Giant tech companies are increasingly turning to international markets as they look to finance trillions of dollars of investment in AI infrastructure over the coming years. And they are doing so at accelerated speeds and scale.
Read this article to understand:
- The scale of fundraising by large cloud service providers (hyperscalers)
- The implications for investors and impact on corporate bond markets
- How markets are increasingly differentiating between these companies
Since we last wrote about ‘hyperscalers’ in the December 2025 edition of Bond Voyage, the pace of investment in new data centres – and the associated scramble for capital – has only accelerated.
In 2025, the ‘big five’ hyperscalers (Alphabet, Amazon, Meta, Microsoft and Oracle) spent approximately US$400bn in total in capital expenditure (capex), largely directed towards data centre build-outs, computing infrastructure and key components such as graphic processing units (GPUs), see Figure 1.
Their latest guidance on capex spending for the 2026 calendar year, points to roughly double that amount at US$800bn. Based on company commentaries, an early look into 2027 indicates expectations of further increases ahead, with some investment bank estimates anticipating a potential move towards, or even beyond, the US$1 trillion mark.
Figure 1: Historical and guided capex by hyperscalers (US$ billion)
Note: Oracle’s financial year-end is on May 31, Microsoft’s is on June 30. The above figures refer to calendar years. In Oracle’s case, December to November period is shown. Estimates for their calendar 2026 capex includes half of their financial year (June 2026 to May 2027) guidance. Projected capex for these hyperscalers is shown using mid-point of their guidance.
Source: Aviva Investors, latest company reports, as of June 2026.
While part of this increase reflects higher component costs driven by exceptional demand and ongoing supply bottlenecks, it is also fuelled by a rapidly expanding ‘contracted cloud backlog’.1 The latest reported combined remaining performance obligations (RPOs) of Amazon, Microsoft, Alphabet and Oracle were above US$2 trillion, up over 25 per cent quarter-on-quarter, and nearly tripling year-on-year.2 Figure 2 shows the current reported backlog.
Figure 2: Reported backlog by hyperscalers (US$ billion)
Note: Oracle’s financial year-end is on May 31, Microsoft’s is on June 30. The figures in this chart refer to calendar years. In Oracle's case, Q126 corresponds to May 2026 financial year-end and Q425 to February 2026 and so on, going back.
Source: Aviva Investors, latest company reports, as of June 2026.
Beyond own data centre developments
Hyperscalers are also increasingly securing long-term leases for future capacity. Such committed but not yet commenced, off-balance sheet leases were estimated at approximately US$800bn in their most recent disclosures, with future lease payments typically used to repay debt raised by the data centre developers for specific projects.
In addition, some of the incremental financing needs stem from both paid and outstanding investment commitments to frontier AI labs – for example by Amazon and Alphabet totalling more than US$120bn, with around US$85bn subject to future milestone-based payments.3
A significant portion of this substantial investment can be funded through strong and growing operating cash flow, alongside sizeable cash balances, reflecting the generally solid investment grade profiles of these companies. However, as spending on the artificial intelligence ecosystem accelerates, many hyperscalers are increasingly turning to previously underused financing sources.
Tapping new sources of finance
In bond markets, this has been evident in record levels of issuance in non-US dollar currencies, with such deals accounting for more than one-third of the roughly US$170bn in hyperscaler issuance as of June 26, 2026, see Figure 3.
Figure 3: Year-to-date hyperscalers’ bond issuance by currency in 2026 (US$ billion)
| USD | EUR | CAD | CHF | GBP | JPY | |
|---|---|---|---|---|---|---|
| Alphabet | 20.0 | 9.0 | 8.5 | 3.1 | 5.5 | 576.5 |
| Amazon | 37.0 | 14.5 | 14.0 | 2.8 | - | - |
| META | 25.0 | - | - | - | - | - |
| Oracle | 25.0 | - | - | - | - | - |
| Microsoft | - | - | - | - | - | - |
Note: We have excluded the recent jumbo fund-raising by Nvidia and SpaceX (US$25bn across several tranches each), who are typically not grouped together with the above-mentioned hyperscalers as their capital spending covers a broad range of other activities.
Source: Aviva Investors, Bloomberg, company announcements, as of June 26, 2026.
For example, following Alphabet’s debut Canadian dollar transaction – CAD 8.5bn multi-tranche issue across maturities up to 30 years – Amazon also launched its first-ever deal in June (CAD 14bn) split across five tranches. These transactions followed the companies’ combined Swiss franc (5.9bn) and euro issuances (23.5bn) and came alongside Alphabet’s sterling (5.5bn) and yen (576.5bn) deals.
Entry into these markets presents both opportunities and challenges for local fixed income investors. On the one hand, expanding the issuer base allows portfolio managers to better meet strong demand from domestic clients for high-quality, home-currency assets. On the other, the sheer scale of these transactions raises concentration concerns.
Alphabet and Amazon are becoming increasingly difficult to ignore
For example, in the Canadian dollar market, Alphabet and Amazon now account for around three per cent of the investment grade corporate universe, including financials, based on bond notional outstanding.4 In the Swiss franc market, their combined share exceeds four per cent.5 As a result, for local currency fixed income investors, these issuers are becoming increasingly difficult to ignore – even for those with a more cautious stance on the broader AI theme.
Markets are also increasingly differentiating between companies based on how well they are performing and how they fund their growth.
Why operational progress and funding plans matter
Alphabet recently announced an US$84.75bn equity capital fundraising programme, increased from US$80bn, about half of which is still available under its at-the-market (ATM) programme.6 Meanwhile, after issuing US$5bn of mandatory convertible preferred stock earlier this year, Oracle has guided towards an additional US$20bn ATM programme in 2026. In contrast, Amazon and Meta have primarily relied on debt markets, while Microsoft has kept a relatively low profile, not coming to market with a major new bond deal in recent years.
Bond valuations among these major players have also meaningfully diverged, reflecting perceptions of relative positioning in the AI race.
Bond valuations have diverged, reflecting perceptions of relative positioning in the AI race
Meta bonds for example, have widened significantly versus peers, as investors grow more concerned about elevated capex, a less diversified business model, and greater uncertainty around AI monetisation. Despite its AA- rating, longer-dated Meta bonds currently trade in line with many BBB-rated corporate issuers.
Importantly, financing the AI infrastructure expansion is not limited to hyperscalers. So-called ‘neocloud’ providers (specialised cloud companies focusing specifically on AI-related workloads) such as CoreWeave, are also active participants, while the market for secured bond issuance to finance data centre projects has grown rapidly across both investment grade and high yield segments. Since Meta’s inaugural transaction in October 2025, issuance in this space is estimated to be approaching US$100bn. Investors therefore need to carefully assess and balance construction and concentration risks, alongside tenant credit quality and the specific terms embedded within lease structures.
While the AI landscape continues to evolve – likely beyond the predictive reach of even the most advanced large language models (LLMs) – it is set to offer significant opportunities for active bond investors increasingly (hyper)focused on this space.
References
- Multi-year customer commitments/contract for cloud and AI services that have not yet been recognised as revenue.
- Remaining Performance Obligations, or RPOs, a measure of revenue backlog, are contracted revenue not yet recognised, including unearned and unbilled revenue.
- Frontier AI labs build the most advanced general-purpose artificial intelligence models available at any given time. These systems can tackle complex tasks, like reasoning, working with different types of information and doing more things on their own.
- Bloomberg, as of July 2, 2026. Note that Amazon and Alphabet represent approximately 0.84 per cent of the broader FTSE Canada Universe Bond Index, which also includes non-corporate issuers, based on market value.
- Bloomberg, as of July 2, 2026.
- An at-the-market (ATM) programme lets a company to continuously sell new shares directly into the market at current market prices.
