Goldman Sachs: Global AI Actual Investment to Exceed $1 Trillion in 2026
N.R. Finch
Goldman Sachs estimates 2026 global AI investment will reach $1.019 trillion, roughly $200 billion above the widely cited $800 billion hyperscaler figure — a gap that signals the AI capex cycle is systematically larger and more durable than the market assumed.
Why is the $800 billion number wrong?
The most-cited ~$794 billion hyperscaler capex figure has four structural flaws.
It misses U.S. private companies and other listed firms — Goldman's data shows hyperscalers account for only 40% of 2026 AI-related supply directly.
It omits non-U.S. investment entirely, especially China and the rest of Asia; it ignores that hyperscalers spent over $150 billion before the AI boom, so part of current capex is non-AI; and U.S. hyperscalers operate globally, meaning a significant share of their capex lands outside the U.S.
This means → the headline number understates the global total by ~$200 billion while overstating U.S. domestic investment by ~$200 billion — both direction and magnitude are off.
Three methods, one answer — is the trillion-dollar figure a coincidence?
Goldman used three independent approaches, all converging: the primary estimate yields $1.019 trillion, a profit-revision cross-check ~$1.06 trillion, and a national-accounts-plus-trade method ~$1.002 trillion.
In plain terms = no matter which angle you use, 2026 global AI investment lands around one trillion dollars — this is not an artifact of one model.
Averaged across the three, cumulative global AI investment from 2022 through end-2026 reaches $1.8 trillion.
Where does the money go?
Goldman estimates ~70% of U.S. hyperscaler capex flows to domestic projects, 15% to Asia, and 9% to Europe.
After adjustments, 2026 U.S. domestic AI investment is about $581 billion; the global total is $1.019 trillion.
This reflects a capex landscape still centered on the U.S. but increasingly global — Asia and Europe together already account for nearly a quarter.
How big is this relative to GDP — and is it historically extreme?
Goldman projects U.S. AI capex as a share of GDP rising from 1.8% in 2026 to 2.5% in 2027 and 2.8% in 2028; the global equivalents are 0.9%, 1.3%, and 1.4%.
This means → those levels are consistent with the 2%–5% peak GDP investment shocks seen in past general-purpose technology build-outs — AI intensity has not yet peaked.
In plain terms = railroads, electricity, and the internet all hit higher peaks — AI is not there yet, and there may still be room to climb.
Spending is rising — so why might the real economic pull shrink?
Goldman flags that 8% of the increase in nominal U.S. AI hardware spending so far in 2026 is attributable to cost inflation, not real investment expansion.
If that trend holds through the second half, the boost from AI spending to real investment in 2026 will be smaller than in 2025 — more money spent, but not proportionally more capacity bought.
U.S. national accounts also exclude semiconductor purchases from investment goods, and AI hardware's high import content is netted out of GDP. This means → even if capex keeps growing fast, its direct contribution to headline GDP faces structural limits.
When will the growth rate slow?
Goldman recommends a "dashboard" approach, tracking leading indicators: semiconductor equipment imports from Taiwan and South Korea, related PMI sub-indices, import prices, memory procurement costs, and GPU rental prices.
All indicators currently sit in the highest range since 2022, but trade data from Taiwan and South Korea suggest a modest deceleration in June and July.
This signals that the timing of an AI capex slowdown remains one of the macro market's core uncertainties — Goldman offers monitoring tools but no call on the inflection point.
Content is for reference only, not financial advice.