By the end of H1 2025, very little around AI investment had been seen in public bond markets, with the sector’s “hyperscalers” said to be generating a considerable level of cash flow and leveraging private equity, private credit and other sources to “fund their capital expenditure (capex).” But with capex set to ramp up over the next 24 months and free cash flow being “constrained when shareholder returns through share buybacks are taken into account, the hyperscalers are leaning on public bond markets.”
Aberdeen Investments has been leaning into this theme through its Short Dated Enhanced Income Fund, taking a “tactical approach by adding select positions at key points in the market.”
Recent moves include a three-year bond issued “by Alphabet and Oracle two-year bonds at the end of October.”
These additions are not a structural overweight, but rather “opportunistic plays in response to market repricing, which can be volatile.”
The Fund also holds bonds issued by SK Hynix, “a South Korean business specialising in memory semiconductors, which is seeing significant growth on rising demand from AI.”
Mark Munro, Investment Director – Fixed Income at Aberdeen Investments and manager of the Short Dated Enhanced Income Fund, says that bond markets care about AI ultimately “because of its potential impact on growth, productivity, employment, inflation and therefore how central banks respond through interest rate policy.”
Munro added that in their view, AI-related corporate bond supply will likely continue to “grow both outright and as a share of the wider bond market.”
They further noted that considering the U.S. investment grade market is close to “$1.7tn in gross supply this year – near to a record year – AI issuance is unlikely to overwhelm the bond market.”
Munro also mentioned that it is likely there will be further “periods of significant issuance from AI-related companies as they continue to invest in their capabilities, causing some indigestion in public credit markets.”
The last three months has seen a glut of debt issuance “in bond markets, including $30bn of bonds issued from Meta, $25bn from Alphabet (owner of Google), $20bn from Amazon and $18bn from Oracle.”
Such issuance has brought firmly into focus how the current AI boom will continue to be financed, with “some of the numbers around the current AI investment cycle far higher than other investment cycles.”
The International Energy Agency estimates that “total data centre electricity consumption will double by 2030.”
Data centres are essential for AI due to the massive computational power and high-speed networking “required to train and run complex AI models, with electricity consumption by AI-optimised servers are expected to increase fivefold by 2030.”
BloombergNEF estimates that by 2035 global data centre power needs will hit 1.6TWh, which will take data centre share of global “power demand from the current 1.3% to closer to 4.4%.”
By 2030 it is estimated that if data centres were “a country, they would be the fourth largest consumer of energy after China, the U.S. and India.”
The power demand required by AI-optimised data centres “underscores the scale of investment needed for businesses to keep up in the AI race.”
The typical cost of an AI specific data centre can be as high “as $50bn depending on the type of chips involved, up to three times the cost of a conventional non-AI data centre.”
Morgan Stanley estimates the “cost of data centre funding at $3tn by 2028, while JP Morgan and McKinsey in the $5-7tn range by 2030.”
Given the investment required, capex guidance and forecasts “are rising significantly.” The capex of the five “hyperscalers” (Amazon, Google, Meta, Microsoft and Oracle) is expected to grow “40% in 2026 to $500bn, and then by a further 17% to £600bn by 2027.”
Oxford Economics suggests that the current run-rate of investment in AI since 2023 is in line with “the digital boom in the 1990s, which saw the rapid growth of the internet.”
Considering the level of required investment, bond markets are beginning to take notice, with credit spreads “12% wider in the U.S. investment grade bond market since the end of September, one driver of which is the recent large amounts of issuance, with markets questioning how this boom in investment will be funded.”
According to Morgan Stanley, just under half of the $3tn required investment to 2028 could be funded from “cash generation, with a quarter from private credit, 10% from other sources such as private equity and sovereign wealth.”
This leaves the remaining “15%, around $450bn, to come from bond markets, with around $200bn to $250bn potentially from investment grade credit markets, with JP Morgan estimating that 14% of the US investment grade debt market is already tied to AI.”
Anthony Merola, U.S. Senior Investment Manager at Aberdeen Investments, adds:
“Within our funds, we view the technology sector as one to play tactically. For funds that focus on shorter dated bonds, big levels of supply provide opportunities to lock into some high-quality names at cheaper levels as markets reprice. For all-maturity or longer dated funds, a much more nimble approach will be required, ensuring control over exposure to the wider AI sector, while creating room in portfolios to add when the next set of major bond issuances come along, of which we expect more in the not too distant future.”
