Does AI spending push up interest rates? What reaches Hong Kong

Artificial intelligence is usually discussed as a story about productivity. In 2026 it has also become a story about borrowing, and therefore about interest rates. Hong Kong does not build many of the data centres, but under the peg it borrows at US-linked rates.

Through which channels could AI raise rates?

Visual guide

Through which channels could AI raise rates?

1. Supply of bonds

   How it works

   Tech companies issue large amounts of debt to fund data centres, competing with governments for investors

   What to watch

   Corporate bond issuance, spreads, term premium

2. Investment demand

   How it works

   A capital spending boom raises demand for savings and can push up the neutral real rate

   What to watch

   Real yields, neutral rate estimates

3. Costs

   How it works

   Memory chips and electricity get scarcer and dearer

   What to watch

   DRAM prices, electricity prices, inflation

Channel	How it works	What to watch
Supply of bonds	Tech companies issue large amounts of debt to fund data centres, competing with governments for investors	Corporate bond issuance, spreads, term premium
Investment demand	A capital spending boom raises demand for savings and can push up the neutral real rate	Real yields, neutral rate estimates
Costs	Memory chips and electricity get scarcer and dearer	DRAM prices, electricity prices, inflation

The first two work through real interest rates. The third works through inflation.

How much are tech companies spending and borrowing?

Guidance from second-quarter results in July 2026 (Platformonomics tally):

Company	2026 capital spending guidance
Amazon	About US$220 billion
Alphabet	US$195-205 billion
Microsoft	About US$175 billion (calendar year)
Meta	US$130-145 billion
Total	About US$720-745 billion, 2.2-2.3% of US GDP

Goldman Sachs expects about US$1.2 trillion for five companies including Oracle in 2027. Much of this is being borrowed. Morgan Stanley expects about US$570 billion of AI-related bond issuance in 2026, with about US$236 billion done by the end of May, four times the same period a year earlier. Alphabet sold a 100-year sterling bond in February, Amazon raised about US$53.8 billion in March in the largest corporate bond sale on record, and Meta raised US$25 billion in April. Spreads on long-dated hyperscaler bonds widened from 108.5 to 118 basis points, according to Reuters.

Is there evidence for the supply channel?

Some, but it is disputed. The rise in spreads and new-issue premiums suggests investors want to be paid more to absorb this supply. But PIMCO looked at six large AI bond deals and found no significant reaction in the 10-year Treasury yield, the term premium or swap spreads around them. Capital Economics judged the effect on rates to be limited.

Real rates or inflation?

The breakdown of the US 10-year yield since the start of 2026 points to real rates.

US 10-year measure	2 January 2026	Early October 2026	Change
Nominal yield (DGS10)	4.19%	5.27% (6 October)	+108 bp
Real yield, inflation-protected (DFII10)	1.94%	2.91% (6 October)	+97 bp
10-year inflation expectations (T10YIE)	2.25%	2.36% (7 October)	+11 bp

Change in the US real 10-year yield and in 10-year inflation expectations, start of 2026 to early October

The real 10-year peaked at 2.95% on 5 October, the highest since November 2008. The New York Fed’s estimate of the neutral rate rose from 0.89% in the fourth quarter of 2025 to 1.01% in the second quarter of 2026, and the FOMC’s median long-run rate projection moved from 3.1% in June to 3.2% in September. These are consistent with the investment-demand channel, though they do not prove it.

How about costs?

Conventional DRAM contract prices rose 93-98% quarter on quarter in the first quarter of 2026 (TrendForce). US electricity prices in the CPI were up 3.8% in August, and the PJM grid’s capacity price for 2027/28 cleared at the cap of US$333.44 per megawatt-day, the third record in a row (Utility Dive). Federal Reserve Governor Lisa Cook said on 28 September: “In the short term, AI appears to be adding inflationary pressures to the economy.” In February Governor Michael Barr said “the AI boom is unlikely to be a reason for lowering policy rates.”

What is the case against?

PIMCO argued on 1 July 2026 that AI could lower the neutral rate in the short to medium term, through precautionary saving, a smaller labour share of income and fiscal expectations. Across 43 days of major AI model releases, it found the 5-year, 5-year forward real rate fell by about 100 basis points in total, while it rose about 100 basis points over the full period. On this view, the 2026 rise in real yields owes more to oil, deficits and the term premium than to AI.

What reaches Hong Kong?

Interest rates, through the peg. Hong Kong’s 10-year yield averaged 2.981% in December 2025 and 3.71% in September 2026, a rise of 73 basis points; the US 10-year rose 85 over the same months. Whatever part of the US rise is due to AI, Hong Kong borrowers and bondholders share it.

Prices, only a little. Hong Kong has no memory chip industry, and it is a re-export hub for electronics.

Hong Kong indicator	Value	Period
Durable goods prices, year on year	−2.8% to +0.8%	January to August 2026
Composite CPI, year on year	+1.7%	August 2026
Total exports, year on year	+50.7%	July 2026
Office machines and data processing equipment exports	+106.5%	July 2026
Export unit value index	+13.7%	July 2026
Import unit value index	+13.6%	July 2026

The August CPI showed durable goods turning positive, but the government named fuel, not electronics, as the main pressure. AI demand has lifted trade volumes (Xinhua), while export and import unit values rose almost equally, so traders buy dearer and sell dearer. The product price side is covered in more detail in our note on chipflation.

For a Hong Kong household, the AI boom is therefore more likely to show up in HIBOR-linked borrowing costs and bond prices than in the CPI. To compare a past amount under consumer prices and money supply, open the past money calculator.

Compare the same amount on several bases →