JPMorgan, UBS project cloud and AI capex nearing $1 trillion

Two of the world's largest banks issued forecasts in recent days pointing to a rapid acceleration of spending on artificial intelligence and cloud infrastructure, with both projecting annual capital expenditures crossing the $1 trillion threshold within the next year.

JPMorgan Sees $954 Billion in Cloud Capex This Year

In its Fall 2026 U.S. Semiconductor and Semiconductor Equipment report released on September 18, JPMorgan ↗ 0.69% projected cloud capital expenditure reaching $954 billion in 2026, rising to $1.41 trillion in 2027 and $1.54 trillion in 2028. The bank said early AI monetization and return-on-investment signals are becoming more visible, with accelerating cloud and AI revenue growth, expanding backlogs, and improving profitability across cloud businesses.kucoin+1

JPMorgan analyst Harlan Sur wrote that semiconductor industry fundamentals are strengthening and that recent sector pullbacks are "constructive." The bank's models project 118% growth for the semiconductor industry in 2026 and 35% in 2027. On custom chips, JPMorgan estimates the AI ASIC market will reach $60 billion to $70 billion by 2026, with Broadcom ↗ 1.13% and Marvell ↗ 5.85% dominating at estimated market shares of 80–85% and 10–12%, respectively.techflowpost+1

The bank's large-cap top picks include Broadcom, Analog Devices ↗ 1.53%, Marvell, Micron ↗ 2.72%, and KLA ↗ 4.25%, with additional optimism toward NVIDIA ↗ 2.30%, Applied Materials ↗ 4.34%, and Lam Research ↗ 5.15%.kucoin

UBS Highlights Memory as the Driving Force

Separately, UBS ↗ 0.77% projected that global AI capital expenditures will approach $1 trillion in 2026 — estimating $998 billion — and climb to approximately $1.4 trillion by 2027. The bank identified surging memory costs as the primary driver. UBS estimates memory-related spending will rise from $71 billion in 2025 to $367 billion this year and $923 billion in 2027, accounting for roughly 60% of the AI capex increase in 2026. By 2027, as spending on other AI components is expected to decline from $631 billion to $525 billion, memory cost increases alone would represent about 90% of the nearly $1 trillion cumulative rise in AI capex between 2025 and 2027.phemex+2

A Converging View on Semiconductor Strength

The two forecasts, while differing in scope — JPMorgan focused on cloud capex broadly while UBS centered on AI-specific spending — paint a consistent picture of infrastructure investment accelerating well beyond previous expectations. JPMorgan noted that supply discipline and diversified end markets have reduced cyclicality in the semiconductor industry, while wafer fabrication equipment spending is expected to grow 31% in 2026 and 38% in 2027. DRAM and NAND pricing are each forecast to rise approximately 250% in 2026, a trend that aligns with UBS's emphasis on memory as the dominant cost driver ahead.news.futunn+2