“GPU” does not stand for “growth potential unlimited,” but it is starting to feel like that is what it should stand for. AI training needs have caused demand for graphic processing units used for AI acceleration to launch skyward, and sent the stock value of market share leader Nvidia soaring. There may even be room aboard the rocket ship for AMD after the company recently said GPU sales helped boost its second quarter revenue.
How fast is the market growing? One indicator can be found in a statement from Dell’Oro Group this week, which suggested that “accelerator revenues,” a segment consisting mostly of GPUs and custom accelerators, are expected to increase at a 38% compound annual growth rate for the next five years. That growth engine will help the worldwide Data Center IT Semiconductors market, which includes the major components for servers and storage systems, to reach $390 billion by 2028, according to Dell’Oro, which included these figures in its Data Center IT Semiconductors and Components 5-Year July 2024 Forecast Report.
As stock markets tumbled sharply early this week, Dell’Oro’s forecast, along with a different market report that showed a jump in chip sales worldwide during June, served as a reminder that AI accelerators and other chips have a lot of runway ahead of them.
“We recently raised our revenue forecast on accelerators because AI-related data center capex for hyperscale Cloud SP and Enterprise markets has surged since last year, and is expected to remain elevated during our forecast horizon,” said Baron Fung, Senior Research Director at Dell’Oro Group, in a statement. “In addition to AI, workloads such as video transcoding, HPC, and inline storage compression and decompression will also benefit from domain-specific accelerators.”
Dell’Oro, like many other research firms, currently sees Nvidia as having a dominant share of the GPU accelerator market, but Fung also suggested that other vendors, including AMD and Intel–as well as hyperscalers, via their own custom accelerators–will gain some share over time as buyers look to diversify their supplier ranks. There also has been a buzz around a group of prospective Nvidia challengers, including Cerebras Systems.
Dell’Oro also said in the report that the CPU market will lag that of other component categories in terms of growth, given the focus on optimizing the footprint of general-purpose servers. Meanwhile, the market for high-bandwidth memory (HBM) and other memory components also is expected to benefit from growing AI infrastructure needs. Storage components also eventually could benefit, given the vast amount of data that needs to be harvested and retained to train data models, Dell’Oro said.