Nvidia , a dominant player in the artificial intelligence computing market, may face increasing competition from custom chip designers in the near future, according to Morgan Stanley. A sharp rise in interest in AI-led technologies, sparked by the release of ChatGPT late last year, has led to a spike in demand for graphic processing unit semiconductors (GPUs). These chips, made almost exclusively by Nvidia today, are expensive to purchase and energy-intensive to operate. Nvidia’s new H100 GPUs, for example, priced at $30,000 per unit, consume nearly twice as much power as a chip made by Intel or AMD. “Budget costs and energy requirements are the two major limitations for future AI computing, in our view,” said Morgan Stanley analysts led by Charlie Chan in a note to clients on June 11. “We therefore expect to see increasingly energy-efficient and low-cost AI custom chip designs ahead, matching or even outpacing the growth of NVIDIA’s and AMD’s general purpose GPUs.” The investment bank estimates that the total AI computing semiconductors market size will be around $43 billion in 2023 and could nearly triple to $125 billion by 2027. Additionally, the bank said there could be room for even more growth if investments in AI significantly enhance enterprise productivity or if more consumer-oriented applications emerge. The firm predicted that these custom chips will take up to 30% of cloud-based AI semiconductor market share by 2027 as increased demand justifies the upfront cost of custom designs. The Wall Street bank named five chip designers that could benefit from the trend: Taiwan Semiconductor Manufacturing Company ( TSMC ), Alchip Technologies , Global Unichip and King Yuan Electronics (KYEC) in Taiwan, and Hong Kong-listed ASM Pacific . Morgan Stanley is “overweight” on the five stocks, and increased their price targets in the June 11 note. TSMC TSMC, which currently holds a 60% market share manufacturing advanced AI chips, is expected to play an essential role in enabling future growth within this sector. As TSMC expands its capacity over time, it could see up to 13% of revenue coming from manufacturing these new chips by 2027, according to Morgan Stanley analysts. Here’s what the investment bank said of the custom chip designers: Alchip Technologies Morgan Stanley raised its price target for Alchip to 1,800 Taiwan dollars ($58.49) from $1,480 Taiwan dollars. “We raise our intermediate growth rate assumption to 14% (from 13%) to reflect accelerated AI ASIC demand. As one of the few pure-play leading-edge design service houses, we believe Alchip (along with GUC) is a key enabler of future custom A.I. chip designs, and therefore a higher intermediate growth rate is justified.” Global Unichip The chip designer has been awarded a contract from Microsoft to work on its new 5nm custom AI chip, which could be deployed across the U.S. company’s cloud computing products. “According to our foundry supply chain checks, Chinese networking customers have turned more aggressive on order pull-ins, creating potential upside to 2023 earnings.” KYEC The bank raised its price target for KYEC to $60.00 from $49.00. “We raise our intermediate and terminal growth rates to 4.5% and 3.0% (from 4.0% and 2.5%) for KYEC given its strategic position of capturing growing AI testing demand as NVIDIA’s major final test provider.” ASM Pacific The investment bank raised its price target to 88 Hong Kong dollars ($11.23) from $66.00 Hong Kong dollars. ASM Pacific manufactures chips that connect GPUs to a server’s memory using specialized “packaging” and “bonding” technology.” “In our Greater China semi coverage, we identify ASM Pacific as a key proxy for advanced packaging equipment makers. We estimate that 5-10% of ASM Pacific’s 2023 revenue will be from [TCB bondinging technology], and the associated contribution will increase with growing demand for chiplet technologies, driven by the proliferating AI semi market.” — CNBC’s Michael Bloom contributed to this report.
