AI hyperscalers drove 47% of 2026 sterling bond issuance so far

Five large technology companies supplied 47% of new sterling corporate bond issuance in 2026 so far, the Bank of England's Financial Policy Committee (FPC) reported on 30 September 2026. The companies named are Meta, Alphabet, Amazon, Microsoft and Oracle — the same five firms the Bank classes as "AI hyperscalers" in its July 2026 Financial Stability Report. None is a UK company, and the Bank has not published the sterling amount behind the 47% figure or the exact cut-off date for "year to date".

The reason this matters to a UK reader is not that these firms are in difficulty. On the Bank's own account, most hold credit ratings of AA− or higher and carry low debt relative to their earnings. The reason it matters is concentration: when a handful of borrowers account for nearly half of new supply in a market, whichever UK investors, bond funds or pension schemes hold that supply may become more exposed, collectively, to changes in those issuers' creditworthiness or bond valuations if AI spending and earnings expectations deteriorate — not to the AI spending itself, since bond proceeds are typically fungible and the public record does not establish how much of this debt specifically financed AI infrastructure — though the public record also does not identify the actual buyers or quantify how much of this debt UK institutions hold.

The FPC's own framing is conditional, not predictive. Its concern is that if AI earnings, adoption or productivity expectations disappoint, the debt, private credit and structured financing now underpinning the sector's expansion could amplify losses and spread a repricing beyond AI-related assets into wider UK financing markets. The Bank has not concluded that a correction is imminent or that AI valuations are in a bubble.

What the 47% figure measures — and what it doesn't

The statistic describes a flow, not a stock. It is the hyperscalers' share of new sterling corporate bond issuance during 2026 to date, not their share of sterling corporate debt already outstanding. A comparable US figure illustrates the gap between the two: at the end of 2025, the five hyperscalers represented just 3% of outstanding US investment-grade debt, yet by early May 2026 they accounted for more than 15% of that year's investment-grade issuance. A high share of new borrowing can sit on top of a much smaller existing debt pile.

The figure also describes currency, not nationality. These are mainly US-domiciled companies choosing to borrow in pounds, among other currencies, for reasons the public record does not explain. The Bank notes separately that hyperscaler issuance in sterling markets remains significantly smaller in absolute terms than their issuance in US dollars and euros — so 47% is a large share of a comparatively small sterling pool, not evidence that sterling is their primary funding currency.

Several details that would help a reader judge the scale of this exposure are not public. The Bank has not disclosed the sterling amount underlying the 47%, the exact data cut-off, or whether the denominator covers all sterling corporate bonds, investment-grade issuance only, or gross issuance before refinancing. It is also not established how much of the proceeds are earmarked for AI infrastructure rather than general corporate purposes — bond proceeds are typically fungible unless a prospectus restricts their use. Nor is it known who holds these bonds: UK pension funds, insurers, asset managers, overseas institutions, or some mix. Without that information, exposure within the UK financial system cannot be quantified from the public record.

Why hyperscalers are turning to debt

Hyperscalers have historically funded data centres, chips and related infrastructure largely from their own cash flow. The Bank's reporting points to a shift towards external finance as AI capital spending has scaled up. Third-party estimates relayed by the FPC give a sense of the pace, though these are forecasts and estimates from investment banks whose underlying methodologies and issuer universes are not public, and they do not represent completed financing.

The OECD, also cited by the Bank, estimated that private credit's share of AI investment financing rose from 9% in 2024 to 34% in 2025. Separately, AI-related issuers accounted for 41% of non-refinancing US high-yield bond issuance in 2026 to the Bank's July cut-off, despite representing only 1% of the JPMorgan US high-yield bond index at the end of 2025 — a US market example, but one that illustrates the same pattern of a small group of issuers becoming disproportionately prominent in new supply across several debt markets at once, not only in sterling.

Why concentration in a bond market can matter

A market can absorb heavy issuance from a few large, highly rated borrowers without strain for a long time — and the Bank says that, so far, it has. The risk the FPC is flagging is not about today's market functioning but about what happens if the underlying AI earnings story changes. If expectations for AI revenue, adoption or productivity gains were revised down, several things could happen together: losses for current holders of the bonds as prices fall and spreads widen; correlated selling if many investors try to reduce AI-linked exposure at the same time; wider borrowing costs across the sterling corporate bond market generally, if investors demand more compensation for holding any corporate credit; and, in a more severe scenario, reduced credit availability for other UK businesses if investor appetite for corporate debt pulls back broadly rather than only from AI names.

The Bank also points to a channel beyond corporate credit. A reassessment of AI productivity expectations could feed into weaker growth expectations more broadly, which can affect sovereign debt markets through changed fiscal expectations or shifts in term premia — a macro channel distinct from any direct link between sterling corporate issuance and gilt pricing.

How risk could reach UK investors

Treasury Committee evidence taken on 14 July 2026 sets out two distinct UK transmission routes. FPC external member Stephen Blyth told MPs that debt, rather than equity or cash flow, could make the downside of an AI correction more consequential. Separately, Governor Andrew Bailey said that a correction centred outside the UK — for instance, in US equity or credit markets — would still affect the UK through globally connected markets, even without direct UK exposure to the specific assets that fell.

Neither statement amounts to a forecast that a correction will happen, or a finding that it has already caused losses to current sterling bondholders. Blyth told MPs it was not clear the market was necessarily a bubble, and that the probability of a valuation reset was hard to assess.

What limits the immediate danger

Several FPC findings, all dated 7 July 2026 unless noted, work against treating 47% as an imminent threat. The stock of AI-related corporate debt entering 2026 was modest. Most hyperscalers carry strong ratings — AA− or higher — and comparatively low leverage. Issuance has, in the Bank's words, been "readily absorbed" by the market. And as of that date, the Bank found little evidence that AI-related borrowing was crowding other businesses or governments out of credit markets, though this is a point-in-time judgement rather than a guarantee against crowding out as issuance volumes grow.

A live test came in July 2026, when AI and semiconductor shares fell sharply. The FPC's September record says market functioning remained orderly through that episode, with no spillover into core markets and no broader systemic stress — evidence that a correction in AI-linked assets can occur without automatically becoming a financial-stability event, though the FPC continued to treat the risk of a sharper future correction as unresolved. On 30 September 2026 the FPC also maintained the UK countercyclical capital buffer — the extra capital buffer UK banks must hold against the risk of a downturn — at 2%, and judged the UK banking system appropriately capitalised and liquid. It announced no AI-specific capital measure and did not identify which, if any, UK banks hold hyperscaler bonds.

The opacity problem

The FPC's headline judgement in September flagged leverage, opaque structures and circular financing arrangements as factors that could make risks harder to assess and losses harder to contain if AI expectations disappoint. This points towards private credit funds, special-purpose financing vehicles and similar structures increasingly used alongside public bond markets to fund data-centre build-out. The Bank describes these as a feature of the financing, not as evidence of wrongdoing, and the public record does not say how widespread they are or quantify their use. That absence of detail is itself part of the concern: a shock that moves through structures investors cannot easily see is harder to size in advance.

What to watch next

The clearest gap in the public record is the sterling amount behind the 47% figure, its exact cut-off date, and the precise scope of "sterling corporate bond issuance" it covers. Readers who want the underlying detail should look to the Bank of England's own publications — the Financial Policy Committee's records and Financial Stability Reports — rather than third-party commentary, since the Bank has not released the data download or methodology note that would resolve these questions.

Beyond that, the markers likely to shape how this story develops include any rating changes or spread movements on hyperscaler bonds, changes in the hyperscalers' share of sterling corporate bond issuance, subject to the Bank clarifying the denominator and methodology, clearer evidence on who holds the UK-denominated debt, and whether future FPC assessments find crowding-out effects that were absent as of July 2026. The next Financial Stability Report and FPC record will be the primary sources to check for an update.