Hyperscaler Bond Spreads in 2026: 7 Deals, and a 37bp Rise at the Long End

Meta borrowed 40-year money at 110 basis points over Treasuries in October 2025. Six months later, with the same Aa3/AA- rating, it paid 147. That 37bp move is the clearest single answer in the 2026 record to whether lenders charge more as hyperscaler debt grows. Hyperscaler bond spreads rose most at the long end, barely moved at 10 years, and split hard by rating.

The short answer. Yes, lenders are charging more as hyperscaler debt grows, but mostly for long maturities. Meta's 10-year spread rose 12bp between deals while its 40-year rose 37bp. Amazon's spreads moved only 5bp to 7bp between March and July 2026. The bigger price difference is still rating: BBB-rated Oracle paid roughly double the AA spread at every tenor.

What a hyperscaler bond spread actually measures

A spread is the extra yield a company pays over a US Treasury of similar maturity. Underwriters quote it in basis points, where 100bp equals one percentage point. A 10-year bond priced at +90bp yields 0.90 points more than the 10-year Treasury it is benchmarked against.

The spread isolates what lenders think of the borrower. The Treasury yield moves with rates and inflation, which no hyperscaler controls. The spread moves with credit risk, supply and demand for that name. If you want to know whether bond investors are worried about AI debt, the spread is the number to read, not the coupon.

Every figure in the main table below comes from a pricing term sheet or a deal report, not from secondary trading. That matters. A spread at issue is what the company actually paid on the day. A spread in the secondary market a month later is what the next buyer pays, and it can drift a long way from the original print.

The long end deserves special attention in 2026. A 40-year bond outlives several generations of the chips it funds. Lenders at that maturity are betting on the company's cash flows decades out, so a rising long-end spread is the purest signal of doubt about the durability of the build. That durability question sits behind every number in the comparison of hyperscaler capex against operating cash flow.

Hyperscaler bond spreads in 2026, deal by deal

Seven US dollar senior unsecured deals are in the table. Six priced in 2026. The seventh, Meta's October 2025 deal, is the baseline that makes Meta's April deal comparable. Where a term sheet published the benchmark Treasury yield, the all-in yield is shown, so you can check that yield minus Treasury equals the spread.

Yields in brackets are re-offer yields from SEC pricing term sheets. Alphabet August figures are coupons from IFR, which sit close to the yield because the notes priced near par. Amazon also sold a 50-year tranche in March at +130bp, left out so the columns match.

How the table was built

Oracle, Meta's April deal and Amazon's March deal come straight from the final pricing term sheets each company filed with the SEC as Form FWP. Those documents list the benchmark Treasury, its yield, the spread and the re-offer yield for every tranche. For Oracle's 10-year, the arithmetic is 4.279% plus 1.45 points, which gives the 5.729% yield on the sheet.

Alphabet's August spreads and coupons come from IFR. Meta's October 2025 spreads come from S&P data reported by The DESK. Amazon's July spreads come from Bloomberg reporting syndicated by Investing.com. Alphabet's February deal is the weakest row: I could verify only the overall range of about +27bp to +95bp, reported by the credit team at Mawer, so the 10-year and 30-year cells say so rather than guess.

Microsoft is absent on purpose. I found no Microsoft US dollar bond deal in 2026 in any source I could open. Its absence is itself a data point: one of the five largest spenders has funded the build without a jumbo bond, at least so far.

Repeat issuers paid more for long hyperscaler debt

The cleanest test of whether lenders charge more as debt grows is the same borrower, same rating, borrowing again. Three companies give you that test in this record. All three paid more on the second deal. How much more depends on the company and on the maturity.

Meta: 12bp more at 10 years, 37bp more at 40

Meta's October 2025 deal was $30 billion. Its April 2026 deal added $25 billion, so Meta sold $55 billion of public bonds in six months. Both carried Aa3/AA- ratings. The 10-year spread rose from +78bp to +90bp. The 30-year went from +98bp to +132bp, and the 40-year from +110bp to +147bp.

That shape is the finding. Lenders were nearly as happy to hold 10-year Meta paper in April as they were in October. They wanted roughly 3 times as much extra compensation to lend for 30 or 40 years. I read that as a judgement about duration risk on AI cash flows, not about Meta's near-term solvency, which no lender in this book appears to doubt.

The all-in number is starker. Meta's April 2066 note yields 6.460%. That is the price of 40-year money for a company rated AA-. Meta also pays a different, higher price for off-balance-sheet structures, covered in the breakdown of the Hyperion bond and who carries its risk.

Amazon: 5bp wider, and half the demand

Amazon's March deal was the largest in this record at $37 billion across 11 tranches. Its July deal added $25 billion. Between them, spreads moved very little: 10-year from +75bp to +80bp, 30-year from +105bp to +110bp, 40-year from +118bp to +125bp. Coverage of the July deal said every tranche priced tighter than initial guidance.

The price barely moved, but the demand did. Bloomberg reported about $126 billion of orders for the March deal. The July deal drew about $62 billion at its peak, per Bloomberg reporting syndicated by Investing.com. Amazon got almost the same price from a book half the size. That is a sign the buyer base was thinner, even if it was still deep enough.

Alphabet's 40-year: from +95bp to +130bp

Alphabet sold $20 billion of US dollar bonds on 9 February 2026, with spreads from about +27bp to +95bp. On 6 August it sold $25 billion across 10 tranches, its largest bond ever, according to IFR. The 40-year priced at +130bp with a 6.50% coupon, 35bp wider than February.

IFR also reported new-issue concessions of 10bp to 15bp on the August deal. A concession is the discount a borrower offers versus its own existing bonds to get a new deal done. An Aa2/AA+ issuer paying a double-digit concession tells you that supply, not credit quality, was doing the pricing. Alphabet's other funding choice this year is examined in the analysis of its $80 billion equity raise.

Investment grade bond spreads are not one price

All five hyperscalers in this record are investment grade. That label covers a wide band. AA-rated Amazon and Alphabet borrow like near-sovereigns. BBB-rated Oracle borrows like a debt-heavy industrial. The table makes the difference measurable.

Oracle priced on 2 February 2026 at +95bp for 3 years, +145bp for 10 years, +180bp for 30 years and +195bp for 40 years. Amazon priced on 10 March at +40bp, +75bp, +105bp and +118bp for the same tenors. Five weeks apart, Oracle paid 55bp to 77bp more at every point on the curve. Mawer's credit team noted that Oracle's book drew more than $127 billion of demand even at those levels.

That rating gap is the part of the story that has worsened since. S&P cut Oracle to BBB- in July 2026, one notch above junk, a move traced line by line in the analysis of what Moody's and S&P said about Oracle. Oracle told the market in February that it would not issue more bonds in 2026, so its next spread will be set in a harder market.

The chart shows something the ratings do not. At 5 years, Meta in April paid +53bp, close to Amazon's +55bp in March. At 40 years, Meta's +147bp sat 29bp above Amazon's +118bp and 48bp below Oracle's +195bp. On long money, the market priced Meta somewhere between its AA- rating and a BBB.

All-in yield narrows the rating gap

Spreads are not what a treasurer pays. Coupons are. Here the Treasury market did part of the work. Oracle's 30-year benchmark yielded 4.914% in February. Meta's yielded 4.990% in April. So Meta's 30-year re-offer yield of 6.310% sat only 40bp below Oracle's 6.714%, despite five rating notches between AA- and BBB.

Alphabet's August 30-year carries a 6.375% coupon. That is 34bp below Oracle's February 30-year yield, for an issuer rated seven notches higher. Treat that comparison as approximate, because a coupon is not a yield. It still makes the practical point: by late summer, the strongest AI borrowers were paying long-term rates that a BBB borrower paid at the start of the year.

Order books are thinning even as deals keep clearing

A spread is a price. The order book is the quantity of money willing to lend at that price. Dividing reported peak demand by deal size gives a rough cover ratio, and the pattern across this record runs one way.

Peak and final books are not the same measure, and order books are inflated by investors who over-order expecting allocation cuts. Meta's April 2026 and Alphabet's February 2026 books are omitted because I could not verify them from a source I opened.

Oracle's 5.2x is the outlier, and it is not a sign of strength. The book was large because the price was generous: a BBB name paying +195bp at 40 years attracts buyers who would never touch a +110bp AA bond. Gimme Credit's Dave Novosel had told IFR before pricing that Oracle would need to offer some very generous spreads. It did, and the book followed.

Amazon's two deals are the cleaner comparison, because the issuer and rating are fixed. Cover fell from 3.4x to 2.5x in four months. Spreads held only because Amazon's $25 billion July deal was still small relative to a $62 billion book. A borrower that needs $40 billion in a single sitting may not get the same treatment.

The market-wide read on hyperscaler bond spreads

Single deals can mislead. The index view confirms the deal-by-deal pattern. J.P. Morgan Asset Management reported in August 2026 that hyperscalers had issued $219 billion of investment grade bonds year to date. Hyperscaler spreads had widened about 30bp over the same period, against 2bp for the broad US investment grade index.

The same note put hyperscaler spreads around 105bp, duration-adjusted, against about 80bp for similarly rated bonds and the broader market. In plain terms, a AA-rated hyperscaler now pays more than a typical company with the same rating. The sector has become a credit risk category of its own, rather than a set of near-risk-free names.

Supply explains much of it. J.P. Morgan estimated that hyperscaler issuance rose from 2% of US investment grade supply between 2022 and 2024 to an expected 9% in 2026. Index funds and insurers have concentration limits per issuer and per sector. When one sector supplies nearly a tenth of new paper, those limits start to bind, and the marginal buyer demands a higher spread.

The note added that hyperscaler-related project finance structures pay around 100bp more than equivalent public hyperscaler bonds when investment grade, and around 200bp more when high yield. That tier is where the money from private credit's exposure to AI debt sits, and it is the far end of the curve that begins with the public bonds in this table. The junk-rated version is covered in the analysis of high-yield data center bonds.

Where this analysis is weakest

The table answers the question in the headline better than any single news story does. It still has four weaknesses you should weigh before quoting it.

The deals are on different days

Spreads move with the whole credit market, not only with the borrower. A comparison between Oracle in February and Amazon in March blends a rating gap with five weeks of market drift. Capital Group reported that the US investment grade index ended August 2026 at 78bp, the same level as January. That limits the drift problem, because the broad market barely moved. It does not remove it.

A spread at issue is not a spread today

Every number here is the price on the day the bonds sold. Secondary spreads can widen or tighten afterwards. A bond that priced at +147bp may trade at +170bp now. If you want the current cost of hyperscaler credit, check secondary levels or credit default swaps, not this table. The table tells you what each issuer agreed to pay, which is a different question.

Seven deals is a small sample

Three repeat-issuer comparisons is the entire evidence base for the headline finding. Each could have another explanation. Meta's April deal came the day after it raised 2026 capex guidance by $10 billion, which may have worried lenders more than the deal size. Alphabet's August deal was larger than its February one, which by itself justifies some extra concession. The pattern is consistent, but it is not proof of cause.

The strongest counter-case is Amazon. It sold $62 billion of US dollar bonds in four months and paid only 5bp to 7bp more. If lenders were simply penalising debt growth, Amazon should show it. My position is that Amazon shows lenders penalise doubt about cash flow, not size as such, and Amazon's cash flow is the least doubted in the group. That reading fits the data. It is still a reading.

What the next hyperscaler bond deal will tell you

The question for the next six months is whether the long-end premium spreads into shorter maturities. If the 10-year spread on a AA name starts moving like the 40-year did, lenders have stopped pricing duration risk and started pricing credit risk.

The funding mix is already shifting. On 26 October 2026, Meta filed a press release for about $35 billion of underwritten equity, combining $22.5 billion of common stock under forward sale agreements with a mandatory convertible preferred. It also flagged a planned at-the-market programme of about $60 billion starting in the first quarter of 2027.

A company that paid +147bp for 40-year debt in April and then turns to equity in October is telling you how it reads its own cost of debt. I would not call that distress. Meta remains rated AA-. I would call it a signal that the cheapest marginal dollar for AI capex is no longer a long bond.

Three readings to watch, in order of importance:

• The 10-year spread on the next AA-rated hyperscaler deal. Above +100bp would mean the repricing has moved beyond duration.
• Whether Oracle, now rated BBB-, keeps its pledge not to issue more bonds in 2026.
• Peak order book divided by deal size. A jumbo deal below 2x cover would be the first sign that supply has outrun demand.

Frequently asked questions

Where to start

Pull one term sheet yourself. Search SEC EDGAR for the company, filter to Form FWP, and open the pricing term sheet for its last bond deal. Every tranche lists the benchmark Treasury yield, the spread and the re-offer yield. It takes ten minutes and it replaces every secondhand headline about AI debt costs.

Then add one row to this table each time a hyperscaler prices a new deal. Record the 10-year and the 40-year spread and the order book. If the 10-year column starts moving the way the 40-year column has, the AI debt story has changed, and you will see it before most commentary does.

Related reading