Nvidia is the maker of the GPUs used to train the most powerful AI models today. Having more of these chips gives you a head start over your competitors. Not having a good supply will leave you behind.
SemiAnalysis analysts Dylan Patel and Daniel Nishball documented the situation in a report last week, dividing the industry into “GPU poor” and “GPU rich” groups. The poor group, which is largely comprised of startups and open-source experts, is struggling with limited GPU supply. From European startups and the Jules Verne supercomputer to well-known AI companies like Hugging Face, Databricks, and Together, all fall into the poor group.
Meanwhile, the richer group includes a number of companies that have more than 20,000 A100 and/or H100 chips from Nvidia. They include OpenAI, Google, Anthropic, Inflection, X (formerly Twitter), and Meta. Several of these and a handful of Chinese companies have ordered more than 100,000 GPUs by the end of 2024.
According to SemiAnalysis, Meta ranks second in the world based on the number of H100 chips it has, behind only Google. The analyst firm calls Google “the richest computing company in the world” with an unbeatable architecture. Google will soon unveil an AI model called Gemini and is already training its next version, threatening OpenAI’s GPT.
(According to Insider)
Source
Comment (0)