UBS now estimates total AI capital expenditure at $998 billion in 2026 and $1.447 trillion in 2027 — but the headline total conceals a composition that tells a structurally different story.
UBS has lifted its AI capital expenditure estimates, as reported by Investing.com, putting the total at $506 billion in 2025, $998 billion in 2026, and $1.447 trillion in 2027. Every one of those figures is a UBS estimate or forecast — not a reported actual, not an audited number, and not an industry disclosure. The 2025 figure carries that status equally with the others.
Within those totals, UBS estimates memory spending at $71 billion in 2025, $367 billion in 2026, and $923 billion in 2027. UBS’s own figures place memory’s share of total AI capex at 37% in 2026 and 64% in 2027. Those share figures are UBS’s own — not derived here.
UBS states that higher memory costs account for about 60% of the increase in AI capex in 2026, and more than the entire net increase in 2027, as spending on other components falls. That second clause is where the structural read begins.
Memory vs. Non-Memory — UBS Estimates + Derived Series
Memory — UBS est.
2025: $71B
2026: $367B
2027: $923B
Non-Memory — Derived*
2025: $435B
2026: $631B
2027: $524B
*Non-memory figures derived by subtracting UBS’s memory estimates from UBS’s total estimates. This is not a series UBS published. All underlying figures remain UBS estimates, not reported actuals.
The key insight: A component that accounts for more than the entire net increase in a total is only arithmetically possible if the rest of the mix is contracting. UBS’s own figures describe exactly that for 2027 — which means a headline total rising toward one and a half trillion dollars conceals a composition moving in two different directions at once.
A total that rises while one component accounts for more than all of the rise describes a mix in which everything else is contracting.
The Structural Read
This is not primarily a forecast that AI spending grows. It is a forecast that what the spending buys changes — and the 2027 line is the one that warrants the closest reading. Read UBS’s own framing slowly: higher memory costs account for more than the entire net increase in 2027 as spending on other components falls. A headline total can rise and still conceal a composition in which most of the constituent parts are moving the other way. That is what UBS’s own figures describe here. This is not a claim about the world, not an endorsement of the forecast, and not a prediction of any outcome.
The derived non-memory series makes the composition visible, and it needs both halves of its description to be accurate. On these figures — derived by subtracting UBS’s memory estimates from UBS’s totals, a series UBS did not publish — non-memory AI capex peaks in 2026 at $631 billion and is lower in 2027 at $524 billion. That is a real directional observation on these estimates. The other half of the description is equally real: $524 billion in 2027 remains above the $435 billion in 2025. This is a mix shifting, not a market disappearing. It is not a collapse, not a crash, and not a bust. No growth rate or percentage is computed from this series here.
UBS — as reported by Investing.com
“Higher memory costs account for about 60% of the increase in AI capex in 2026 and more than the entire net increase in 2027, as spending on other components falls.”
There is a precise distinction that a compositional forecast requires. A forecast that one input comes to dominate a spending total is a statement about relative cost and about mix. It says nothing about what happens to volumes, to units purchased, or to the physical capacity deployed. Spending and volume are different quantities — a line item can grow while the units behind it shrink, and it can shrink while the units behind it grow. Nothing here infers a shortage, a price movement, or a supply constraint. None of those claims is established in the material this piece relies on, and none is made here.
The total tells you how much. The mix tells you who you are dealing with.
A buyer planning against a rising total prepares to spend more money — that is a budgeting exercise. A buyer planning against a shifting mix has to prepare for something harder: a different negotiation, with a different set of suppliers, on different terms, with a different coststructure sitting underneath the same budget line. The total and the mix are not the same planning problem. That is a general property of the difference between the two — not a claim about any buyer, supplier, or this market’s structure, and not advice of any kind.
Three Implications
IMPLICATION 1 — THE HEADLINE TOTAL IS NOT THE SIGNAL
On UBS’s estimates, total AI capex rises from $506 billion in 2025 to $998 billion in 2026 and $1.447 trillion in 2027. That headline is true and also incomplete. The structural content of the forecast sits in the composition, not the sum. A total that is rising while one constituent is contracting is a different planning environment than a total that is rising because all constituents are rising. The two require different analytical tools, different supplier conversations, and different budget assumptions — even when the top-line number is the same.
IMPLICATION 2 — SPENDING AND VOLUME ARE NOT THE SAME VARIABLE
The derived non-memory line in UBS’s estimates peaks in 2026 and is lower in 2027 — but that is a statement about dollar spend, not about units, capacity, or deployment. Those are distinct quantities, and a compositional forecast establishes nothing about them. Analysts and planners who read a declining spend line as a declining volume line are reading a different forecast than the one UBS published. The distinction matters every time a budget figure is used to draw an inference about physical scale.
IMPLICATION 3 — MIX SHIFTS CHANGE THE NEGOTIATION, NOT JUST THE BUDGET
The operational consequence of a mix shift is not captured by updating a total spend figure. When the composition of a large capital program changes — when one category comes to account for 64% of the total, as UBS’s estimates describe for memory in 2027 — the procurement logic, the supplier concentration, the contract terms, and the coststructure of the entire program change with it. Budgeting for a total is one exercise. Negotiating a changed mix is another. The second is harder, and it is the one that UBS’s figures point toward.
Business Engineer Framework
Map of AI — Where a Forecast Sits in the Stack Changes What It Means
The Map of AI framework maps 200+ companies and technologies across nine layers of the AI stack — from infrastructure and memory to models, interfaces, and applications. Understanding which layer a spending forecast is concentrated in changes the entire strategic read. When composition shifts faster than totals, the layer map is the tool that makes it visible. UBS’s estimates are a prompt to ask not just how much, but where — and what that means for every layer above and below.
UBS’s estimates — all of them forecasts, none of them reported actuals — describe a total that rises to nearly one and a half trillion dollars by 2027 and a composition that tells a structurally different story underneath it. On these figures, UBS puts memory at 37% of the total in 2026 and 64% in 2027, while everything else, taken together, peaks in 2026 and is lower in 2027 than in 2026 — though still above 2025. A mix shifting that fast is not merely a budget update. It is a forecast that the planning problem itself changes: from how much to spend, to who you are negotiating with, on what terms, and with what leverage. Those are not the same question, and the headline total answers only the first one.
Every figure above is a UBS estimate or forecast, including the 2025 figures. None is a reported actual, an audited number or an industry disclosure, and nothing above endorses or refutes the forecast or predicts whether it is met. The non-memory series is derived here by subtracting UBS’s memory estimates from UBS’s totals. It is not a series UBS published. On those estimates it is lower in 2027 than in 2026 and remains above its 2025 level. The memory share figures of 37% and 64% are UBS’s own. No other figure is derived above. Spending and volume are different quantities. Nothing above claims anything about units, capacity or deployment, and nothing above claims or implies a shortage, a supply constraint, a price movement or any cause for the forecast composition. No company is named above, no share price, valuation, rating or analyst view is stated, and no view is expressed on any security or sector. Nothing above says who benefits or is harmed, and no guidance is given to any buyer or supplier. This is businessanalysis of a published forecast. It is not investment advice and no recommendation is made.
Gennaro is the creator of FourWeekMBA, which reached about four million business people, comprising C-level executives, investors, analysts, product managers, and aspiring digital entrepreneurs in 2022 alone | He is also Director of Sales for a high-tech scaleup in the AI Industry | In 2012, Gennaro earned an International MBA with emphasis on Corporate Finance and Business Strategy.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
