity and restraint. The piece reflects on Grok’s positioning as a nonpartisan chatbot, user expectations around transparency and answers, and the interplay between human oversight and machine output. It highlights tensions between offering useful, filtered responses and avoiding overt intervention or ideological slant, noting how community feedback (e.g., Reddit threads) shapes perceptions. This matters for AI developers, platforms, and policymakers because expectations about neutrality, content filtering, and alignment affect trust, adoption, and regulation of conversational AI.

xAI lost senior pretraining lead Jun-tang Zhuang, who announced his departure after two years leading Grok model pretraining, voice models for X and Tesla, and xAI’s enterprise API models. Zhuang previously spent two years at OpenAI contributing to GPT-4’s technical report, GPT-4o and DALL·E 3 development, GPT4-Turbo 128k algorithm work, and building OpenAI’s embedding model; he also authored the AdaBelief optimizer in academic research. His exit comes amid a wave of departures after xAI was folded into SpaceX as “SpaceXAI,” with reported staff turnover exceeding 80 people in a year. Despite personnel churn, Elon Musk says Grok continues development on Colossus 2 and recently gained connector features for productivity apps.

解散xAI ，马斯克想用造火箭的方法，重造AI公司

The article argues that dedicated AI hardware is becoming the definitive competitive moat for companies in the generative-AI race, spurred by recent moves from Anthropic and xAI. It contends that while open-source models will grow, access to custom accelerators and large-scale infrastructure — not just model code — dictates performance, cost, and product differentiation. The piece weighs whether organizations should buy new hardware now or wait, highlighting trade-offs: immediate investment captures short-term gains and control, while waiting may yield cheaper, more efficient chips and better software-hardware stacks. It matters because procurement timing affects startups and enterprises' competitiveness, operational costs, and ability to leverage state-of-the-art AI services.

Anthropic has struck a deal to lease all capacity of xAI/SpaceX’s Colossus 1 data center to run Claude, a move that alleviates Anthropic’s compute constraints but raises environmental and reputational concerns because Colossus 1 has a troubled pollution record. xAI will retain its larger Colossus 2 for Grok development, and Elon Musk framed the deal as providing compute to AI firms that "do the right steps" while reserving the right to reclaim resources if AI harms humanity—introducing potential supply-chain and vendor-risk for Anthropic. The deal also followed xAI’s abrupt deprecation of Grok 4.1 Fast, which frustrated customers relying on that model. This matters for AI infrastructure, ethics, and platform dependency.

@macrotradecn: 马斯克把xAI的团队拆了，xAI并入SpaceX，同时将超算集群租给Anthropic

Anthropic has leased the entire Colossus 1 data center from SpaceX/xAI to ease compute constraints, while xAI retains the larger Colossus 2 for its own work. The deal drew criticism because Colossus 1 has a fraught environmental record—gas turbines initially ran without proper permits and have been linked to local air-quality health impacts—prompting pushback from advocates like Andy Masley. The timing also followed xAI’s abrupt deprecation notice for Grok 4.1 Fast, which angered customers. Elon Musk framed the lease as support for companies “doing the right steps” on AI safety and signaled a right to reclaim compute if an AI “harms humanity,” highlighting a new operational and supply-chain risk for Anthropic. The move matters for AI infrastructure, ethics, and public perception.

SpaceX’s IPO filing reveals governance terms that would give Elon Musk near-total control if the rocket company goes public, combining supervoting shares, mandatory arbitration, limits on shareholder proposals and Texas corporate law to constrain legal challenges and votes. The filing says only Musk can remove him as CEO, effectively locking in executive authority and reducing typical investor protections. While governance experts and some investors warn these provisions are unprecedented and risky, many institutional buyers may accept restrictions to access what could be a massive IPO—potentially up to $75 billion in proceeds and a $1.75 trillion valuation—fearing they’ll miss out on outsized returns. Critics warn it could set a template for future founder-led IPOs.

Simon Willison / Simon Willison's Weblog : While Anthropic will use the Colossus 1 data center, which has a really bad environmental record, xAI retains the larger Colossus 2 for its own AI training — There weren't a lot of big new announcements from Anthropic at yesterday's Code w/ Claude event, but the biggest by far …

Elon Musk / @elonmusk : Elon Musk says SpaceX reserves “the right to reclaim the compute” from Anthropic if its “AI engages in actions that harm humanity” — @MobofJoggers @nottombrown Just as SpaceX launches hundreds of satellites for competitors with fair terms and pricing, we will provide compute to AI companies that are taking the right steps to ensure it is good for humanity. We reserve the right to reclaim the compute if their AI engages in actions that

Anthropic has struck a deal to use all capacity of xAI/SpaceX’s Colossus 1 data center to run Claude, a move that eases Anthropic’s compute bottleneck but raises environmental and reputational concerns. Colossus 1 has a controversial record—its gas turbines once operated without proper Clean Air Act permits and have been linked to local air-quality health impacts—prompting criticism from observers like Andy Masley. xAI will retain its larger Colossus 2 for in-house work, so this isn’t a full exit from Grok development, though xAI recently deprecated Grok 4.1 Fast with short notice, which angered customers. Elon Musk framed the deal as conditional support for companies acting “good for humanity,” implying a new supply-chain risk tied to his discretion.