Neocloud sell-off is funding risk, not AI demand weakness: UBS; backs hyperscalers, chip stocks

In its view, UBS draws a line between two asset classes that had moved in tandem last week: neoclouds that carry high leverage and depend on cheap borrowing, and well-capitalised hyperscalers and chip companies whose profits remain on the rise. Should that distinction become the prevailing view, a rotation inside the technology sector could unfold rather than an outright sell-off, with chip companies and major cloud platforms outperforming specialist GPU renters. Interest rates are the variable that matters most. With government bonds across developed economies at elevated levels, an additional rise in borrowing costs would hit the neocloud space hardest. In the coming weeks, hyperscaler earnings will test the thesis, as UBS believes cloud revenue gains will pick up.

---

The UBS research note had come out before Monday's recovery, bringing credit to their analysis.

---

AI cloud specialists are under pressure from rising interest rates, yet UBS contends the issue is a funding challenge limited to one segment of the market rather than a fissure in AI demand.

Summary:

• Stocks of neoclouds – firms that lease out AI compute capacity – faced severe selling last week as investors worried about their business models in a prolonged high-rate environment.
• That sell-off pushed the Nasdaq down roughly 4.5% for the week, its steepest decline since June, with pressure also coming from Chinese competition and revenue concerns at AI research laboratories.
• UBS states the sell-off does not undercut the wider AI capex cycle and projects a 33% increase in AI capital expenditure to $1.2 trillion by 2027, with neoclouds accounting for less than 10% of that amount.
• GPU rental rates keep climbing: charges for four-year-old hardware have risen nearly 30% so far this year, while two-year-old chips about 35%.
• UBS notes that robust earnings from TSMC and ASML back the view of sustained AI demand.
• The bank is cautious on neoclouds, prefers well-capitalised cloud companies, high-quality semiconductor and hardware stocks, and defensive technology sectors.

---

UBS attributes last week's steep decline in neocloud stocks to funding and execution hazards in a specific part of the AI market, not to a demand drop-off for artificial intelligence, and the bank maintains its bullish view on the industry's structural expansion.

Neoclouds are firms that specialize in leasing GPU computing capacity for AI tasks. Their shares experienced heavy pressure last week as investors scrutinized their models with interest rates expected to stay elevated. Chinese competition and worries about revenue at top AI firms contributed to the selling, which drove the Nasdaq roughly 4.5% lower for the week, its worst weekly fall since June.

UBS believes the key difference lies between neoclouds and hyperscale providers. Neoclouds generally have a single product and need to make large upfront investments before revenue arrives, making them vulnerable to higher financing costs. Hyperscalers are also increasing borrowing for AI, but their diverse cloud platforms bring in multiple revenue streams. UBS puts their debt cost at a manageable 6.5% to 7% versus returns of roughly 25% or more from new data centre investments.

UBS foresees limited impact on overall AI investment as computing capacity stays tight. The bank projects a 33% rise in AI capital spending to $1.2 trillion by 2027, with neoclouds making up under one-tenth. Still, a big additional jump in financing costs could affect the neocloud part, UBS noted, but strong credit of customers and big pre-payments could offer some shield.

Rental rates give one of the strongest clues. UBS reported that charges for four-year-old GPUs increased 7% in the past month, bringing the year-to-date gain to nearly 30%, and two-year-old GPU rates have risen 35% in 2024. Higher costs for older equipment suggest demand remains above supply.

Q3 earnings back that conclusion. UBS pointed out TSMC stating demand is strong and anticipating its capex from 2026 to 2028 to be significantly above the prior three years. ASML reported solid orders for 2028, with demand extending to 2030. UBS also looks for hyperscalers to show faster cloud revenue increases.

UBS recognized that financing and delivery risks can amplify each other, as stricter credit decelerates build-out and postpones the cash required to meet debt payments. The bank advocates a targeted strategy: reducing exposure to neoclouds, favouring well-financed hyperscalers and top-tier semiconductor and hardware stocks, and exploring defensive tech sectors like payment systems, certain consumer electronics and specific software.