AnswersPublished23 sources
AI Infrastructure Spending Could Top $1 Trillion a Year in 2026—Memory Is the Key Variable
JPMorgan and UBS both project annual cloud or AI related infrastructure spending near $1 trillion in 2026 and above $1.4 trillion in 2027. JPMorgan forecasts cloud capex of $954 billion in 2026, $1.41 trillion in 2027, and $1.54 trillion in 2028; UBS’s reported AI capex outlook is $998 billion, $1.447 trillion, and...
Published byEdited with GPT-5.6 TerraImages generated with GPT Image 2
Research answer
Key takeaways
- JPMorgan and UBS both project annual cloud or AI related infrastructure spending near $1 trillion in 2026 and above $1.4 trillion in 2027.
- JPMorgan forecasts cloud capex of $954 billion in 2026, $1.41 trillion in 2027, and $1.54 trillion in 2028; UBS’s reported AI capex outlook is $998 billion, $1.447 trillion, and $1.619 trillion, respectively.
- The common signal is broad semiconductor demand: memory, custom AI chips, and chipmaking equipment all remain central to the buildout.
AI Prompt
openai.comCreate a landscape editorial hero image for this Studio Global article: How do JPMorgan’s and UBS’s latest forecasts indicate that global cloud and AI infrastructure capital expenditures are set to approach or ex. Article summary: JPMorgan and UBS independently point to an AI-infrastructure investment cycle at roughly $1 trillion per year beginning in 2026 and well above it in 2027. Their estimates differ in scope—JPMorgan’s is cloud capex, UBS’s . Topic tags: general, general web, user generated. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, charts with fa
The headline in the latest JPMorgan and UBS projections is not simply that AI spending is rising. It is that cloud and AI infrastructure outlays could reach roughly $1 trillion in 2026 and move well beyond that level in 2027. The important qualification is that the banks measure different scopes, while UBS attributes much of its projected increase to memory-price inflation. 4
18
The forecasts: near $1 trillion, then $1.4 trillion or more
JPMorgan’s reported global cloud-capex forecast is:
- $954 billion in 2026
- $1.41 trillion in 2027
- $1.54 trillion in 2028
18
UBS’s updated AI-related capex figures are reported at:
- $998 billion in 2026, up from $506 billion in 2025
- $1.447 trillion in 2027
- $1.619 trillion in 2028, implying about $4.1 trillion across 2026–28 in a separate reported UBS outlook
4
12
These are not like-for-like estimates: JPMorgan’s figure is framed as cloud capex, while UBS’s is AI-related capex. Still, their direction is consistent. Both point to a buildout that reaches approximately $1 trillion annually in 2026 and rises into a roughly $1.4 trillion–$1.6 trillion range thereafter. 4
18
12
Why UBS says memory drives most of the increase
UBS estimates that about 90% of the increase in AI capex from 2025 to 2027 is attributable to rising memory expenditure. Its reported memory-spending figures rise from $71 billion in 2025 to $367 billion in 2026 and $923 billion in 2027. 3
4
That distinction matters. A larger capex total does not necessarily mean a proportionate increase in the number of AI systems installed. If the memory needed for AI servers becomes much more expensive, the cost of building a given amount of compute capacity rises even when deployment plans are broadly unchanged.
UBS’s forecast therefore combines two effects:
- Continued demand for AI infrastructure.
- A much higher memory bill embedded in each server and cluster.
The bank’s conclusion is not that all incremental spending represents additional computing capacity. It is that memory scarcity and pricing have become a major determinant of what AI infrastructure costs. 3
4
JPMorgan’s case extends beyond memory
JPMorgan’s semiconductor outlook supports the idea of broad infrastructure demand, even though memory is a major part of the forecast. It expects global semiconductor revenue to grow 118% in 2026 and 35% in 2027. Excluding memory, its reported growth assumptions are still 32% and 18%, respectively—evidence that the expected expansion is not solely a memory-price story. 19
Custom AI chips are becoming a larger part of the mix
JPMorgan estimates a $60 billion–$70 billion custom AI ASIC market in 2026, with a projected 40%–50% compound annual growth rate in subsequent years. It expects ASIC/XPU shipments to account for 54% of AI-accelerator unit shipments in 2027, surpassing GPUs. 35
36
The reported high-end ASIC market is highly concentrated: Broadcom is estimated to hold 80%–85%, while Marvell holds 10%–12%. 23
This does not mean GPUs are disappearing. Rather, it suggests large cloud providers may increasingly supplement general-purpose accelerators with custom silicon designed for particular AI workloads.
Chipmaking capacity must expand too
JPMorgan also raised its wafer-fab-equipment outlook to:
- $163 billion in 2026
- $225 billion in 2027
- $263 billion in 2028
24
The forecast implies that the AI cycle reaches beyond chips already available for sale. It also requires investment in the equipment used to manufacture memory and leading-edge logic chips. JPMorgan cited DRAM and TSMC-related demand as key reasons for its upgraded equipment view. 24
The memory assumptions are unusually aggressive
JPMorgan’s reported pricing assumptions illustrate why memory has become so consequential to the capex outlook. It expects blended DRAM pricing to rise about 250% in 2026 and 30% in 2027, while NAND pricing is forecast to rise about 250% and 25%, respectively. 23
45
Its broader research has also estimated that DRAM prices will have increased by more than 400% from the start of 2024 to the end of 2026. 48
The underlying constraint is supply. JPMorgan expects DRAM demand growth to exceed supply growth through 2027, with the imbalance most acute in that year; its model shows DRAM supply growth potentially moving ahead of demand by 2028, while NAND remains slightly short. 42
43
What the two forecasts mean
UBS and JPMorgan converge on a central point: the AI buildout is becoming a full-stack infrastructure cycle, spanning cloud data centers, accelerators, custom chips, memory, and semiconductor-production equipment.
But investors and operators should separate spending growth from physical capacity growth. UBS’s analysis indicates that surging memory prices account for a large share of the higher dollar totals. JPMorgan’s outlook adds a broader demand signal—semiconductor revenue excluding memory is still expected to grow, custom silicon is scaling, and fab-equipment spending is rising. 3
19
24
Together, the forecasts support a sustained infrastructure cycle through 2028. They do not guarantee that every additional dollar of capex translates into an equivalent increase in deployed AI compute.
Studio Global AI
Continue your research
This page includes a source-backed answer you can continue inside Studio Global.
People also ask
What is the short answer to "AI Infrastructure Spending Could Top $1 Trillion a Year in 2026—Memory Is the Key Variable"?
JPMorgan and UBS both project annual cloud or AI related infrastructure spending near $1 trillion in 2026 and above $1.4 trillion in 2027.
What are the key points to validate first?
JPMorgan and UBS both project annual cloud or AI related infrastructure spending near $1 trillion in 2026 and above $1.4 trillion in 2027. JPMorgan forecasts cloud capex of $954 billion in 2026, $1.41 trillion in 2027, and $1.54 trillion in 2028; UBS’s reported AI capex outlook is $998 billion, $1.447 trillion, and $1.619 trillion, respectively.
What should I do next in practice?
The common signal is broad semiconductor demand: memory, custom AI chips, and chipmaking equipment all remain central to the buildout.
Sources
news.futunn.comUBS Group: Global capital spending in the AI sector will rise to $1.4 trillion next year, with 90% of the increase driven by memory price hikes.