Passive bond funds look simple and low-cost, but the way bond indices are built risks leaving investors overexposed to the biggest borrowers, forced to trade at the wrong moments and lagging the indices they track.
Article last updated 15 October 2026.
The biggest story in bond markets is undoubtedly the huge and unprecedented borrowing spree from the US tech giants to pay for the buildout of AI. The largest AI hyperscalers (which include Amazon, Alphabet and Meta) were once able to finance almost all their weighty capital expenditure from the enormous amounts of cash their businesses generated. But training increasingly sophisticated AI models requires massive computing power. And that, in turn, demands more data centres, more semiconductors, more networking equipment and a lot more electricity.
Even America’s tech giants can't pay for all this through their operating cashflows alone. As a result, they’ve been issuing investment grade bonds at pace. The numbers are phenomenal. AI-related borrowing has rocketed from almost nothing in 2024 to an estimated $500 billion in the first nine months of 2026 alone. Some estimate that AI now accounts for about a quarter of all new corporate bond issuance. With the AI giants constantly upping their spending plans, and buildout costs rising, no one really knows how much they'll end up borrowing. JPMorgan Chase estimates that more than $4 trillion of AI-related debt could be issued by 2030.
A lesson from history?
Recent history provides a cautionary tale. In the late 1990s and early 2000s, many telecom companies borrowed heavily to finance fibre networks, wireless infrastructure and spectrum purchases. Their debt issuance surged as they raced to build capacity, and the sector became a big part of corporate bond markets. Excess capacity then put pressure on prices and profitability, leaving many firms struggling to generate the cashflows investors had expected. Credit spreads across the sector widened sharply as investors questioned whether all the borrowing would ever be repaid. Some high-profile firms, including WorldCom and Global Crossing, eventually went bust, leaving bondholders nursing heavy losses.
The big hyperscalers are, of course, a very different kettle of fish to the dot.com-era telcos. They have much stronger balance sheets, vastly superior profitability and much greater financial flexibility. But no one can say with confidence how all this will play out. The AI buildout could generate eye-watering profits. But it could also lead to overinvestment and disappointing payoffs. And even very strong credits will struggle to perform well if there’s too much supply chasing limited demand.
As more and more AI issuance floods into key credit indices, investors in the passive portfolios that track them are making an increasingly large bet on a concentrated cluster of tech and tech-adjacent borrowers, whether they realise it or not.
A web that’s hard to untangle
Over-supply concerns are intensified by the fact that, on top of the deluge of new bonds from the hyperscalers themselves, chipmakers, data centre operators and utilities linked to the AI buildout are all borrowing from public and private credit markets too. They’re also borrowing from and financing one another, which makes it hard for investors buying their debt to assess their true level of exposure to the sector.
Some of the financing is taking place through increasingly complex structures. Hyperscalers are establishing standalone companies, known as Special Purpose Vehicles (SPVs), to build and operate data centres and borrow against them. Because these vehicles sit off the hyperscalers’ balance sheets, investors are repaid primarily from the cashflows the data centres generate. Long-term lease agreements and capacity commitments from the hyperscalers often underpin these arrangements, helping data centre operators secure cheaper financing. (Meta, for example, has issued bonds both through its corporate balance sheet and via SPVs tied to individual data centre projects.) The resulting maze of guarantees, SPVs and long-term contractual commitments makes it very tricky to gauge exactly how much is being borrowed, who ultimately stands behind that borrowing and where the risks truly sit.
And with a substantial share of data centre financing being funded through private credit and off-balance sheet structures, companies' publicly available balance sheets and bond indices may no longer capture the full financing picture of the buildout. That makes the ultimate distribution of risk harder to spot.
The big credit ratings agencies have become increasingly uneasy about this network of borrowing arrangements, revenue commitments and cross-company dependencies. If one company stumbles, how many others could be dragged down too? In July, S&P downgraded Oracle’s credit rating to BBB- (one notch above high yield), citing concerns about the scale of its borrowing and its dependence on a single major customer (OpenAI) for a significant proportion of its future revenues. These interconnections add extra layers of complexity – and, in our view, extra layers of risk – to the AI borrowing boom.
Here, there and everywhere
Meanwhile, the hyperscalers are increasingly looking beyond the US investment grade market to fund their AI ambitions. In 2026 they’ve issued a lot of new bonds in currencies beyond the dollar, including sterling, euros, Swiss francs, yen and Australian and Canadian dollars, in a bid to make their debt more appealing to overseas investors and secure more favourable borrowing costs. These massive multi-currency deals have set historical records for foreign issuance in several big corporate bond markets, and hyperscaler debt now makes up a much bigger slice of overall corporate bond issuance in lots of countries beyond the US.
As the AI-related share of credit markets grows around the world, so do the risks associated with a downturn in a sector that now influences almost every other capital market to some degree.
Why passive investors should pay attention
For passive bond investors, this creates real challenges. All this borrowing is reshaping the composition of many investment grade corporate bond benchmarks in three ways.
- Concentration. Tech has historically accounted for a modest slice of the investment grade universe. That’s changing fast. AI-related issuance (including the utilities and industrials tied to the buildout) is already the largest sector in the US investment grade index, according to JPMorgan. That suggests that concentration in tech, long an equity-market phenomenon, is creeping into bonds too. Extra-worrying given that many investors expect their bond exposure to provide effective diversification from their equity allocations.
- Duration. The AI giants favour jumbo deals that are mostly long-dated (meaning the debt won’t be repaid for many years to come), including substantial issuance at 30 years or more. As these bonds flood into the major credit indices, they lift the tech sector’s weighting and stretch the average duration of the benchmarks at the same time. That adds a meaningful chunk of interest-rate sensitivity to the passive portfolios that track them. Longer-dated bonds are more sensitive to movements in both government bond yields and credit spreads, so small changes in market conditions can have a particularly large impact on their prices. Passive investors are therefore becoming exposed not only to tech sector-concentration, but also to greater duration risk. The shift shows up clearly in DV01, the measure of how much a bond’s price moves if yields shift by 0.01%. Goldman Sachs research showed that, at the start of May, Oracle had become the largest risk-adjusted name in the US investment grade universe, with Amazon fifth and Meta eighth – a dramatic turnaround from 19th and 51st respectively a year earlier.
- Visibility. With so much AI-related financing moving into private credit and off-balance sheet vehicles, a tracker can only buy what the index can see. Passive investors may be taking on risks that aren't obvious from the benchmark itself.
The opportunity for active investors
Recent market moves suggest investors are becoming more cautious. The relentless flow of new supply has driven bouts of spread widening across parts of the hyperscaler and AI-adjacent credit universe even as broader credit markets have remained relatively resilient. That’s left many AI-related bonds trading materially wider than where they were issued. Uncomfortable news for passive investors, who had to buy them at higher prices when they entered the major credit benchmarks. Combined with considerable uncertainty about the size of the future AI pipeline, this weak performance in secondary markets has meant the hyperscalers' more recent new issues have attracted less demand, and they've had to pay higher borrowing costs to get these deals away.
AI financing needs continue to rise, while investors’ capacity to absorb more and more long-dated debt is not limitless. There’s a big risk that supply will remain the dominant force driving AI-related bond prices in coming years, regardless of how strong company fundamentals remain.
That makes active management particularly valuable. Active bond investors get paid to assess more than just default risks. It's also our job to judge whether technical supply pressures could overwhelm otherwise solid credit fundamentals. Rather than blindly accepting benchmark concentration, we can choose where we believe spread compensation adequately reflects the risks. That’s why we've steered clear of long-dated hyperscaler and hyperscaler-adjacent debt: the credit fundamentals look solid, but the technical risks don’t stack up.
The AI infrastructure boom may end up transforming the global economy. But with no end in sight to the AI borrowing splurge, the case for actively managing your bond exposures is stronger than ever.
