Why AI companies are starting to build their own chips

Anthropic has officially confirmed it is building a custom silicon team. Silicon is the physical material used to make computer chips. The company plans to co-design its hardware and its AI models—the complex sets of instructions that allow AI to process information—to make them run faster. They are not alone. OpenAI is building its own chips, and companies like Google and Meta have been designing their own hardware for years. Meanwhile, smaller companies are exploring how to run AI locally on your device rather than in a distant server, a process known as on-device inference. This is the act of the AI actually doing its job and generating an answer, performed right on your phone or laptop instead of in the cloud.

Think of an AI model like a high-performance race car and the chips as the fuel and engine design. For years, these companies have used general-purpose chips designed for many different tasks. These are powerful, but they are not specialized. When you build a custom chip, you are essentially tailoring the engine to the specific race track the model runs on. By designing the model and the chip to work together, engineers can strip away unused pathways and focus all the chip’s energy on the specific math the AI needs to do. This makes the system more efficient, meaning it uses less electricity and creates less heat while getting the answer to you faster. For on-device AI, this is even more critical; a phone cannot handle the massive cooling systems found in a data center, so the chip has to be incredibly efficient to do the same work without draining your battery or turning your phone into a hand warmer.

We are moving past the early phase where any big computer will do. As AI becomes more common, the sheer volume of processing power required is becoming a bottleneck. Relying on someone else’s hardware is expensive and limits how much a company can scale its services to new users. By controlling their own hardware, these companies gain more independence and can better manage their own costs. For you, the user, this shift might mean that in the near future, the AI on your phone becomes much faster and more capable, even when you do not have an internet connection. We are moving toward a world where the intelligence is no longer just in the cloud, but built into the very foundation of the devices we carry every day.