NVIDIA AI GPU supply shortage is about to end, Delivery time shortened to 8 to 12 weeks

Since last year, Taiwan Semiconductor Manufacturing Company (TSMC), responsible for the manufacture and packaging of NVIDIA’s AI chips, has experienced a tight capacity in advanced packaging, leading to a persistent shortfall in the supply of AI GPUs. This shortfall has consequently resulted in extended delivery cycles, with servers built around the H100 architecture requiring anywhere from 36 to 52 weeks for delivery.

According to a report by Wccftech, citing a spokesperson from Dell, there has been a significant reduction in the delivery time for AI servers to customers, with the cycle now reduced by at least half—from 40 weeks to between 8 and 12 weeks. Dell remains optimistic about future supply capabilities. The company is notably significant in this sector, accounting for approximately 24.5% of the mainstream AI server market—a near quarter share that is on an upward trend.

Dell attributes this improvement to NVIDIA’s multifaceted efforts over the past few months, including the addition of new partners and the diversification of its supply chain, particularly in advanced packaging. TSMC has significantly increased its capacity for CoWoS packaging, planning to double this capacity within the year. Additionally, NVIDIA has recently placed orders with Samsung to meet the rising demand for advanced packaging capacities.

Last month, at the GTC 2024 conference held at the San Jose Convention Center in California, NVIDIA’s founder and CEO, Jensen Huang, introduced the Blackwell architecture GPUs, including the B200 GPU set to replace the H100/H200, as well as the GB200, which integrates with the Grace CPU. Industry concerns that the arrival of these new products might exacerbate the strain on advanced packaging capacities have been momentarily alleviated by the current supply situation.