UBS: Global AI Capex Approaching $1 Trillion in 2026, with 90% of Incremental Spending Driven by Memory Price Increases
nashnova research
UBS forecasts global AI capex will hit $998 billion in 2026, nearly doubling from 2025, but roughly 90% of the increase stems from rising memory prices — shifting profits from U.S. tech giants to Asian memory producers.
Where is all this AI money actually going?
UBS's latest forecast: global AI capex rises from $506 billion in 2025 to $998 billion in 2026, then to $1.447 trillion in 2027.
The critical detail: of the roughly $1 trillion in incremental spending, about 90% comes from higher memory costs — not new data centers or more GPUs.
This means → the economics of AI investment are shifting structurally — from "build more compute" to "pay more for the components compute requires."
How did memory go from a side cost to the biggest line item?
Global memory spending jumps from $71 billion in 2025 to $367 billion in 2026 and $923 billion in 2027 — a 13× increase in three years.
The share shift is even more striking: memory's slice of AI capex goes from 14% in 2025 to 37% in 2026, then 64% in 2027.
In plain terms = by 2027, out of every $100 spent on AI, $64 goes to memory; two years earlier that figure was just $14.
Non-memory AI capex actually falls from $631 billion in 2026 to $525 billion in 2027 — spending on other components is shrinking.
How much of the growth is just price inflation?
In 2026, memory price increases account for roughly 60% of the year-on-year rise in AI capex.
By 2027, the memory spending increase exceeds the total net increase in AI capex — because other component spending declines at the same time.
This means → the headline number keeps rising, but strip out memory inflation and actual new investment is contracting.
Who pockets the money? Where are the profits flowing?
UBS flags a crucial distinction: if the spending increase is driven by price hikes rather than volume growth, its contribution to U.S. real GDP is very limited.
This reflects a deeper signal: the money is not turning into new American factories or jobs — it flows as profit to memory producers.
The world's major memory makers are concentrated in Asia — Samsung and SK Hynix in South Korea, and even U.S.-based Micron runs significant capacity in Asia — so Asian economies capture the more direct GDP benefit.
Put simply = U.S. companies write the checks, but Asian chipmakers cash them — the real economic boost of AI investment to the U.S. may be overstated.
市场有风险，内容仅供研究参考，不构成投资建议。