AI roundup: Mythos shelved
A prominent AI roundup on X listed several major industry moves this weekend, saying Anthropic shelved its controversial “Mythos” model and Meta struck a roughly $21 billion deal with CoreWeave. The same thread reported OpenAI paused its UK “Stargate” deployment over energy costs, flagged delays for Nvidia’s Rubin GPUs, and noted new AI‑powered glasses from Snap and Qualcomm. (x.com)
Anthropic has limited access to its new Mythos model days after launch, while Meta signed a roughly $21 billion cloud deal with CoreWeave through 2032. (techcrunch.com) (coreweave.com) Anthropic published a system card for Claude Mythos Preview on April 7 and said the model showed unusually strong cybersecurity capability, including finding software exploits. The company’s safety team said it limited release rather than offering broad public access. (anthropic.com) (red.anthropic.com) CoreWeave said on April 9 that Meta expanded their agreement to about $21 billion in AI cloud capacity through December 2032. CoreWeave said the capacity will be spread across multiple sites and include some of the first deployments of Nvidia’s Vera Rubin platform. (coreweave.com) The common thread is compute: companies need more chips, more power, and more data center capacity to train models and run them for users. OpenAI’s Stargate project, announced in January 2025 as a plan to invest $500 billion over four years in United States AI infrastructure, put that buildout at the center of the industry race. (openai.com) That race is now running into physical limits outside the United States. Reuters reported on April 9 that OpenAI paused its main United Kingdom data center project over high energy costs and regulation, months after OpenAI announced Stargate UK with Nvidia and Nscale on September 16, 2025. (msn.com) (openai.com) The chip roadmap is under pressure too. Nvidia introduced Rubin in January and said in March that the platform was in full production, but supply-chain analysts this week said slower qualification of High Bandwidth Memory 4, new networking gear, and heavier cooling needs could push shipments back and reduce 2026 volumes. (nvidia.com) (nvidianews.nvidia.com) (trendforce.com) Consumer hardware is moving on a parallel track. Reuters reported on April 10 that Specs, Snap’s smart-glasses unit formed earlier this year, signed a multi-year deal to use Qualcomm Snapdragon XR chips in devices planned for later in 2026. (money.usnews.com) Meta, Snap, OpenAI, Anthropic, and Nvidia are all pointing at the same bottleneck from different angles: safer models need tighter release controls, and bigger products need more electricity, memory, and specialized cloud capacity. The next test is whether those constraints ease before the next wave of models and devices ships later in 2026. (anthropic.com) (coreweave.com) (trendforce.com)
