Broadcom Expands Chip Deals with Google, Anthropic Amid AI Demand; Anthropic Unveils New Model

Anthropic has significantly expanded its computing deal with Google and Broadcom for multiple gigawatts of TPU capacity, with new capacity set to come online starting in 2027. This expansion comes amid skyrocketing demand for AI compute power, as reported by TechCrunch AI here and detailed by CNBC Tech here. The enhanced agreement aims to secure the substantial computing resources Anthropic needs to further develop and improve its advanced AI models, including its flagship model, Claude.

In a significant parallel development, Anthropic has also debuted a preview of its powerful new AI model, Mythos, as part of a new cybersecurity initiative. This move, reported by TechCrunch AI here, is part of a broader effort, detailed by Wired AI here, where Anthropic is teaming up with rivals to address AI security concerns, including the potential for AI to be used in hacking. This initiative, dubbed Project Glasswing, aims to proactively defend against malicious uses of AI.

This substantial investment in infrastructure underscores the intense competition and rapid growth within the AI sector. In a related development, Uber has reportedly chosen Amazon's AI chips, highlighting the diverse strategies companies are employing to secure necessary hardware, as previously covered by TechCrunch AI here. The partnership between Anthropic and Google is expected to have a significant impact on the AI industry, particularly in the development of large language models. With the increased computing capacity, Anthropic will be able to train and deploy more advanced AI models, potentially leading to breakthroughs in areas such as natural language processing and machine learning.

Beyond infrastructure and model development, Anthropic, alongside Google and OpenAI, is also collaborating to combat the unauthorized copying of their AI models by competitors, particularly those in China, as previously reported by The Decoder here. This initiative aims to protect the intellectual property of these leading AI firms and prevent the misuse of their cutting-edge technologies.

In parallel, OpenAI has been active on multiple fronts. The company recently announced the 'OpenAI Safety Fellowship' to bolster its safety research efforts here. However, internal challenges persist, with recent reports suggesting that a 'safety brain drain' at OpenAI may be attributed to factors as subjective as 'Sam Altman's vibes,' according to The Decoder here. Furthermore, OpenAI is reportedly seeking investigations into Elon Musk's alleged 'anti-competitive behavior' from California and Delaware attorneys general, ahead of a significant April trial, as detailed by CNBC Tech here.

OpenAI has also articulated a broader vision for the future of AI, proposing concepts such as public wealth funds, robot taxes, and a four-day workweek to manage the economic and societal impacts of artificial intelligence, as explored by TechCrunch AI here. In other Google AI news, Google Maps can now write captions for photos using AI here, and the company quietly launched an AI dictation app that works offline here. Meanwhile, General's GEN-1 robotics model has achieved 99% reliability in tasks ranging from folding boxes to fixing vacuums, as reported by Ars Technica AI here.