ASIC or GPU for artificial intelligence? NVIDIA’s Jensen Huang says he’s not afraid of competition

During a roundtable discussion at GTC 2025, NVIDIA CEO Jensen Huang stated that specialized ASIC devices do not compete in the AI industry. But are GPUs really safe?

ASICs (Application-specific integrated circuits) have practically ousted video cards from the cryptocurrency mining market. These are chips designed for specific narrow tasks, may be cheapermore productive and energy efficient. So far, they do not «threaten the tasks of artificial intelligence, but this may change over time.

According to DigiTimes, Jensen Huang gave a rather vague answer to the question about possible competition with ASICs. He says that even if manufacturers manage to create their own ASICs, they will not have the necessary scalability for many reasons. However, when artificial intelligence becomes mainstream, the industry will definitely explore opportunities to get hardware capacity elsewhere than the de facto monopolist NVIDIA — not only because of the shortage, but also because of the cost of one-size-fits-all solutions.

The publication reports that large companies such as Google, OpenAI, Microsoft, and Broadcom are developing their own solutions optimized for specific AI workloads. As an example, Microsoft’s Maia-100 processor, which Microsoft showed back in 2023. However, it is not exactly an ASIC, «sharpened» for a specific job, but rather a programmable chip for a wider range of tasks.

Past ITC.ua wrote about the Sohu AI chip from the Etched startup. According to the company, the processor is claimed to be 10 times faster and cheaper than NVIDIA B200 chips. A single Sohu server is capable of processing Llama tokens 20 times faster than an H100 server and 10 times faster than a B200 server. 8 of these chips are supposedly capable of replacing 160 NVIDIA H100s. So, NVIDIA’s dominance does not seem so unquestionable in the future — creating more efficient ASICs is possible, albeit difficult.

Source: Wccftech