NVIDIA’s AI GPU Demand Skyrockets as Orders Get Pushed to December
NVIDIA’s AI GPU sales are booming thanks to the expanding artificial intelligence market and the fast development of AI-based applications by major tech companies like Google, Microsoft, and OpenAI. It has been reported that because to the overwhelming demand, the shipment of AI GPUs has been postponed until December 2023.
According to MyDrivers (Citing Kuai Technology), NVIDIA’s A100 GPU has been steadily increasing in price since December of last year. Its total price rise over the previous five months reached 37.5% as of the beginning of April this year, while the total price increase of A800 GPU reached 20% over the same time period.
Additionally, the GPU supply cycle has become longer at the same time. As opposed to the old delivery cycle of about a month, the current one often takes three months or more. Deliveries on certain new orders “may not occur until December.”
At present, NVIDIA offers at least 9 distinct graphics cards for AI applications, 4 of which are high-performance versions (the V100, A800, A100, and H100).
According to a DigiTimes report, NVIDIA has placed additional orders with TSMC for chips that use the CoWoS packaging technology. CoWoS, or Chip-on-Wafer-on-Substrate, is a packaging technology used in NVIDIA’s top data center and cloud GPUs that make use of HBM memory.
TSMC’s monthly production capacity is between 8,000 and 9,000 CoWoS wafers, but NVIDIA has placed an order for an extra 10,000 wafers for the full year of 2023. This will need TSMC to increase production by between 1,000 and 2,000 wafers per month. TSMC’s confidence in the future of its CoWoS technology has been bolstered by this uptick in demand.
Massive demand is also coming from China, which is attempting to acquire NVIDIA’s newest AI GPUs despite being supplied a type that is pared down.