AI fundamentals remain robust despite market volatility

Podcast: Jump Start | Earnings season in focus (5 mins)
Podcast: Across the Pond: Europe’s bond stress test, on Apple , Spotify (22 mins)
Video: Investors Club: Managing inflation, UST yields, and tech pressure (6 mins)
Video: The AI Show |The AI trade's two-sided test (4 mins)
Video: Data with Donovan | The Fed's rate path ahead (5 mins)

Stocks of major neoclouds, specialized AI cloud computing infrastructure providers, came under intense pressure last week as investors questioned their growth models amid the prospect of higher-for-longer interest rates. These concerns helped drive a 4.5% decline in the Nasdaq, its biggest weekly loss since June. Competitive risks from China and worries over revenue growth at leading AI labs also contributed to the weakness.

But the pressure on neoclouds does not undermine the broader AI investment cycle. Financing and execution risks are rising, but recent semiconductor results and GPU rental prices suggest that demand for AI compute remains robust. In our view, this supports the broader AI growth story.

Neoclouds and hyperscalers have very different growth models. Neoclouds are single-product, AI-first infrastructure providers focused on renting GPU compute capacity. They therefore generally require substantial upfront investment before contracted revenue translates into cash flow. While hyperscalers are increasingly tapping debt markets to fund their AI buildout, their general-purpose cloud platforms offer a broad range of services and more diversified revenue streams. We believe hyperscalers’ cost of debt also remains manageable at 6.5-7%, particularly given that the return on invested capital from their new data center investments is around 25% or higher.

The impact on overall AI-related expenditures should be limited, as compute supply remains constrained. We estimate AI capex will grow 33% to USD 1.2tr in 2027, with neoclouds accounting for less than 10% of this investment. While a further significant increase in funding costs could pose a risk to this segment of the AI buildout, the credit quality of many neocloud customers and the volume of advance orders should offer some relief. More importantly, GPU rental prices have continued to rise, pointing to robust demand for AI compute capacity amid supply constraints. The latest data show that rental rates for four-year-old GPUs have risen 7% over the past month, bringing the year-to-date increase to nearly 30%. Rates for two-year-old GPUs, meanwhile, have risen 35% so far this year.

Resilient 3Q earnings should reinforce confidence in AI fundamentals. Both ASML and TSMC posted positive quarterly results last week, with an encouraging outlook for the industry. TSMC said demand remains robust and expects its aggregate capital expenditure in 2026-28 to be meaningfully higher than in 2024-26, while ASML highlighted strong order intake for 2028, with interest extending into 2030. Without taking a view on individual companies, we think these results provide further evidence that AI demand remains strong and continues to grow. We also expect incoming results from hyperscalers to show accelerating cloud revenue growth.

So, we maintain our positive outlook on the structural growth of AI, while acknowledging that funding and execution risks could reinforce each other across the AI infrastructure ecosystem. Tighter credit could constrain deployment, while delivery delays could postpone the cash generation needed to service debt.

For investors, this backdrop calls for a selective approach. We continue to limit exposure to neoclouds amid funding and construction pressures, instead favoring well-funded hyperscalers for their better risk-reward. We also like high-quality semiconductor and hardware beneficiaries of AI spending, and think investors should consider exposure to defensive areas of tech such as payment networks, select consumer electronics names, and select software companies.