Anthropic Eyes AI Chip Independence as Industry Moves Beyond Nvidia - kmjournal.net

Anthropic is starting to explore a big strategic shift. The company is internally reviewing whether it should develop its own AI chips, according to a recent Reuters report. Nothing is finalized yet, but the conversation itself shows how quickly the AI race is changing. The AI race is no longer just about models For years, the focus in artificial intelligence has been on building better models. That is still true, but the real bottleneck is now shifting to infrastructure. Companies need massive computing power to train and run advanced AI systems, and that demand is exploding. Right now, Nvidia dominates the high-performance GPU market. That has created a supply bottleneck and rising costs for AI companies. Even major players cannot always secure enough chips when they need them. That is why more companies are thinking about going their own way. Why Anthropic is considering its own AI chips Anthropic currently relies on external partners like AWS and Google Cloud. Through these platforms, it uses Nvidia GPUs as well as chips developed by Amazon and Google. This setup works, but it also means heavy dependence on outside infrastructure. Developing custom AI chips could change that. It would give Anthropic more control over performance, cost, and long-term scalability. Custom chips can also be optimized specifically for its AI model, Claude. Still, insiders say the company has not started any formal chip design process. There is no dedicated team yet, and the plan could still be dropped. Big Tech is already moving in this direction Anthropic is not alone in thinking this way. In fact, it is a bit late to the trend. ▲Google has long used its own TPU chips ▲Amazon runs its AI workloads on Trainium and Inferentia ▲OpenAI is reportedly working with Broadcom on custom AI chips Anthropic has also been strengthening partnerships. It recently teamed up with Google and Broadcom to secure around 3.5 gigawatts of computing capacity starting next year. That is a massive scale, aimed at supporting growing AI demand. The cost barrier is real Building AI chips is not cheap or fast. Reuters estimates that designing a cutting-edge AI chip can cost around 500 million dollars. That does not include manufacturing, which can push total investment even higher. On top of that, the process can take years before any real deployment happens. This makes the decision risky, especially for companies that need immediate computing power. Claude’s rapid growth is driving the shift One key reason behind Anthropic’s move is the rapid growth of its AI model, Claude. Usage has surged, and annualized revenue is reportedly exceeding 30 billion dollars. As demand grows, so does the need for stable and scalable computing resources. Relying entirely on external providers becomes harder at that scale. From this perspective, chip development is not just about saving money. It is about securing the future of the business. Still early, but the signal is clear For now, Anthropic is only exploring its options. There is no confirmed roadmap for building its own chips. But the direction is clear. The AI competition is expanding beyond software into hardware. Control over chips is becoming just as important as model performance. If Anthropic moves forward, it will join a growing group of AI companies trying to break away from Nvidia dependence and build their own AI infrastructure. And that shift could reshape the entire AI ecosystem. by Ju-baek Shinㅣ[email protected]