As of June 2023 there are chip shortages throughout the AI ecosystem, and NVIDIA cannot make enough chips.

There are two parts, as of June 2023: (a) AI chips were in short supply across the AI ecosystem; (b) NVIDIA could not make enough chips to meet demand. True: contemporaneous reporting (trade press, cloud providers, AI companies, NVIDIA's own earnings statements on supply constraints) shows widespread shortages of AI accelerators such as H100/A100 GPUs, affecting startups, cloud providers and large labs, with NVIDIA supply falling short of demand. Mostly true: both hold but the shortage was concentrated in high-end NVIDIA data-centre GPUs rather than 'throughout' the ecosystem. Mixed: only one part is supported. Mostly false: there were isolated or brief supply constraints, but supply broadly met demand. False: contemporaneous evidence shows no meaningful AI chip shortage and that NVIDIA was meeting demand.

Set before the research began.

1. E1 AI Capacity Constraints - CoWoS and HBM Supply Chain SemiAnalysis · 5 July 2023

   “GPU sales are going up like a rocket and the supply chain has struggled to keep up with the demand for GPUs. Companies are scrambling to get their hands on GPUs or cloud instances.”

2. E2 AI Capacity Constraints - CoWoS and HBM Supply Chain SemiAnalysis · 5 July 2023

   “The highest-end Nvidia GPU, H100, will remain sold out until Q1 of next year, despite Nvidia’s attempts to increase production drastically.”

3. E3 IFTLE 563: Is CoWoS Capacity Causing a GPU Shortage? - IMAPS 3D InCites Content Platform IMAPS 3D InCites · 24 July 2023

   “However, due to the severe shortage, Nvidia has started exploring options with its secondary supplier, placing orders with Amkor Technology and United Microelectronics (UMC).”

4. E4 AI GPU Bottleneck Explained: Causes and Prospects for Resolution TrendForce · 9 August 2023

   “Boyle clarified that the current GPU shortage is not a result of NVIDIA misjudging demand or constraints in Taiwan Semiconductor Manufacturing Company’s (TSMC) wafer production. The primary bottleneck for GPUs lies in the packaging process.”

SemiAnalysis (July 2023) reported that the supply chain had struggled to keep up with GPU demand and that companies were scrambling for GPUs or cloud instances (E1). It also said NVIDIA's H100 would stay sold out into the next year despite NVIDIA's efforts to sharply increase production (E2). Later reports describe a severe shortage that led NVIDIA to explore options with a secondary supplier, placing orders with Amkor Technology and UMC (E3), and a TrendForce report quoting Boyle on a current GPU shortage whose primary bottleneck lies in the packaging process (E4). All of this evidence comes from shortly after the June 12 date and focuses mostly on high-end NVIDIA data-centre GPUs and their packaging. That supports both parts of the claim, but it shows a shortage concentrated in those GPUs more than one running "throughout" the ecosystem.