Anthropic weighs building its own AI chips amid crunch, Pg21

Key Highlights:

• Anthropic, an artificial intelligence lab, is considering designing its own AI chips due to a shortage of these chips.
• The company's revenue has increased to over $20 billion, up from $9 billion at the end of 2025, due to high demand for its AI model Claude.
• Anthropic has signed a long-term deal with Google and Broadcom to use Tensor Processing Units (TPUs).
• US Treasury Secretary and Federal Reserve Chair warned bank CEOs about cyber risks related to Anthropic's new Mythos model.

Detailed Insights:

• Anthropic currently uses chips from Google and Amazon to develop and run its AI software and chatbot Claude.
• Designing an advanced AI chip can cost around $500 million, requiring skilled engineers and a defect-free manufacturing process.
• Other tech companies like Meta and OpenAI are also exploring designing their own AI chips.
• Anthropic's Mythos model has the capability to identify and exploit weaknesses in major operating systems and web browsers, raising cybersecurity concerns.
• Anthropic has proactively briefed US government officials and industry stakeholders about the Mythos model's offensive and defensive cyber capabilities.

Key Concepts Involved:

• AI Chips: Specialized processors designed to accelerate artificial intelligence and machine learning tasks.
• Tensor Processing Units (TPUs): AI accelerators developed by Google for neural network machine learning.
• Cybersecurity Vulnerabilities: Weaknesses in a system that can be exploited to cause harm.