UBS now expects AI capex to reach nearly $1tn this year and around $1.4tn by 2027

UBS predicts that artificial intelligence capital expenditure will reach nearly $1 trillion in 2026, followed by a significant rise to around $1.4 trillion in 2027. The primary driver behind this surge is the sharp increase in memory costs. The bank's updated estimates indicate that total AI capital expenditure for the current year amounts to $998 billion, more than double the $506 billion spent in 2025. Spending is expected to continue its upward trajectory to $1.447 trillion by next year.

Memory expenses are anticipated to rise from $71 billion in 2025 to $367 billion in 2026 and $923 billion in 2027. Other AI-related costs are projected to total $631 billion in 2026, before declining to $525 billion the following year. Consequently, higher memory costs will account for approximately 60% of the increase in AI capital expenditure this year, and more than the entire net increase in 2027, as spending on other components is expected to fall.

According to UBS, around 90% of the nearly $1 trillion increase in AI capital expenditure from 2025 to 2027 will stem from higher memory spending. This shift will dramatically alter the composition of AI investment, with memory's share rising from 14% of total AI capital expenditure in 2025 to 37% this year and 64% in 2027. The economic implications of this transition will depend on whether the additional spending results from higher memory prices or increased volumes.

UBS suggests that price-driven increases would add little to real U.S. gross domestic product, instead representing a transfer of income and profits to memory producers in Asia. This would have a positive GDP impact in those economies. The estimates underscore how the economics of AI investment are shifting from infrastructure spending alone to the rising cost of components required to power increasingly powerful computing systems.