Nvidia dominates the AI accelerator market. But is there room for others?

Nvidia (NASDAQ:NVDA) is likely to keep dominating the artificial intelligence accelerator market, but there is room for companies that make application specific integrated circuits as well, investment firm Jefferies said.

Analysts at the firm called out Marvell (NASDAQ:MRVL) as one beneficiary, given it is building the Trainium 2 chip for Amazon (AMZN), which is said to be “ramping well.” And Broadcom (NASDAQ:AVGO) and its custom-built business has made inroads at Google (GOOG) (GOOGL) and Meta (META) with their offerings: Google’s tensor processing units and the Meta Training and Inference Accelerator chip, respectively.

The Jefferies analysts said they expect the AI accelerator market to grow revenue at a 58% compound annual rate from 2023 to 2027, boosted by a 30% increase in volumes and a 22% increase in the average selling prices.

“Hyperscalers are accelerating [their] spend on GPUs but also their internal ASIC programs,” the analysts wrote. “We expect that these ramps will exceed current expectations, and provide solid upside to current estimates.” There is the most near-term upside for Nvidia and Marvell, but Broadcom should benefit later next year, the analysts added.

AMD (AMD) on the other hand, which recently reported third-quarter results, is something of a “show-me story,” the analysts said, though they are starting to see customers react positively on the AI front, including Meta and Microsoft (MSFT). AMD is now starting to produce “meaningful” volumes, the analysts added, as evidenced by the fact the Dr. Lisa Su-led company recently upped its revenue guidance for AI products to $5B for 2024.

Other beneficiaries

As the demand for AI increases, so too does the need for more memory.

Companies that produce high-bandwidth memory have benefited from the AI spending boom, and Jefferies research suggests that there is likely to be “continued” growth of 40% to 50% for HBM die growth, which should benefit companies like Camtek (NASDAQ:CAMT) and Onto Innovation (ONTO), the analysts said.

“This increasing [high-bandwidth memory] requirement will require continued capacity expansion at memory players with [high-bandwidth memory] die volumes growing at an estimated ~50% CAGR from 2024 to 2027, presenting a great growth outlook for [Camtek and Onto],” the analysts said.