Are we at peak GPUs?

UBS analysts anticipate a "relative near-term peak" in GPU and AI compute spending as a percentage of overall AI capital expenditures in 2025, projecting a decline from 59% to 50-55% by 2026 as investments broaden into areas like memory and networking. While AI compute revenue experienced a 120% CAGR from 2023 to 2025, UBS forecasts a more moderate 20% CAGR through 2030, advising investors to diversify across AI-related segments and exercise caution after the Nasdaq's recent rebound, as current valuations may already reflect positive spending commitments.

Analysis

UBS analysts project that while absolute spending on AI compute infrastructure is not nearing its peak, its proportion of total AI capital expenditure is expected to reach a "relative near-term peak" in 2025. AI compute's share of AI capex is forecast to increase from 53% in 2024 to approximately 59% in 2025, before declining to a 50-55% range from 2026 onwards. This anticipated shift is attributed to a broadening of AI investments into other critical areas such as memory, networking, and industrial AI. While AI compute revenue has demonstrated a robust 120% compound annual growth rate (CAGR) from 2023 to 2025, UBS foresees a transition to "more normalized spending growth," projecting a 20% CAGR through 2030. This moderation reflects maturing demand for GPUs and custom AI chips, alongside continuous performance enhancements and deeper integration with segments like memory, networking, and advanced packaging techniques. Despite recent positive spending commitments from major cloud providers ('Big 4'), sovereign AI initiatives, Chinese AI entities, and neocloud customers, UBS notes these developments are likely "somewhat priced in" given the Nasdaq Composite's significant rebound of over 25% since its April lows, underscoring a cautious outlook despite positive underlying trends.

Key Decisions for Investors

• Investors should consider adopting a more balanced and diversified approach to AI-related investments, looking beyond pure-play GPU and AI compute to include opportunities in memory, networking, and industrial AI applications.
• Given that recent positive spending announcements may already be reflected in current market valuations, particularly after the strong Nasdaq rally, caution is advised against over-exuberance in high-beta technology names, which may face increased volatility.
• Portfolio allocations should reflect the anticipated shift from hyper-growth in AI compute to more normalized, albeit still strong, growth rates, while acknowledging the broadening nature of AI infrastructure spending into adjacent sectors.