Why AI spending pushes interest rates up, and what reaches Malaysia

Artificial intelligence is usually discussed as a technology story. In bond markets it has become an interest-rate story too, because building data centres needs enormous amounts of money and electricity. Malaysia sits on an unusual side of this: it sells chips and hosts data centres while borrowing costs are set mostly abroad.

Three ways AI can push rates up

Visual guide

Three ways AI can push rates up

1. Supply

   Mechanism

   Tech companies issue bonds to fund capex, adding to the supply of long-term debt

   What to look for

   Wider spreads and new-issue premiums, higher long yields around big deals

2. Investment demand

   Mechanism

   Strong demand for capital raises the neutral real interest rate

   What to look for

   Real yields rising more than inflation expectations

3. Costs

   Mechanism

   Memory chips and electricity become scarce and expensive

   What to look for

   Price increases in components, power and electronics

Channel	Mechanism	What to look for
Supply	Tech companies issue bonds to fund capex, adding to the supply of long-term debt	Wider spreads and new-issue premiums, higher long yields around big deals
Investment demand	Strong demand for capital raises the neutral real interest rate	Real yields rising more than inflation expectations
Costs	Memory chips and electricity become scarce and expensive	Price increases in components, power and electronics

How much are Big Tech spending and borrowing?

Item	Figure
2026 capex guidance (July results)	Amazon about US$220bn, Alphabet US$195-205bn, Microsoft about US$175bn, Meta US$130-145bn
Total	about US$720-745bn, 2.2-2.3% of US nominal GDP (Platformonomics)
2027 estimate	US$1.2 trillion for five firms including Oracle (Goldman Sachs); consensus about US$1.1 trillion
AI-related bond issuance, 2026	about US$570bn expected; US$236bn by end-May, four times a year earlier (Morgan Stanley via Reuters)
Notable deals	Alphabet sterling 100-year bond (6.125% coupon); Amazon about US$53.8bn in March, the largest corporate bond sale on record (IFR); Meta US$25bn in April

Is the supply channel visible?

Partly, in credit markets. Spreads on long hyperscaler bonds (over 20 years, median) widened from 108.5 bp to 118 bp, and the median new-issue premium rose from 2.25 bp in 2025 to 12 bp in 2026, according to a Reuters tally. Investors are asking a bit more to absorb the volume.

Government bond yields are another matter. PIMCO (21 September 2026) looked at six large issues and found no significant reaction in the 10-year Treasury yield, the term premium or swap spreads around them. Capital Economics also described the effect as limited.

Real yields did most of the work

Change in the US 10-year real yield against 10-year breakeven inflation, early January to early October 2026

US 10-year (FRED)	Early January	Early October	Change
Nominal (DGS10)	4.19%	5.27% (6 Oct)	+108 bp
Real (DFII10)	1.94%	2.91%	+97 bp (high 2.95% on 5 Oct, highest since November 2008)
Breakeven inflation (T10YIE)	2.25%	2.36% (7 Oct)	+11 bp

When the real yield carries almost all of the rise, it fits the investment-demand channel better than an inflation scare. The New York Fed’s estimate of the neutral real rate rose from 0.89% in 2025 Q4 to 1.01% in 2026 Q2, and the median FOMC longer-run rate projection went from 3.1% in June to 3.2% in September. Fed Governor Cook said on 28 September: “In the short term, AI appears to be adding inflationary pressures to the economy.”

The cost channel

• Standard DRAM contract prices rose 93-98% quarter on quarter in 2026 Q1 (TrendForce).
• US electricity CPI was up 3.8% year on year in August, and the PJM grid’s 2027/28 capacity price hit its cap of US$333.44 per MW-day for the third record in a row.
• In Malaysia, a 32GB DDR5 RAM kit went from about RM700 to RM2,000 in shops by March 2026, but the CPI item for computers, laptops and tablets rose only 1.8% in August (see chipflation).

The opposite argument

PIMCO (1 July 2026) argues that AI could lower neutral rates over the short to medium term: households may save more as a precaution, labour’s share of income may fall, and fiscal expectations may change. On the 43 days when major AI models were released, the 5-year, 5-year-forward real rate fell a combined 100 bp or so, while it rose about 100 bp over the whole period. Both sides can point to data; neither has settled it.

What reaches Malaysia?

Indicator	Value	Period
Electrical and electronics exports	+66.5% year on year, 48.4% of total exports	August 2026
SSD exports	+468.8%	August 2026
Data centre and cloud projects, approved	RM95.8bn, about 44% of RM218.5bn total	First half 2026
10-year MGS yield	3.506% → 3.867%, +36 bp	End-2025 to end-August 2026
MGS 10-year minus US 10-year	−67 bp → −88 bp	End-2025 to end-August 2026
OPR	2.75%, unchanged	Since July 2025

The export figures are from DOSM; the investment figure is from MIDA, as reported by The Edge. Malaysia earns from the AI build-out through chips, storage and data centres. The rate side arrives mainly through US yields: the US 10-year rose faster than the Malaysian 10-year, so the gap widened. For Malaysian borrowers on floating rates, the OPR still matters more than any of this.

Check what a past ringgit amount buys today →