Research investment-grade bond issuance by Microsoft, Alphabet, Amazon, Meta, and Oracle from 2024 through the latest available data in 2026.

The five hyperscalers have turned the investment-grade bond market into a core funding pipe for AI capex. From 2020–2024 they were a rounding error; in 2025–2026 they became a primary-market driver, terming out 20- to 100-year money while order-book coverage fell and new-issue concessions rose. Microsoft is the exception that proves the rule: it has barely issued senior unsecured debt and is pushing capacity into lease and project vehicles instead.

The regime change

Until 2024 these companies funded servers, data centers, and chips from operating cash flow. J.P. Morgan Asset Management puts combined gross bond issuance at about $17 billion in 2024, then $109 billion in 2025 and $194 billion in the first half of 2026 alone. Vanguard, using Bloomberg data, shows a 2020–2024 average of roughly $35 billion a year, $93 billion in 2025, and about $132 billion of U.S. bonds through July 31, 2026. The gap between those series is definitional—currency, whether related project debt is included, and cutoff date—not a dispute about the direction. [1] [2]

The mechanism is a cash-flow crossover, not a distress event. UBP estimates 2026 capex of about $820 billion against operating cash flow of about $750 billion. Goldman Sachs, cited by Reuters in late July, had capex near $750 billion versus operating cash flow near $778 billion, and expected the five to issue roughly $250 billion of bonds in 2026 and $400 billion in 2027. Either way, multi-year GPU and power contracts no longer fit inside a single year’s free cash flow, so treasurers are locking long-dated unsecured debt while the rating is still high. [3] [4]

They are also leaving the dollar market on purpose. CPR Asset Management counted about $223 billion of hyperscaler bonds since the start of 2026, of which about $132 billion was in dollars, plus €23.5 billion, 5.9 billion Swiss francs, and £9.75 billion. Dollar IG remains the workhorse; euros, sterling, Swiss francs, Canadian dollars, yen, and Australian dollars are how they avoid saturating one buyer base. [5]

Publicly reported deals

Figures below are gross proceeds of publicly reported senior bond sales. Off-balance-sheet project bonds (Meta’s Hyperion vehicle, Oracle’s Project Jupiter loans, Microsoft-tenanted QTS deals) are noted separately and are not the issuer’s own unsecured debt. Sources disagree on some 2025–2026 aggregates; deal-level amounts are from contemporaneous reports.

Issuer	Date	Size	Tenors / structure	Spread at issuance	Order book
Group	2024	~$17B combined	Light, mostly routine	Not a market event	—
Oracle	Sept. 24, 2025	$18B, 6 parts	5s to 40s (2030–2065)	+75 / +90 / +105 / +115 / +125 / +137 bp	Peak ~$88–90B (~5x)
Meta	Oct. 30, 2025	$30B, 6 parts	5s to 40s	+50 / +70 / +78 / +88 / +98 / +110 bp	~$125B (then a record)
Alphabet	Nov. 3, 2025	$17.5B USD + €6.5B (~$7.5B)	USD 3s to 50s	50-year at +107 bp (IPT +135)	~$90B on the USD leg
Amazon	Nov. 17, 2025	$15B, 6 parts	Out to 40 years	40-year tightened to +85 bp from IPT +115	Peak ~$80B
Oracle	Feb. 2–4, 2026	$25B, 8 parts	3s to 40s, plus a $500M floater	+95 / SOFR+111 / +115 / +130 / +145 / +170 / +180 / +195 bp	Peak ~$129–155B; final books ~$109B
Alphabet	Feb. 9–10, 2026	~$32B equivalent	$20B USD (7 parts, to 40 years); £5.5B (~$7.5B) incl. £1B century; ~$4.0B Swiss francs	USD 3-year +27 bp, 40-year +95 bp; century +120 bp over gilts, 6.125% coupon	USD book ~$95–100B+; century ~10x (£9.5B orders)
Meta	Apr. 30, 2026	$25B, 6 parts	To 2066	2066 at +147 bp (IPT up to +180); October 40-year had been +110	Peak $96B (~3.8x)
Amazon	Mar. 10, 2026	$37B USD + €14.5B (~$16.6B)	USD 2s to 50s (2076); euro to 38 years	2076 at +130 bp (IPT +155)	USD peak ~$126B (~3.4x)
Amazon	July 7, 2026	$25B, 8 parts	3s to 40s (2066)	Longest tightened ~20 bp to about +125 bp; 18–21 bp concession on the long end	Peak $62B, settled near $41B (~1.6–2.5x)
Alphabet	Aug. 6, 2026	$25B, 10 parts	2s to 40s (2066), incl. floaters	Coupons 4.50% (2028) to 6.50% (2066); final spreads not fully public	Peak ~$115B
Amazon	Sept. 9, 2026	£4.25B ($5.76B)	3s, 6s, 12s, 19s	+53 bp (3-year) to +93 bp (19-year) over gilts; yields ~5.2% to 6.7%	>£10.65B (~2.5x)
Microsoft	2025–2026	No jumbo senior unsecured sale	—	—	Still the AAA outlier

Oracle’s September 2025 coupons were 4.45% (2030, $3B), 4.80% (2032, $3B), 5.20% (2035, $4B), 5.875% (2045, $2.5B), 5.95% (2055, $3.5B), and 6.10% (2065, $2B). The February 2026 print landed 40–58 bp wider than that curve. Meta’s October 2025 coupons were 4.20% ($4B fives), 4.60% ($4B sevens), 4.875% ($6.5B tens), 5.50% ($4.5B twenties), 5.625% ($6.5B thirties), and 5.75% ($4.5B forties), with final books of $15.8B, $20.5B, $24.3B, $15.4B, $23.8B, and $16.5B and concessions of 10–15 bp. [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25]

One reporting conflict: some secondary write-ups call Oracle’s February deal $30 billion. Primary syndicate accounts and the filing detail a $25 billion note offering, paired with an equity and mandatory-convertible plan that together targeted $45–50 billion for 2026. [26] [27]

Not in the table, but part of the same funding stack: Meta’s October 2025 Blue Owl joint-venture bond of about $27 billion for the Louisiana Hyperion campus (debt sits in a vehicle, not on Meta’s unsecured curve); roughly $18 billion of Oracle Project Jupiter loans quoted in September 2026 at 89–91 cents; and Microsoft-linked QTS paper—$4.6 billion in April 2026 at Treasuries plus 137.5 bp (Baa2, peak demand $12.5 billion) and a Project Odyssey deal marketed around $3.9 billion in August at a high-6% coupon and a low-to-mid-7% yield, investment-grade on rating but junk-like on yield. [28] [29] [30] [31]

Amazon is the volume leader. By early September 2026 it had sold the equivalent of more than $92 billion across dollars, euros (a record €14.5 billion corporate deal), Swiss francs, Canadian dollars (a record C$14 billion, about $10.3 billion), and sterling. Alphabet set local records in sterling, Swiss francs, yen (¥576.5 billion, about $3.6 billion, the largest yen bond by a non-Japanese issuer), and Australian dollars. [32] [33] [34]

How the paper is priced, and why books are shrinking

The product design is deliberate duration. Across 2025 and 2026 supply, hyperscalers placed 41% of issuance beyond 15 years, versus 16% for the ICE BofA U.S. Corporate Index, and their weight in the 26-year-plus bucket is about 31% versus 13% for the broad market. Amazon has gone to 2076; Alphabet sold a 2126 sterling century, the first technology century bond since Motorola in 1997. That matches a 15- to 20-year data-center lease and power contract better than a 5-year bond, and it parks supply with insurers and pensions rather than total-return credit funds. [25] [35]

The cost of that access has risen inside a still-tight investment-grade market. Reuters’ read of LSEG data showed median spreads for Amazon, Alphabet, Meta, and Oracle moving from 30 bp to 40 bp on 2- to 4-year bonds, 50 bp to 60 bp on 5- to 7-year bonds, and 108.5 bp to 118 bp on bonds longer than 20 years, comparing 2025 with 2026 through late July. The median deal-level concession rose from 2.25 bp in 2025 to 12 bp in 2026. Of 91 hyperscaler bonds issued in 2026 with comparable pricing, 78 were trading at higher yields on July 28 than at issue; the median gap was about 22 bp. UBP’s count as of September 3 was that just over half of 2026 bonds from Oracle, Meta, Amazon, and Alphabet were at least 5 bp wider than the new-issue spread, while the broader IG market sat near its tightest levels since the financial crisis. [4] [36]

Coverage tells the same story in sequence. Meta’s October 2025 book was about 4.2 times. Oracle’s February book peaked near 5 times. Amazon’s March dollar book was 3.4 times. Alphabet’s August book was still about 4.6 times ($115 billion on $25 billion). Amazon’s July deal fell to roughly 1.6–2.5 times, and its September sterling book was about 2.5 times, versus roughly 5 times on Alphabet’s February sterling sale. Apollo data cited in September showed hyperscaler cover ratios falling from about 5 times in February to about 2 times in July. The market is still clearing jumbo deals; it is no longer clearing them for free. [1] [23]

Oracle is where secondary trading has detached from the rating. On October 2, 2026, the 6.55% bonds due February 2046 were at 84.91 cents, a yield to maturity of 8.11%, about 247 bp over the 20-year Treasury (5.64% on October 1). Every Oracle bond due after 2044 was yielding between 7.85% and 8.44%. The 6.70% 2056s crossed 8%. MarketAxess noted 30-year spreads had widened about 150 bp since the September 2025 deal and were trading wide of several BB names. Oracle’s own August 31, 2026 10-Q marked $125.0 billion of senior notes at a fair value of $105.7 billion, about 84.6 cents. Five-year CDS hit roughly 203 bp in July, an 18-year high, and set fresh records in late September. [37] [38] [29] [39]

Ratings: four fortresses, one cliff

Microsoft is AAA/Aaa with stable outlooks, one of two U.S. public companies still triple-A at both S&P and Moody’s, and rated above the U.S. sovereign. It is also the only hyperscaler that, as of early September 2026, had not termed out 2026–27 capex in its own bond curve. The implication is mechanical: the next large long-end print the market does not yet have in the price is Microsoft’s, if and when cash flow no longer covers the build. [40] [25] [41]

Moody’s affirmed Alphabet at Aa2, outlook stable, on March 30, 2026. Meta’s October 2025 deal was Aa3/AA−. On the April 2026 sale, S&P kept a stable outlook and said leverage should stay “well below” the downgrade threshold for at least two years, while noting that AI investment was “starting to affect credit metrics.” UBP groups Alphabet, Amazon, and Meta as AA− or better, with more than $500 billion of cash among the four highest-rated names, and argues they could add another $400 billion of debt over 18 months without a rating event, and about $600 billion before a slide to single-A. That is capacity, not a forecast of issuance. [42] [43] [16] [3] [36]

Oracle is the break in the complex. S&P cut the issuer from BBB to BBB− on July 9, 2026, outlook stable, and cut short-term and commercial paper from A-2 to A-3. The agency said it had underestimated the scale of AI investment, that the infrastructure business is diluting a previously strong business-risk profile, and that adjusted leverage stays above 4x. It forecast a fiscal 2027 free-operating-cash-flow deficit of nearly $42 billion and flagged OpenAI as a key credit risk—about half of a roughly $638 billion remaining-performance-obligation backlog at the time of the review. Moody’s is Baa2 with a negative outlook; Fitch is BBB. A further notch at S&P would push a stock of bonds that secondary commentary puts near $120 billion out of investment-grade indexes. That forced-selling channel, not the coupon, is the systemic piece. [44] [45] [46] [29]

Moody’s July 2026 note is the cross-issuer view: capex of $785 billion in 2026 and about $1 trillion in 2027 is eroding free cash flow, and stock sales plus off-balance-sheet structures “threaten credit quality” at Microsoft, Amazon, Alphabet, Meta, Oracle, and CoreWeave. It also said the four strongest balance sheets are unlikely to lose investment-grade status imminently. Pressure is concentrated at Oracle and at high-yield neoclouds. [47]

Index math is already moving. By August, Amazon was about 1.4% of the Bloomberg U.S. Corporate Index (fifth-largest issuer) and Oracle about 1.3% (eighth). UBP puts the five hyperscalers plus Nvidia at 4.6% of the U.S. IG index. That is large enough to reprice the long end of credit, not large enough to “break” a multi-trillion market—Pimco found no statistically significant rise in 10-year Treasury yields around the past six mega AI offerings. [48] [3] [49]

What share of AI capex this channel funds

Gross bonds are not the same as net capex funding. Companies still pay for most of the build with operating cash flow; bonds also refinance, support buybacks, and pre-fund years that have not been spent. Leases, project finance, and chip-backed structures sit outside the unsecured tally. UBP puts off-balance-sheet leases, purchase commitments, and guarantees near $3 trillion; Oracle alone disclosed $288 billion of additional data-center lease commitments not yet on the balance sheet. [3] [29]

The cleanest sourced ratio is Goldman’s, via Reuters on July 29, 2026: hyperscaler debt issuance equivalent to about one-third of 2026 capital spending and about 35% in 2027, on an expected $250 billion of bond issuance this year against roughly $750 billion of capex, and $400 billion of issuance in 2027. [4]

A range, not a point estimate:

• 2026 capex. Goldman via Yahoo Finance in late September: about $800 billion this year and $1.2 trillion next year. UBP: about $820 billion. Moody’s: $785 billion, then about $1 trillion in 2027. Use $750–820 billion for 2026. [49] [3] [47]
• 2026 bond issuance by the five. Realized totals depend on the cutoff. Capital Group counted $240.7 billion year-to-date through August 31 across currencies, of which $66 billion was non-dollar, and expected more than $25 billion more from the five in the last four months of 2026. CPRAM had about $223 billion since the start of the year. Goldman’s full-year marker is about $250 billion. A full-year gross figure around $250–280 billion is the reasonable band if late-year supply arrives as those desks expect. [50] [5] [4]
• Implied share. $250–280 billion of gross bonds against $750–820 billion of capex is about 30–35%. That matches Goldman’s one-third. It is an upper-bound description of this channel’s role in the funding mix, not a claim that a third of every dollar of servers was borrowed.

The cash gap is smaller than the bond print. On UBP’s $820 billion versus $750 billion, the operating-cash shortfall is about $70 billion—so a large slice of 2026 issuance is terming out future spend, replacing commercial paper, or sitting as liquidity, not plugging a same-year hole. Goldman’s earlier July snapshot even had operating cash flow slightly above capex. The strategic point is forward: 2027 capex estimates of $1.0–1.3 trillion (UBP) or $1.2 trillion (Goldman) against issuance forecasts of about $400 billion (Goldman) to $420 billion (Goldman, cited by Reuters as a 60% rise from 2026) keep the bond share in the mid-30s even if cash generation grows. [51] [1]

Wider AI-linked borrowing is a different numerator. The IIF said global bond issuance by AI-linked companies had already passed $400 billion in 2026 and was running above a $500 billion annualized pace, with U.S. companies about 90% of that. Goldman has put investor financing to AI-linked groups near $500 billion this year, of which the five hyperscalers are about $200 billion. Those figures include chipmakers, data-center developers, utilities, and private credit. They should not be divided into hyperscaler capex. [49] [51]

What this means if you are trying to fund, or lend against, the same build

The unsecured IG bid is still open for AA balance sheets, but the clearing price is now set by supply technicals, not by the last deal’s spread. Each jumbo print has cheapened the prior one: Amazon’s July $25 billion sale pushed an existing 30-year about 20 bp wider. Issuers that can still diversify currency—and Alphabet and Amazon have shown they can set records in five markets in one year—will pay less than issuers that have to keep returning to the same 30-year dollar buyer. [52]

Microsoft’s absence is the residual supply risk. A first large, long-dated Microsoft deal would land into a long end that is already 2.5 times overweight hyperscaler paper versus the corporate index. The market has demonstrated it can take $25–37 billion from AA names; it has not demonstrated it can take that from the AAA name on top of another Oracle or Amazon print in the same month.

Oracle is the template for what happens when the same capex story meets a BBB− rating, customer concentration, and negative free cash flow. The bonds still clear—February’s book was the largest of the cycle—but they now trade with high-yield yields and high-yield liquidity behavior while remaining inside IG portfolios. The binding constraint is no longer coupon capacity. It is the index bid that disappears if S&P moves one more notch, and the lease and project-finance markets that are being asked to hold what the unsecured curve will not.

---

Recent Findings Supplement (October 2026)

Alphabet and Amazon have led a surge in multi-currency, multi-tranche investment-grade bond issuance since mid-2026 to fund AI-driven capex, with deals often oversubscribed but showing signs of cooling demand and wider spreads amid rising Treasury yields and heavy supply. Microsoft has signaled a major upcoming issuance, while Meta tapped the market again and plans euro entry; Oracle has faced secondary-market pressure and ratings pressure without prominent new IG bond prints in the period. Bonds represent a growing but partial share of the financing mix for the five companies’ combined ~$820 billion 2026 capex (versus ~$750 billion operating cash flow), supplemented by equity, project finance, and other channels.[1][2]

Alphabet’s Record Multi-Tranche Deals

Alphabet executed two large USD offerings in 2026 (post-April), extending its curve to ultra-long tenors while maintaining strong (though not record) demand. These support its elevated capex guidance (raised to $195–205 billion for 2026).[3]

• October 2026 ($20 billion, 7 tranches, 2029–2066 maturities): Priced with yields from ~3.829% (3-year, +27 bp spread) to 5.794% (40-year, +95 bp spread). Net proceeds ~$19.85 billion. Ratings Aa2/AA+ (stable). Joint coordinators BofA, Goldman, JPM.[4]
• August 2026 ($25 billion, 10 tranches including floaters, 2028–2066): Coupons 4.5%–6.5%; spreads 33–130 bp over Treasuries (tightened 25–27 bp from IPTs). Strong demand supported the size; part of ~$88 billion+ issuance since late 2025.[5][3]

Implication for competitors: Alphabet’s ability to issue across currencies and tenors (including prior euro, sterling, etc.) demonstrates broad investor appetite for AA-rated hyperscaler paper, but repeated large prints are testing absorption capacity and contributing to modestly wider AI-issuer spreads (~115 bp vs. ~78 bp broader IG).[6]

Amazon’s Diversified Currency Push and Upcoming Mega-Deal

Amazon has been the most prolific issuer, adding sterling and planning simultaneous USD/euro raises while its 2026 capex guidance sits around $200–220 billion.[7]

• October 2026 launch (up to $42 billion total): Targeting $25–30 billion USD (up to 11 tranches, 2–50 years) plus up to €10 billion euro (up to 8 tranches, 2–38 years). One of the largest-ever corporate offerings; initial talks for longest tranche ~+155 bp over Treasuries.[8]
• September 2026 (£4.25 billion / ~$5.76 billion sterling debut, 4 tranches, 3–19 years): Demand >£10.65–12 billion (2.5x+ coverage). Yields ~5.2% (3y) to 6.7% (19y). Brings 2026 borrowing past $92–98 billion equivalent.[9][7]
• July 2026 ($25 billion USD, 8 tranches, 2029–2066): Peak demand $62 billion (~2.5x); fixed coupons ~4.6%–6.25%. Used for general corporate purposes including capex.[10][11]

Implication: Amazon’s multi-currency strategy (euro, Swiss franc, CAD, now sterling) diversifies funding and taps local demand, but lower coverage ratios versus earlier 2026 deals signal potential investor fatigue with sustained hyperscaler supply.[6]

Meta’s Continued USD Access and Euro Plans

Meta followed its large 2025 issuance with another sizable USD deal and is preparing its first euro outing.[12]

• May 2026 ($25 billion, 6 tranches): Demand peaked at $96 billion. Spreads wider than its prior $30 billion deal, reflecting some investor caution on capex ramp (2026 guidance $130–145 billion). Ratings stable (S&P noted leverage expected to stay below downgrade thresholds).[13][14]
• Autumn 2026 euro debut planned: First foray into euro market amid broader hyperscaler shift.[12]

Implication: Meta demonstrates that even with equity-market volatility around AI spending, bond investors remain receptive to its AA-/Aa3 paper, though concessions are increasing.

Microsoft’s Planned Large Issuance and Linked Financing

Microsoft has not completed a major direct IG bond print in the immediate post-April window highlighted in results but announced a substantial upcoming deal and participated in linked project financing.[15]

• September 2026 announcement (at least $14 billion, 7 tranches, 3–40 years): Expected to be one of the year’s largest non-M&A deals; underwriters Barclays and HSBC. Proceeds for general corporate purposes amid expected rate rises. Cash position ~$123 billion.[15][15]
• August 2026 Microsoft-linked QTS “Project Odyssey” data-center bond (~$3.9 billion upsized from initial target): Demand >$8 billion; marketed at junk-like yields (~6% coupon, low-to-mid 7% yield, single-B equivalent) despite expected IG rating (Baa2 from Moody’s on related paper). Finances Georgia facility with Microsoft as tenant.[16][17]

Implication: Microsoft’s pristine credit (often AAA/Aaa) supports low-cost direct issuance when needed, while linked structures allow specialized data-center financing at higher yields.

Oracle’s Secondary-Market Pressure and Mixed Financing Approach

Oracle has relied more on prior issuance, equity, and project structures amid capex intensity and ratings scrutiny, with no prominent new large IG bond deals detailed post-April.[18]

• Secondary trading (as of late Sept/early Oct 2026): Long-dated bonds (e.g., 2046–2056) yielding 7.85%–8.44% (e.g., 6.55% 2046 at ~8.11% YTM); 2056 bonds breached 8% for the first time. 5-year CDS at records. Fair value of ~$125 billion senior notes/borrowings estimated at ~84.6 cents on the dollar.[19][20]
• Ratings: S&P downgraded to BBB– (stable) in July 2026, citing high capex ($90–95 billion FY2027 guidance) and negative FCF trajectory; one notch above junk. Fitch affirmed BBB/stable on earlier notes.[21]
• Financing plans: Expects ~$40 billion in debt + equity raises (including $20 billion ATM equity); earlier $25 billion notes in Feb 2026 referenced. Project Jupiter loans trading at discounts.[19][18]

Implication: Oracle illustrates the limits for lower-rated (BBB) issuers in the group—strong backlog supports the buildout, but market pricing reflects leverage and execution risks more acutely than for AA/AAA peers.

Estimated Bond Share of AI Capex Funding

The five hyperscalers’ combined 2026 capex (~$820 billion) exceeds operating cash flow, creating a funding gap filled partly by bonds.[1]

• Hyperscaler bond issuance: ~$194 billion in H1 2026 (or through July/Aug per varying reports); part of ~$400–500 billion total AI-related financing/debt raised in 2026 so far, with hyperscalers accounting for roughly $200 billion.[6][22][23]
• Bond channel share: Described as covering “most” of the dollar IG financing need, with the balance from other currencies, project finance, leveraged loans, securitization, equity (~$115 billion announced in one period), and off-balance-sheet leases/commitments (~$3 trillion noted in one analysis). One estimate positions hyperscaler USD IG issuance at a substantial portion (~40% of AI debt supply in some breakdowns). Goldman Sachs projects hyperscaler gross debt issuance could reach ~$420 billion in 2027.[6][1][2]

Overall implication: Bonds have become a core, scalable channel (especially IG USD and multi-currency) as capex outpaces cash flow, but absorption depends on continued investor appetite amid wider spreads and competition from Treasuries/other issuers. Companies with stronger ratings (Microsoft, Alphabet, Amazon, Meta) retain more flexibility than Oracle. Future supply could pressure spreads further if capex guidance rises. Data is drawn from post-April 2026 reporting; exact secondary spreads and full order-book details for every tranche are not uniformly disclosed in public sources.