Interview Questions144

    Hyperscaler Bonds: AI Capex Reshaping IG Issuance

    Hyperscaler bonds topped $228 billion in 2026 as Amazon, Alphabet, Meta and Oracle borrowed for AI capex, and investors began demanding higher yields.

    |
    15 min read
    |
    1 interview question
    |
    Share

    Introduction

    Hyperscaler bonds have become the largest single theme in investment-grade credit. The five companies building most of the world's AI infrastructure (Amazon, Alphabet, Meta, Microsoft and Oracle) raised about $121 billion of bonds in 2025, which Bank of America counted as more than four times their average over the prior five years. That was the opening wave. In 2026 the borrowing roughly doubled: Alphabet, Amazon, Meta and Oracle had issued more than $228 billion of bonds by October 13, according to LSEG data cited by CNBC, with deals in seven currencies from dollars and euros to sterling, Swiss francs and yen.

    The scale changed the market's pricing as well as its composition. Deals that once priced through guidance on record order books now pay visible new-issue concessions, and most 2026 hyperscaler bonds trade at higher yields than where they were sold. This article covers the capex behind the borrowing, the landmark deals, each company's financing strategy, credit quality, investors and what the theme means for DCM bankers.

    The Scale of Hyperscaler Capex

    The hyperscaler capex story has grown from a contained technology-sector phenomenon into one of the largest concentrated capital deployment programs in corporate history.

    Capex Growth Trajectory

    Aggregate hyperscaler capex has grown dramatically:

    YearCombined hyperscaler capex
    2022$162B
    2023~$210B
    2024~$330B
    2025$448B
    2026E~$810-840B (latest company guidance)

    The 2026 figure sums each company's latest guidance and is roughly 1.8x the 2025 level and about five times the 2022 level. JPMorgan's estimate of about $800 billion for 2026 lands in the same place, and the bank expects annual spending to exceed $1 trillion in 2027.

    2026 Individual Capex Guidance

    The latest company guidance, mostly from second-quarter results in July 2026, moved higher almost across the board:

    Company2026 capex guidancePeriod
    Amazon~$220B (from $200B)Calendar 2026
    Alphabet$195-205B (from $180-190B)Calendar 2026
    Microsoft~$175B (after lease reclassification)Calendar 2026
    Meta$130-145B (from $125-145B)Calendar 2026
    Oracle$90-95B grossFiscal year to May 2027
    Combined~$810-840BMixed periods

    Two cautions apply. Microsoft's figure fell only because it lengthened the useful life of its buildings and moved more future data center leases from finance to operating leases, per CNBC's report on its fiscal fourth-quarter call, and Oracle reports on a fiscal year ending in May. Amazon attributed its $20 billion increase mainly to higher memory prices, according to CNBC. The company detail and the M&A the buildout generates are covered in the AI investment cycle and infrastructure buildout.

    What Drives the Capex

    The spending rests on reinforcing drivers: training compute for ever-larger foundation models, inference capacity to serve AI products at scale, enterprise cloud demand, a competitive arms race among the five, and data center expansion abroad to meet data residency and latency needs.

    Funding Sources

    Hyperscalers fund capex through internal cash flow and external capital, and in 2026 the balance shifted decisively toward the capital markets:

    1. 1.Operating cash flow: Still the largest source. Goldman Sachs projects about $778 billion of 2026 operating cash flow for the group against about $750 billion of capex, per Reuters, which leaves little for buybacks and dividends without outside funding
    2. 2.Debt issuance: Goldman expects bonds to fund about a third of 2026 capex and about 35% in 2027
    3. 3.Equity and equity-linked capital: New in 2026, as Oracle and Alphabet paired bonds with common stock and mandatory convertibles

    The squeeze shows in free cash flow, the metric bond investors now watch most closely. Amazon's trailing twelve-month free cash flow swung to an outflow of $7.6 billion by June 2026, and Meta generated just $784 million in the second quarter. Microsoft remains the exception, expecting positive free cash flow in fiscal 2027, while Oracle has the heaviest external funding need.

    Hyperscaler Bond

    A bond issued by one of the major cloud computing and AI infrastructure providers (Amazon, Alphabet, Meta, Microsoft, Oracle) to fund AI-related capital expenditure. The category emerged as a distinct sub-segment of US IG corporate bonds in 2025, with about $121 billion issued, and grew into one of the market's largest sources of supply in 2026. Hyperscaler bonds typically feature very large multi-tranche deals ($15-37 billion per dollar offering), ratings from AAA down to BBB-, multi-currency execution, and ultra-long tenors of 30, 40 and 50 years, including one 100-year sterling bond. Because a handful of similar issuers supply so much paper, the category concentrates risk in IG indices.

    The Defining 2025 and 2026 Transactions

    A handful of mega-transactions established the category in 2025; a larger set in 2026 tested how much the market could absorb.

    2025: The Opening Wave

    The 2025 deals set the template of jumbo size, long tenors and heavy oversubscription:

    1. 1.Oracle, September: $18 billion, the second-largest IG deal of the year and notable because Oracle's capex already exceeded its operating cash flow
    2. 2.Meta, October: $30 billion across six tranches out to 40 years on a $125 billion order book, at the time the largest IG deal not tied to M&A
    3. 3.Alphabet, November: a $17.5 billion dollar offering with a 50-year tranche plus a €6.5 billion euro deal, about $25 billion in total, per Cleary Gottlieb's deal note
    4. 4.Amazon, November: $15 billion, a preview of the much larger program it ran in 2026

    Amazon: $54 Billion in March 2026

    Amazon went from occasional issuer to one of the market's largest borrowers. In March it sold $37 billion in 11 US tranches from 2 to 50 years on about $126 billion of orders, the fourth-largest US corporate bond sale and the biggest not tied to an acquisition, then followed with a €14.5 billion euro debut, about $54 billion in total, per Bloomberg. Its July return showed how quickly investor demand cooled: a $25 billion eight-tranche deal paid 18 to 21 basis points of extra yield on its longest bonds, with an order book of about 2.5 times the deal against 3.2 times in March, according to Fortune.

    Alphabet: A Century Bond and Seven Currencies

    Alphabet's February financing raised about $31.5 billion: a $20 billion dollar deal plus inaugural sterling (£5.5 billion) and Swiss franc (CHF 3.055 billion) offerings, per Cleary Gottlieb. The sterling deal included a £1 billion 6.125% note due 2126, the first century bond from a technology company since Motorola in 1997, which drew about £9.5 billion of orders according to TwentyFour Asset Management. Alphabet then added euro, Canadian dollar, yen and Australian dollar deals and, in August, its largest dollar deal yet: $25 billion in 10 tranches out to 40 years on a book that peaked near $115 billion, but with a new-issue concession of 10 to 15 basis points, per IFR.

    Meta and Oracle: Repeat Supply at Wider Spreads

    Oracle opened 2026 on February 2 with a $25 billion eight-part deal from 3 to 40 years, the debt half of a $50 billion debt-and-equity plan built to protect its investment-grade rating, per IFR. Meta returned on April 30 with $25 billion in six tranches, but peak orders of $96 billion compared with about $125 billion for its 2025 deal, and nearly every tranche priced wider than in October, per Bloomberg. In July, Meta's El Paso data center joint venture with BlackRock sold about $12.5 billion of 2048 bonds through the Sopaipilla Investor vehicle at 7.50% on roughly $20 billion of orders, which IFR described as uncomfortably low coverage.

    Issuer2025 headline deals2026 headline deals
    Amazon$15B (Nov)$37B + €14.5B (Mar); $25B (Jul)
    Alphabet$17.5B + €6.5B (Nov)$20B + £5.5B + CHF 3.1B (Feb); $25B (Aug)
    Meta$30B (Oct)$25B (Apr); $12.5B JV bond (Jul)
    Oracle$18B (Sep)$25B (Feb)
    MicrosoftNo new bondsNo new bonds
    Total~$121B (five issuers, BofA)$228B+ (four issuers to Oct 13, LSEG)

    Hyperscaler Financing Strategies

    The five hyperscalers have adopted different financing strategies based on cash flow, balance sheet capacity, and how much equity dilution or off-balance-sheet complexity they will accept.

    Alphabet and Meta: Direct Issuance Plus New Channels

    Both remain heavy direct issuers, but in 2026 each added a second channel. Alphabet diversified by currency to reach investors who rarely buy dollar bonds, and in June raised $20.7 billion of common stock, $19.3 billion of mandatory convertible preferred and a $10 billion placement with Berkshire Hathaway, alongside a $40 billion at-the-market program, per Cleary Gottlieb. The equity-linked side of the AI buildout is covered in the 2026 convertible record. Meta leans on off-balance-sheet joint ventures, in which a partner owns most of a data center campus and the vehicle issues bonds supported by Meta's lease payments and guarantees, a structure explained in project finance bonds.

    Amazon: From Programmatic to Jumbo

    Amazon historically issued programmatically, with periodic benchmarks sized for its diversified retail, AWS and advertising profile. In 2026 it switched to jumbo, multi-currency deals: the March dollar and euro sale, a Swiss franc debut in May and the July dollar deal. The shift reflects a capex budget that now exceeds what even AWS cash generation can cover alone.

    Oracle: Debt-Heavy Necessity

    Oracle is the most debt-reliant hyperscaler. Its February plan split the $50 billion roughly evenly between bonds and equity (common stock and mandatory convertible preferred), and Moody's has said Oracle needs to fund a meaningful share of its AI buildout with equity to keep its investment-grade rating. The equity component signals how close Oracle sits to the limits of bond-market capacity.

    Microsoft: Off-Balance-Sheet Approach

    Microsoft funds most capex from operating cash flow, supplemented by infrastructure partnership structures (where it contributes equity to a partnership that borrows at the partnership level) and data center leases from third-party developers, obligations that sit outside its reported debt. It is the one hyperscaler that stayed out of the bond wave entirely, and its fiscal 2026 annual report shows no proceeds from new debt issuance in fiscal 2025 or 2026.

    Credit Quality and Rating Considerations

    A Wider Ratings Ladder

    The group no longer sits uniformly at the top of the scale. At S&P, Microsoft is AAA, Alphabet AA+, Amazon AA and Meta AA-, while Oracle was cut to BBB-, one notch above high yield, on July 9, 2026, per Fortune; Moody's rates it Baa2 with a negative outlook. Investors now price the credit gap between the strongest and weakest names as much as the group label.

    Rating Implications of Heavy Capex

    The aggressive capex programs are now visible in rating considerations. Oracle's downgrade followed guidance for $90-95 billion of fiscal 2027 capex, and S&P projects negative free cash flow through that year. For the AA names, Moody's has said the investment-grade ratings face no imminent risk, but the leverage trajectory depends on capex converting into commensurate revenue growth.

    Cash Flow Coverage Analysis

    EBITDA-to-interest coverage remains very strong for most hyperscalers (typically 20x+). Debt-to-EBITDA is climbing but stays under about 1.5x for most of the group; Oracle is the outlier, with S&P expecting leverage in the mid-4x area for fiscal 2027, according to IFR.

    The Investor Base for Hyperscaler Bonds

    The hyperscaler bond category has attracted a distinctive investor base, with implications for how the market operates.

    Insurance Companies

    Insurance balance sheets are the anchor buyers, drawn by:

    1. 1.High credit quality matching insurance liability profiles
    2. 2.30-year, 40-year and 50-year tranches matching long-dated liabilities
    3. 3.A yield pickup over Treasuries that widened in 2026

    Pension Funds

    Pension funds with long-duration liabilities join life insurers as the natural buyers of 30-year and longer tranches, which is why the longest parts of each deal skew to these two investor types.

    Bank Treasuries

    Bank treasuries buy top-rated hyperscaler paper in shorter tenors (3-7 years) that match their liquidity needs, holding it as HQLA-eligible liquidity at a haircut rather than as Level 1 assets.

    High-Quality Liquid Assets (HQLA)

    Assets that bank regulators allow lenders to count toward the liquidity buffers required under Basel III's Liquidity Coverage Ratio, because they can be sold or pledged quickly in stress with little loss of value. HQLA is tiered: Level 1 (cash, central-bank reserves, top-rated government bonds) counts in full, while lower tiers (including some highly rated corporate and structured paper) count at a haircut. Bank treasuries buy short-tenor AAA bonds, such as top-rated hyperscaler paper, partly to hold them as HQLA-eligible liquidity.

    Bond Mutual Funds and ETFs

    IG mutual funds and ETFs such as LQD hold hyperscaler paper in proportion to index weights. Hyperscalers now make up about 4.5% of the ICE BofA US Corporate Index, according to TwentyFour Asset Management, so passive demand has grown alongside active demand.

    The Forward Issuance Trajectory

    Wall Street forecasts for hyperscaler borrowing keep being overtaken.

    Forecasts Overtaken by Issuance

    In November 2025, Bank of America expected about $100 billion more hyperscaler debt in 2026, per the same Fortune report that counted the 2025 total. Four issuers passed $228 billion by mid-October 2026. Forecasters underestimated both how far capex would rise and how much of it would be debt-funded.

    Goldman Sachs: A Record $420 Billion in 2027

    Goldman Sachs expected about $250 billion of hyperscaler bonds in 2026 when Reuters reported its July estimate, with debt funding about 35% of capex in 2027. By September it projected a record $420 billion of gross hyperscaler issuance in 2027, up 60% from its 2026 estimate, per Reuters. The same report found portfolio managers asking for bigger concessions and rethinking concentration limits rather than doubting repayment.

    Implications for IG Market Capacity

    The market has absorbed every deal so far, including $25-37 billion single dollar offerings, but each has cleared at a higher price. Goldman's path tests the market's structural depth, because the core buyers would need to keep raising allocations just as single-issuer limits start to bind.

    Comparing Hyperscalers to Prior Mega-Issuance Themes

    Prior concentrated issuance waves provide useful context.

    Telecom Buildout Era (Late 1990s)

    The late 1990s telecom buildout produced massive concentrated bond issuance from carriers building out fiber and broadband infrastructure. The cycle ended badly with telecom defaults in 2001-2002 producing some of the worst credit losses of the era. The parallels are inexact but instructive: concentrated capex, uncertain demand assumptions, technology risk and concentrated investor exposure.

    M&A Mega-Deal Era (2014-2018)

    The 2014-2018 period featured multiple $25-50 billion M&A-driven IG benchmarks (Verizon's $49 billion Vodafone financing, AT&T's Time Warner financing, Bayer's Monsanto financing) that proved the market could absorb mega-transactions. The hyperscaler deals follow the same playbook for capex rather than M&A, and Amazon's March 2026 sale ranks among the largest of either kind.

    Implications

    The 2020-2021 issuance surge was broad-based and driven by cheap money, unlike today's concentrated, capex-driven wave. History suggests concentrated mega-issuance themes can run for 3-7 years before moderating or ending in credit stress, so the hyperscaler wave may have years left, but bankers and investors should manage it with those precedents in mind.

    Implications for the Broader IG Market

    Spread Effects: Pricing Turns Against Issuers

    In 2025 the spread cost of heavy hyperscaler supply was muted by strong overall IG demand. In 2026 it arrived. A Reuters analysis of deals from Amazon, Alphabet, Meta and Oracle, together with later reporting, shows the shift:

    • Concessions: the median new-issue concession on hyperscaler deals rose to 12 bps in 2026 from 2.25 bps in 2025
    • Spreads: the median 5-to-7-year hyperscaler spread rose to 60 bps from 50 bps, and bonds beyond 20 years to 118 bps from 108.5 bps
    • Cover ratios: orders per dollar offered fell from nearly 5x in February to below 2x in July, according to Apollo data cited by Fortune
    • Sector spreads: by September, Goldman data put AI-related issuers at about 115 bps against 78 bps for the broader IG market, per the Reuters report on concentration limits

    Outside AI, IG spreads stayed near historically tight levels, so the widening is a sector-specific premium rather than a broad credit repricing.

    Sector Concentration

    Technology and communications issuers supplied 15-20% of 2025 US IG volume, an unusual concentration. The hyperscaler weight in the ICE BofA US Corporate Index is only about 4.5%, but their bonds skew far longer than the index: TwentyFour puts 31% of hyperscaler debt in the 26-year-plus maturity bucket, against 13% for the index as a whole.

    Long-Duration Demand

    Hyperscalers placed 41% of their 2025-2026 issuance beyond 15 years, against 16% for the ICE BofA US Corporate Index, according to TwentyFour Asset Management. That long-end supply competes with Treasuries for liability-matching money, and it arrived as the 30-year Treasury reached its highest level since 2007 in August 2026. TwentyFour treats it as a marginal technical rather than the main driver of long yields, a distinction explored in the 2026 Fed hike and the bond market.

    Investor Portfolio Considerations

    IG portfolio managers now face binding concentration limits. Many pension and insurance investors cap a single issuer at roughly 2-3% of assets, and Wellington has said some institutions are nearing those limits, with debt from parent-backed data center vehicles aggregated back to the parent company.

    How DCM Bankers Approach Hyperscaler Coverage

    The scale and pace of the AI capex theme demand distinctive DCM coverage.

    Dedicated Coverage Teams

    Major bulge brackets maintain dedicated technology DCM teams covering the hyperscaler clients, working in close partnership with the broader technology corporate coverage teams that own the underlying client relationships. The DCM teams need a working grasp of AI infrastructure economics and each client's capex plan.

    Strategic Capital Markets Advisory

    Beyond individual deals, DCM teams provide strategic capital markets advisory: multi-year debt capacity, tenor mix, currency strategy (now central as Alphabet and Amazon tap euro, sterling, Swiss franc and yen investors) and rating strategy as leverage profiles diverge.

    Cross-Product Coordination

    Hyperscaler engagements now span bonds, bank facilities, commercial paper, mandatory convertibles, common equity and joint venture bonds for individual data center campuses. Effective coverage requires coordinating across product specialists rather than relying solely on bond-focused engagement.

    Forward Pipeline Management

    DCM teams help hyperscalers plan multi-year issuance pipelines that rotate across currencies and tenors and manage maturity profiles. With concessions rising, sequencing deals so that two hyperscalers do not hit the same window has become part of the pricing advice.

    The hyperscaler bond phenomenon is the defining structural story of IG credit across 2025 and 2026 and a critical topic for any DCM banker covering technology issuers or IG investors. The next article turns to the broader refinancing wave and maturity walls facing issuers into 2026.

    Interview Questions

    1
    Question #1Medium

    What's the AI-capex / hyperscaler bond theme?

    The hyperscalers (Microsoft, Alphabet, Amazon, Meta, Oracle) have turned to the bond market in size to fund AI data-center and infrastructure capex that now exceeds even their large operating cash flows. That drove a surge in high-grade issuance, including very large multi-tranche deals and ultra-long tenors, and attracted record order books as investors wanted exposure to "AI debt."

    It matters for DCM because it has become a major source of IG supply and a live example of capex-driven issuance and ultra-long-tenor demand.

    Rate yourself:

    Explore More

    How to Value a Bank: FIG Valuation Explained

    How to value a bank when EV/EBITDA breaks down: master P/TBV, ROE, the justified P/B formula, and the dividend discount model for FIG interviews.

    July 20, 2026

    Buybacks vs Dividends: How Companies Return Cash

    Buybacks vs dividends explained: how each returns cash to shareholders, the tax and signaling differences, the EPS effect, and when each makes sense.

    June 25, 2026

    Why Are Investment Banking Bonuses So High?

    Why investment banking bonuses are so high: the revenue-per-head economics, the compensation ratio, why pay is bonus-heavy and cyclical, and the real catch.

    May 27, 2026

    Ready to Transform Your Interview Prep?

    Join 5,000+ students preparing smarter

    Join 10,000+ students who have downloaded this resource