However, the manager cautioned that a closer look at the generative artificial intelligence (AI) infrastructure theme reveals that markets are still in early days of understanding its potential, pointing to the fact that less than 2 per cent of the 12 million servers sold in 2023 were AI enabled.
“A snapshot of the global server environment shows that this theme is very nascent,” said Bradley Amoils, managing director and portfolio manager of US-based Axiom Investors, which manages the Pengana Axiom International Fund for Australian retail investors.
The term AI servers refers to servers specifically built to handle the demands of AI workloads and they play an important role in the AI technology landscape, supporting AI workloads across the entire pipeline – from data prep and training/fine-tuning to deployment and ongoing management, according to Intel.
Amoils noted that even with Nvidia’s projected sale of 450,000 AI-enabled servers in 2024, these would still constitute only a small fraction of the global supply.
“If we look further ahead, there are suggestions AI-enabled servers might account for 5.3 per cent of servers sold in 2026, which would still fall well short of demand, and suggests we’re only at the beginning,” he said.
Despite this, Amoils believes that Nvidia remains well positioned to generate solid revenues from generative AI in the upcoming quarters.
“How high Nvidia can go is tough to say, and we don’t like to forecast these things, but if we look at the forward revenues for the next few quarters, there is no sign of this trend slowing down,” he said.
Axiom’s director also reminded investors that producing a chip takes approximately three to four months, noting that “if a chip goes into the foundry this quarter, it creates revenue the next quarter”.
He believes that this makes stocks connected to Nvidia, such as its foundry partner TSMC and Vertiv, attractive to investors.
“These companies are providing interesting data and show the AI theme extends from mega cap to small cap positions,” he said.
Two AI cycles
Other fund managers share the view that although the current AI trend may be “overhyped”, the long-term transformative potential of AI and investment opportunities in AI are valid.
Mark Casey, Capital Group’s portfolio manager, noted that the markets tend to overestimate mega trends in the short term while underestimating them over the longer term.
“What is remarkable about AI is its broad potential utility. Because it can take on a multitude of human tasks, I consider the AI market to be unknowably massive,” he said.
Another portfolio manager at Capital Group, Cheryl Frank, expects overcapacity and excesses in the near-term, as enterprises are still experimenting with AI and determining how to use it to gain a competitive advantage.
However, Frank warned that the lack of adequate power capacity, basic materials or capital equipment may slow the AI expansion.
“I think there will be two AI cycles,” Frank said, noting that currently, the market is in the middle of an advertising-driven consumer AI cycle, which is likely to experience a pullback.
“But the enterprise AI cycle will be a much longer and slower build,” she said.
Discussing trends and investment opportunities in the AI space, Capital Group highlighted the “four-layer technology stack”, which enables AI workloads and the supply chains required for AI infrastructure, as one of the key opportunities where an AI spending spree could pay off for investors.
According to the firm, some companies will capitalise at each layer of the stack, while others will launch products at two or three layers.
“The question is, which companies will execute best – and which will stumble? That’s what I am focused on,” Casey said.
The AI stack includes semiconductors, cloud infrastructure, large language models such as ChatGPT, and applications for end users.
Meanwhile, chipmakers like Nvidia and ASML operate at one level while tech giants like Alphabet, Microsoft and Amazon seek to dominate multiple layers.
