UBS has sharply revised its artificial intelligence capital expenditure projections, with the rapid escalation of memory costs emerging as the primary catalyst behind this new wave of spending—a shift that is fundamentally reshaping the structure and economic rationale of the AI investment cycle.
According to the latest UBS estimates, global AI capital expenditure will reach $998 billion in 2026, nearly doubling from $506 billion in 2025. By 2027, this figure is projected to climb further to $1.447 trillion. The significant upward revision stems primarily from the rapid ascent of memory prices rather than an overall expansion in infrastructure investment scale.
The explosive growth in memory spending signals a structural transformation in the distribution of AI investment gains. UBS notes that if spending growth is driven mainly by price increases, the ripple effect on US real GDP will be relatively constrained, with the impact manifesting more as a transfer of revenue and profits toward Asian memory manufacturers—thereby contributing positively to the GDP of those relevant economies.
Memory Spending: From Supporting Role to Leading Driver
Memory is rapidly becoming the dominant source of AI capital expenditure growth. UBS estimates that global memory spending will surge from $71 billion in 2025 to $367 billion in 2026, before expanding further to $923 billion by 2027.
Meanwhile, other AI-related expenditures are trending in a markedly different direction. UBS projects that non-memory AI capital expenditure will reach approximately $631 billion in 2026, but is expected to retreat to $525 billion by 2027.
Looking at the structural composition, memory's share of total AI capital expenditure is undergoing a fundamental rebalancing. In 2025, memory spending accounted for roughly 14% of total AI capital expenditure; UBS anticipates this proportion will rise to 37% in 2026 and then surge to 64% by 2027.
Breaking Down the Sources of Incremental Spending
UBS's calculations illuminate the underlying logic driving this round of AI investment expansion. Among the year-over-year increase in AI capital expenditure for 2026, rising memory costs contribute approximately 60%; by 2027, the incremental growth in memory spending alone will exceed the net increase in total AI capital expenditure, as spending on other components declines during the same period.
In aggregate, UBS estimates that of the nearly $1 trillion in additional AI capital expenditure accumulated between 2025 and 2027, approximately 90% will originate from rising memory spending.
This data indicates that the economic engine of the AI investment cycle is undergoing a profound shift—moving from pure infrastructure scale expansion toward the escalating costs of critical components needed to support increasingly powerful computing systems.
Macroeconomic Impact: A Tailwind for Asian Memory Producers
UBS specifically highlights that the nature of memory spending—whether it is price-driven or volume-driven—creates fundamentally different macroeconomic consequences.
If the incremental spending primarily reflects rising memory prices, the contribution to US real GDP will be quite limited. Under this scenario, capital flows will increasingly represent a profit transfer from AI infrastructure investments toward memory manufacturers, and since the world's leading memory producers are concentrated in Asia, the relevant economies will gain more direct positive GDP contributions as a result.
This assessment carries meaningful implications for investor asset allocation: as the AI investment cycle deepens, the beneficiaries within the Asian memory supply chain may outperform earlier market expectations.