ChatGPT owner OpenAI is reportedly working on an AI model that combines 10 million NVIDIA GPUs.
Wang Xiaochuan, businessman and founder of Chinese search engine Sogou, told Wccftech that OpenAI is already working on a more advanced AI computing model that utlises more advanced training methods compared to the current iteration of ChatGPT.
Wang recently invested in Baichuan Intelligence, which aims to be a competitor of OpenAI in China. Baichuan recently released its Baichuan-13B language model that can run on consumer-level hardware like the NVIDIA GeForce RTX 3090 GPU.
The Sogou founder claimed that China needs an OpenAI alternative.
The 10 million GPU feat
Reaching 100,000 GPUs is a feat of its own, but this AI model could hold the capacity to connect 10 million AI GPUs together. NVIDIA is said to have already supplied OpenAI with 20,000 of its brand-new GPUs.
The semiconductor company is believed to be capable of producing a million AI GPUs per year. At this rate, supplying OpenAI with 10 million GPUs would take at least a decade.
NVIDIA has contracted Taiwan Semiconductor Manufacturing Co. to support and increase supply of its next-generation chips. However, reaching 10 million connected GPUs is a mammoth task, with challenges like cost, power, and finding appropriate infrastructure.
NVIDIA became the first chip manufacturer to reach a $1 trillion valuation in June 2023, a full $4 billion above analyst expectations, fuelled by the increasing demand for AI-driven servers. The company reportedly has 80% control of the GPU market, selling 30.3 million consumer GPUs in 2022.
With each of NVIDIA’s flagship A100 GPU for AI sold at $10,000 per chip, the semiconductor company is expected to make huge gains, especially if the 10 million GPU ambition comes to fruition.
Hungry for more tech news?
Sign up for your weekly tech briefings!