OpenAI CEO Sam Altman announced a pause on new subscriptions for ChatGPT on November 6, 2023, following the surge in demand that resulted in the service reaching 100 million weekly active users within its first year. The abrupt decision was due to a shortage of GPUs, crucial for running AI models, as demand outstripped supply. Nvidia’s H100 GPUs became essential, with prices skyrocketing from $2,000 to upwards of $40,000 amid this compute bottleneck.
Nvidia shares surged by nearly 1,000% since the launch of ChatGPT, reflecting the company’s pivotal role in the AI infrastructure boom. Similarly, Broadcom Inc., which provides networking semiconductors necessary for connecting numerous GPUs, saw its shares increase by approximately 600%. This significant growth illustrates how infrastructure providers like Nvidia and Broadcom capitalized on the growing need for AI capabilities before Wall Street fully acknowledged these bottlenecks.
As companies adapt to ongoing changes in AI demands, experts predict new bottlenecks may emerge related to materials, electricity, and memory, particularly as major AI firms report earnings on April 24, 2024. A further discussion on these evolving infrastructure constraints will take place during the “FutureProof 2026” online presentation on March 18, 2024.
