As the AI industry grows, so does the demand for chips used in model training. NVIDIA is the biggest player in this demand. The company, which has dozens of customers in countries from China to the US, wants to open a new chapter in semi-custom AI chips.
NVIDIA to develop semi-custom AI chip
NVIDIA wants to design a semi-custom AI chip to meet the specific AI requirements of cloud and data center customers. It turns out that the company has formed a new team to work on this chip.
NVIDIA currently makes AI and high-performance computing chips with its A100 and H100 series. But some large customers require customized AI chip designs to better fit their workloads and infrastructure needs.
The Semi-Custom Product Solutions unit established by the hardware giant will offer customized chips to customers. Artificial intelligence chips to be developed specifically for model training will be offered in close cooperation with customers.
These semi-custom chips will be optimized for partners’ specific AI and machine learning applications. Customers will be able to contribute to chip specifications and design while continuing to use NVIDIA’s technologies. The chips will also be integrated into third-party hardware as needed.
NVIDIA estimates that its semi-custom chip operations will address a $30 billion market. With this move, the company is enabling it to better serve the changing needs of large customers and partner with itself.