UBS has raised its estimate for global AI capital spending, saying total capex will reach $998 billion in 2026, nearly double the $506 billion projected for 2025. The figure is expected to rise further to $1.447 trillion in 2027.
The bank said the sharp upward revision is mainly the result of rapidly rising memory prices, not a broad-based expansion in infrastructure investment.
Memory spending drives the revision
UBS estimates memory spending will climb from $71 billion in 2025 to $367 billion in 2026, then expand again to $923 billion in 2027. Non-memory AI capital spending, by comparison, is seen at about $631 billion in 2026 before falling to $525 billion in 2027.
Memory is also taking a much larger share of total AI capex. UBS puts that share at about 14% in 2025, rising to 37% in 2026 and then 64% in 2027.
Most of the increase comes from higher memory costs
UBS said about 60% of the year-over-year increase in AI capital spending in 2026 will come from higher memory costs. In 2027, the increase in memory spending is expected to exceed the net increase in total AI capex because spending on other components is projected to decline over the same period.
From 2025 to 2027, nearly $1 trillion in additional AI capital spending is expected, and UBS estimates about 90% of that increase will come from rising memory spending.
Limited effect on real U.S. GDP
UBS said that if the increase in spending is driven mainly by higher prices, the effect on real U.S. GDP would be fairly limited. Instead, the shift would show up more clearly as revenue and profit moving toward Asian memory producers, giving a more direct boost to GDP in those economies.
With major global memory manufacturers concentrated in Asia, UBS said the region’s memory supply chain may benefit more than the market had previously expected as the AI investment cycle deepens.
A different way to track the AI investment cycle
UBS said the change points to a structural shift in the economics of AI investment. Rather than being driven only by larger infrastructure buildouts, the cycle is increasingly being shaped by rising costs for key components needed to support more powerful computing systems. That, in turn, suggests investors may need to adjust how they track AI capital spending.