UBS raises its AI capital expenditure forecast to nearly $1 trillion by 2026, with 90% growth driven by memory costs.

Huoxing Finance reports that, according to UBS’s latest estimate on September 19, global AI capital expenditures are projected to reach $998 billion in 2026, nearly doubling from $506 billion in 2025, and further rise to $1.447 trillion in 2027. This significant upward revision is primarily driven by rapid increases in memory prices, rather than broad expansion in infrastructure investment. Memory spending is expected to surge from $71 billion in 2025 to $367 billion in 2026, and then to $923 billion in 2027; during the same period, non-memory AI capital expenditures are estimated at approximately $631 billion in 2026 but are projected to decline to $525 billion in 2027. The share of memory in total AI capital expenditures will rise from around 14% in 2025 to 37% in 2026, and further jump to 64% in 2027. UBS estimates that about 60% of the year-over-year increase in AI capital spending in 2026 stems from rising memory costs, while in 2027, the increase in memory spending alone will exceed the overall net increase in AI capital expenditures, as spending on other components declines; approximately 90% of the nearly $1 trillion increase in total AI capital spending between 2025 and 2027 is attributable to higher memory expenditures. UBS specifically notes that if the additional spending is primarily driven by price increases rather than volume growth, its impact on U.S. real GDP will be limited, with the primary effect being a transfer of income and profits to Asian memory manufacturers, thereby delivering more direct positive contributions to the GDPs of related economies. Since global memory producers are concentrated in Asia, the benefits to Asia’s memory supply chain as the AI investment cycle deepens may surpass prior market expectations. This structural shift indicates that the economic driver of the AI investment cycle is transitioning from pure infrastructure scale expansion to rising costs of critical components required for more powerful computing systems, necessitating corresponding adjustments to investors’ frameworks for tracking AI capital expenditures.