AI Is Holding Up Global Growth, and the IMF Sees the Risk

Kristalina Georgieva is the Managing Director of the International Monetary Fund. She spoke in Singapore on 7 October 2026. The annual meetings in Bangkok were days away. She started from a trade figure. The text of her speech^[1] says that AI hardware and related products account for more than a tenth of global goods trade.

In the same speech the Fund names a sharp earnings disappointment as the weak point.

My thesis is blunt. Global growth rests on the AI cycle alone. The spending that holds up that growth is the spending the Fund identifies as the breaking point.

The Fund measures AI against railways, the power grid and telecoms

Georgieva chose railways, the power grid and telecommunications as her three precedents. Her speech says that AI investment as a share of GDP will match and perhaps exceed that of all three.

That is a forecast.

I take the comparison literally. Those networks share a common structure. Capital is committed for decades. Revenues are expected in advance. Debt covers the gap between the two. The Fund measures the comparison as a share of GDP. I stick to that yardstick.

One distinction matters here. The network is structural and lasts decades. Its financing is cyclical and lasts as long as creditors' patience.

A tenth of trade and half of US growth

The numbers on this dependence come from Hung Tran's notes for the Atlantic Council^[2]. The notes are dated 12 October 2026 and cite Goldman Sachs.

• Global AI investment: more than $1 trillion in 2026, or 0.9% of world GDP.
• Estimate for the following years: 1.3% in 2027, 1.4% in 2028.
• AI and high-tech spending: between 1.8% and 2.5% of US GDP.
• Contribution: roughly half of US growth in recent quarters.
• Stock market: ten companies are approaching 40% of the S&P 500 index.

The Fund adds its own estimate. Georgieva cited the Fund's research on 7 October. That research says AI could add up to 0.5 percentage points a year to global growth. Over ten years that equals an economy the size of ASEAN.

The Fund also says that AI-linked economies have grown faster than the rest. I treat that as a correlation. The Next Web^[3] also covered the speech.

The next reading is mine. The jump from 0.9% to 1.3% is 0.4 points. The jump from 1.3% to 1.4% is 0.1 points.

Growth feeds on the change in spending more than on its level. If Goldman Sachs's estimates hold to the decimal, the boost in 2028 would be a quarter of the one in 2027.

The slowdown is already written into the baseline scenario.

How a sector shock becomes a financial shock

The mechanism is in the Fund's own words. Concentration in AI can amplify systemic risk.

It points to two channels:

• The debt of the large cloud operators, which carries the disappointment into credit.
• The large global holdings of US shares, which carry it into portfolios on every continent.

My reading parts ways here with those who watch share prices. I look first at who has lent. A share that falls shrinks a fortune. A loan that goes bad halts a building site.

The backdrop narrows the margin. That speech puts ten-year yields in the United States, Germany and Japan at their highest since 2007, 2009 and 1996. Global public debt is heading towards 100% of GDP.

The bill falls on the creditors of the cloud operators, on global holders of US shares and on the economies that grew faster thanks to AI. Governments pay too. They would have to absorb the blow with rates at their highs and debt heading towards 100% of GDP.

The public cushion is thin.

Three moves for those who allocate capital or sign contracts

1. Within 12 months: count how much you really depend on ten companies. Ten companies are approaching 40% of the index. They enter a portfolio through passive funds, credit, suppliers and customers. Add all of it up on a single sheet.
2. Within 18 months: tie multi-year contracts to the right number. Every long-term commitment for cloud capacity, energy or supplies depends on the 1.3% estimated for 2027. Write exit clauses for the case in which the other party cuts spending.
3. Within 36 months: add the scenario missing from risk models. The missing scenario is an earnings disappointment while yields are at their highs. Shares and bonds then fall together. Models built on past price swings treat that as rare.

The last point concerns the story told to investors. Anyone presenting US growth as broad-based should remember the half that comes from a single line of spending.

My forecast for the end of 2027

By 31 December 2027 the weight of the top ten companies in the S&P 500 index will remain below 40%.

I base this on the Goldman Sachs figure. The boost from spending falls between 2027 and 2028. Markets set prices on the year ahead. My forecast is a judgement on the cycle. The building of the networks remains structural.

• Confidence: 55 out of 100.
• Horizon: 31 December 2027, 444 days from today.
• Verification: the index weights at the year-end close.
• What would prove me wrong: the top ten companies at 40% or more of the S&P 500 at the close on 31 December 2027.

The numbers to watch and what would prove me wrong

Four series signal in advance where the cycle is heading:

• The quarterly earnings of the large cloud operators, set against their announced spending.
• New debt issuance by the same operators.
• Ten-year yields in the United States, Germany and Japan relative to their current highs.
• Goldman Sachs's revisions to the 2027 and 2028 shares.

Two facts would change my mind. The first is that the share of US growth due to AI and high tech falls well below half while total growth holds up. The second is that economies far from AI go back to growing as fast as those tied to it.

In that case growth would rest on more than one cycle and my thesis would fall.

Until then the number to keep on the table is 1.3% of world GDP in 2027.