Anthropic is expanding its partnership with Google to gain access to up to one million of Google’s AI chips, known as TPUs, in a deal worth tens of billions of dollars. Starting in 2026, the company will receive more than one gigawatt of computing power to train future versions of its Claude AI models. Anthropic said it chose Google’s TPUs for their efficiency, strong price-to-performance ratio, and because it already has experience using them. The agreement reflects the intense competition and soaring global demand for AI computing power. Google will also provide additional cloud services, positioning its TPUs as an alternative to Nvidia’s in-demand chips.
At the same time, rival OpenAI has signed massive agreements that could exceed one trillion dollars to secure around 26 gigawatts of computing power and continues to rely heavily on Nvidia and AMD chips. Industry estimates suggest that one gigawatt of AI computing capacity can cost roughly $50 billion. Anthropic remains focused on AI safety and enterprise applications, and its Claude models have contributed to the rise of tools like the AI-assisted coding platform Cursor.
Credit : CGTN