Apollo: AI Accounts for 65% of U.S. GDP Growth; Goldman Warns Growth Rate to Peak Next Year

nashnova research
今天发布阅读约 9 分钟

Apollo estimates AI investment drove roughly 65% of US GDP growth over the past three quarters, but Goldman Sachs — using a different methodology — puts the figure at just 8%. Both readings point to the same question: can this engine keep running into 2027?

01

Where does the 65% figure come from?

Apollo chief economist Torsten Slok calculates that AI investment added about 1.3 percentage points to real US GDP growth over the past three quarters, against a cumulative pace of roughly 2% — about 65% of the total.
The breakdown: software led at 0.45 pp, computers and peripherals contributed 0.37 pp, communications and electrical equipment 0.21 pp, R&D 0.16 pp, and data-centre construction (concrete and steel) just 0.07 pp.
This means → the GDP lift is overwhelmingly "soft" — software and chip procurement, not pouring concrete. Construction accounts for less than 6%.
02

Why does Goldman get only 8%?

Goldman's independent report puts AI's direct contribution to real GDP growth at just 0.2 percentage points — or roughly 0.8 pp after adding back chip and chip-design exports.
The gap comes down to methodology: Slok counts gross spending; Goldman strips out imported chips and inflation-inflated nominal hardware costs, counting only value that stays onshore.
In plain terms = US companies are spending heavily on AI, but a large share flows to offshore chip suppliers. In GDP accounting, spending ≠ domestic output.
03

When does the growth rate peak?

Goldman projects AI-related spending will surpass $900 billion annualised by end-2026 (up ~$550 bn from 2025), then rise to roughly $1.4 trillion by end-2027 (up another ~$500 bn).
The absolute level keeps climbing, but the annual increment shrinks from $550 bn to $500 bn — the second derivative has turned negative.
This means → AI investment is still growing, but the pace is slowing. Its marginal pull on GDP starts to fade in 2027.
04

AI is supposed to be deflationary — so why is it pushing inflation up?

Goldman warns that the energy demand and input-price pressure from AI buildout could keep US core PCE inflation at 3.2% through 2026.
The bank expects the Fed to hike 25 basis points in both December and March, lifting the fed-funds rate ceiling to 4.25%–4.5%, with a possible early move in January.
In plain terms = AI carries a "cost-cutting, efficiency-boosting" label, but data centres devour electricity and raw materials — the "deflationary technology" is itself generating inflation.
05

What is the key checkpoint for 2027?

Slok states directly: "Any outlook for 2027 equities and credit is, at its core, an implicit assumption about the pace and scale of AI data-centre buildout."
Apollo's and Goldman's conclusions converge on a single question: whether AI capex can hold its pace in 2027 — or begins to contract.
This reflects a deeper vulnerability: US economic growth and corporate earnings are heavily dependent on a single driver. If the buildout rhythm slows, there is no visible successor engine.

市场有风险,内容仅供研究参考,不构成投资建议。