There's been a lot of discussion about the current stage of artificial intelligence: building infrastructure. This has obvious benefits for chipmakers like Nvidia and hyperscalers like Amazon, Microsoft, and Alphabet. Hyperscalers provide the enormous cloud computing power required for AI applications, and analysts predict a growing need for more data centers to house the massive computing power required for AI workloads. But technology analysts say the next bottleneck in AI infrastructure — and the bottleneck to invest in — is the network. A network in normal tech terms refers to a network of devices that can transmit and share information via physical or wireless communication. But in AI, the requirements are higher because of large language models and other AI applications that require very high bandwidth and low latency. “While Nvidia and its graphics processing units have been getting most of the headlines for generative artificial intelligence, we see the network as a critical companion to the hardware that underpins models and applications such as ChatGPT,” Morningstar analysts said in a June 2024 report. “Until now, it has been primarily a case of [graphics processing units]The actual AI chip. This is, of course, the most important piece of the puzzle. “But we see the network as the next bottleneck,” Liontrust Asset Management portfolio manager Claire Pleydell Bouverie told CNBC Pro Talks in May. That's because the “large-scale systems” coming to market, such as Nvidia's rack-scale systems, require “much more” infrastructure content, such as networks, she said. Morningstar technology equity analyst William D. Kerwin and technology equity strategist Brian Colello said they believe the need for high-speed networks in generative AI translates directly to “strong secular growth for well-positioned network vendors.” The firm said AI network spending will grow 34% over the next five years due to increased investments in training and inference of generative AI models. That would translate to $34 billion in spending in 2028, up from Morningstar's $8 billion estimate for 2023. “The network creates the performance bottleneck for generative AI model development,” Morningstar analysts said. “Advantageously positioned networking companies are the best secondary derivatives to invest in generative AI,” they added. “While the majority of generative AI spending goes to GPUs, networks are the key infrastructure that enables GPU performance.” Stocks to ride the trend Marvell Technology is Morningstar's top pick to ride the generative AI network trend, saying the company is currently “attractively undervalued,” giving investors an “immediate opportunity” to capitalize on rising generative AI network investments. Other potential winners in this network trend, according to Morningstar, are Arista Networks, Nvidia, and Broadcom. However, analysts believe the generative AI opportunity is “largely priced in” for these three stocks, as their stocks have already experienced a “robust” upside. “However, patient investors can wait for a pullback, as the long-term fundamental opportunity is strong,” Morningstar said. The company added that it is bullish on the adoption of Ethernet, a type of networking standard, in generative AI networks. Arista would be the biggest beneficiary of the move to Ethernet, according to Morningstar. The technology currently in common use is InfiniBand. On network infrastructure, Pleydell-Boubery added, “There are very few companies that can really get into providing this infrastructure.” He cited Meta and Broadcom as stocks that will ride this trend. Broadcom is a “leader” in networking chips and will benefit from the rise of Ethernet, he added. “Ethernet networking is becoming the de facto standard for scaling these AI workloads, and Broadcom has the best-in-class chips to underpin this Ethernet networking,” Pleydell-Boubery said.
