The AI selloff of October and early November 2026 did something more interesting than break the trade: it made the funding of the trade cheaper. Convertible issuance set records through the year even as CoreWeave's five-year credit default swaps blew past 1,000 basis points, and the new issues carried coupons as low as zero and 0.5%. What looks like a win for convertible buyers is a transfer — the coupon line gets cheaper while the share-count line gets more expensive, and the AI capex being financed comes attached with dilution.
Gap between the two prices
~1,300 bp
Same issuer, seven weeks apart
Nscale pre-IPO convertible
$3.36B
Announced September 25, 2026; $1 billion NVIDIA tranche due mid-November
Zero-coupon global convert deals in Q2 2026
44 of 96
$84.6 billion issued in the quarter, up 67% year over year
CoreWeave contracted backlog
$132.5B
As of September 30, 2026, against $13.0–13.3 billion of FY2026 revenue guidance
The market backdrop
The Sellout Repriced AI Credit, Not the Convertible Market
Two things happened in the first week of November 2026, and only one of them reached the AI capital structure. Credit repriced hard: the Nasdaq 100 fell 5% on the week, the VIX closed above 28 on Thursday, and CoreWeave's five-year credit default swaps pushed past 1,000 basis points to record highs. The convertible market did not stop printing. US investment-grade corporate issuance reached $1.88 trillion year to date, already a record for a full year, and convertible supply kept setting records of its own.
| Indicator | Level | Date | What it means for funding |
|---|---|---|---|
| Nasdaq 100 | Down 5% on the week | Week ended Nov 6, 2026 | Equity de-rated; the funding window did not close |
| VIX | Above 28 Thursday, about 25 Friday | Nov 5–6, 2026 | Equity volatility is exactly what makes convertibles cheap |
| CoreWeave five-year CDS | Above 1,289 bp, easing to about 1,155 bp | Nov 5–6, 2026 | AI credit is where the stress actually landed |
| US high-yield premium | Highest since the April 2025 tariff shock | Nov 6, 2026 | Sub-CCC debt near its widest since 2020 |
| US 10-year Treasury | About 50 bp below a peak above 5% | Nov 6, 2026 | Risk-off, not risk-off and inflationary |
| US IG corporate issuance | $1.88 trillion year to date | Nov 6, 2026 | A full-year record already set by AI borrowers |
| Cash funds | $140 billion inflow in one week | Week to Nov 4, 2026 | Money rotated into cash, not out of credit |
This matters for the funding question because a convert's coupon is effectively paid out of the value of the embedded equity option. When the equity falls, the same dollar of principal buys more shares, so investors accept a lower cash coupon and the issuer gets cheaper money — paid for in shares. The equity holder, not the lender, picks up the bill for a build-out that is now being financed almost entirely with paper.
The pricing evidence
The Coupon Ladder Shows Who Can Actually Get the Cheap Money
The 2026 convertible record is not one trade; it is two. Issuers whose equity is deeply volatile — the neoclouds and the AI infrastructure names — are clearing zero-coupon and 0.5% paper at conversion premiums above 40%. Issuers with investment-grade ratings and enormous cash generation are paying 6.25% and 6.50% in cash dividends on mandatory convertible preferred. Issuers get cheap money only by handing over the equity option, which is why the two ends of the ladder look nothing alike.
The 2026 AI-linked coupon ladder
Cash coupon or dividend rate on AI-linked convertible paper priced in 2026. The spread from 0.00% to 6.50% is a credit-and-volatility story, not a rate story. Nscale's pre-IPO convertible is excluded because its terms were not disclosed.
Unit: % cash coupon or dividend rate
| Issuer | Instrument | Priced | Size | Coupon / dividend | Conversion price | Premium to reference |
|---|---|---|---|---|---|---|
| CoreWeave | Convertible senior notes due 2033 | Sep 17, 2026 | $4.2B | 2.875% | $97.85 | 22% over the $79.88 close |
| CoreWeave | Convertible senior notes due 2031 and 2032 | Apr 14, 2026 | $4.0B | 1.75% | Not disclosed | Not disclosed |
| Nebius | Convertible senior notes due 2030 | Aug 19, 2026 | $3.0B of $5.0B | 0.50% | $313.46 | 40% over the $223.90 close |
| Nebius | Convertible senior notes due 2034 | Aug 19, 2026 | Balance of $5.0B | Higher than 0.50% | Not disclosed | Not disclosed |
| Alphabet | Series A and B mandatory convertible preferred | Jun 2, 2026 | About $15B | 6.25% | 2.2520–2.8420 shares per $1,000 | Capped calls struck near 50% |
| Oracle | Mandatory convertible preferred | Feb 2, 2026 | $5.0B | 6.50% | About $160.06 | 25% over the Feb 2 close |
| Penguin Solutions | Convertible senior notes | Jul 14, 2026 | $650M | 0.00% | About $116.70 | Not disclosed |
| Nscale (private) | Pre-IPO convertible loan notes | Sep 25, 2026 | $3.36B | Not disclosed | Not disclosed | Not disclosed |
MacKay Shields, which runs one of the largest convertible mandates in the world, reports that nearly a quarter of the convertible benchmark's weight is now tied to AI development, and that several 2026 issues came with coupons under 50 basis points and conversion premiums above 40%. Global convertible issuance hit $84.6 billion across 96 deals in the second quarter alone, up 67% year over year, with 44 of those 96 priced at zero coupon. The mechanics are simple and unfriendly to equity: the investor's coupon falls while the issuer sells a bigger claim on its own upside.
The clearest single piece of evidence
CoreWeave Prices the Spread: 2.875% and 16% for One Credit
There is a cleaner way to see this than a coupon ladder. On September 17, 2026, CoreWeave priced $4.2 billion of 2.875% convertible senior notes due 2033, upsized from $3.7 billion, with an initial conversion price of about $97.85 and capped calls costing $566.2 million. On November 6, 2026, the market demanded more than 16% to hold the same issuer's 9.625% senior notes due 2032. Seven weeks apart, one credit, roughly 1,300 basis points of difference — and the market prices that gap at the equity option, not at the interest rate.
| Instrument | Coupon | Status | Pricing in November 2026 |
|---|---|---|---|
| Convertible senior notes due 2033 | 2.875% | Issued Sep 22, 2026 | $4.2B at a $97.85 conversion price |
| Convertible senior notes due 2031 and 2032 | 1.75% | Issued April 2026 | $4.0B outstanding |
| Senior notes due 2032 | 9.625% | Outstanding | Yield above 16% on Nov 6, 2026 |
| Senior notes due 2030 | 9.250% | Outstanding | Traded on the same stressed curve |
| Senior notes due 2031 and 2032 | 9.000% and 9.750% | Outstanding | Traded on the same stressed curve |
| Five-year credit default swaps | Not applicable | Nov 5, 2026 | Above 1,289 bp, easing to about 1,155 bp |
CoreWeave: interest expense has eaten about a quarter of revenue for five straight quarters
Interest expense as a percentage of quarterly revenue, third quarter 2025 through third quarter 2026.
Unit: % of quarterly revenue
Q3 2025
22.8%
Q4 2025
24.7%
Q1 2026
25.8%
Q2 2026
24.9%
Q3 2026
23.1%
The 2.875% is the cheaper price only if the equity rallies far enough for the notes to convert. CoreWeave's market capitalization of $33.5 billion against 559 million diluted shares values the equity at roughly $60 a share, well below the $97.85 strike set in September. The option is close to worthless, so the notes will not be settled in stock — the 2.875% becomes straight-debt pricing dressed up as a coupon, and the discount the market is already applying is the tell.
The cost side of the trade
What the Cheap Coupon Actually Costs: A Dilution Ledger
Cheap coupons are real money only if the equity recovers. If it does not, the issuer has swapped a funding problem for a share-count problem. Here is what has already been sold, using each issuer's own disclosure.
| Issuer | Principal | Conversion price | Shares on conversion | Dilution | Hedge cost |
|---|---|---|---|---|---|
| CoreWeave | $4.2B, Sep 2026 | $97.85 | 42.9M at issue, 52.6M at the cap rate | 7.7% of 559M diluted shares, rising to 9.4% | $566.2M capped call, cap price $199.70 |
| Nebius | $3.0B of $5.0B, Aug 2026 | $313.46 | 9.6M on the 2030 notes | 4.3% of 220.4M shares from one tranche | Not disclosed |
| Oracle | $5.0B, Feb 2026 | About $160.06 | 31.2M | 1.1% of roughly 3.0B diluted shares | Mandatory conversion January 2029 |
| Alphabet | About $15B, Jun 2026 | 2.2520–2.8160 shares per $1,000 | Capped at the strike | Capped by capped calls struck near 50% above the offering price | Capped call premium disclosed |
- CoreWeave paid $566.2 million for capped calls — 13.5% of the $4.2 billion gross — to cap dilution at $199.70 a share. The hedge premium buys no capacity and shrinks the equity cushion.
- Nebius closed roughly $5.75 billion of gross convertible proceeds against $1.36 billion of trailing revenue, with capital spending running at 8.2 times revenue. Debt sized at more than four times revenue now rests on a strike roughly double the share price.
- Oracle's $5.0 billion of mandatory convertible preferred pays 6.50% in cash until it converts in January 2029, at a strike of about $160.06 set in February 2026. Even the investment-grade names now pay a real coupon to defer the equity decision.
- Alphabet can afford to cap its own dilution at roughly 50% above the June offering price because its balance sheet is an order of magnitude larger. Alphabet holds ten times the equity cushion of Oracle, at $677.5 billion against $66.8 billion.
Dilution is not a rounding error at the bottom of the AI capital structure; it is the price of the money. A builder that cannot issue straight debt at any sensible coupon issues equity-linked paper instead, and the cost lands in the share count rather than the income statement. Issuers choose between paying interest now and paying interest later at a worse price — and in 2026, a growing number of them are choosing the second.
The template for everything that has not listed yet
Nscale Writes the Next Template: The Pre-IPO Convertible
Nscale is private, so it cannot issue equity-linked paper with a traded strike. On September 25, 2026 it announced a $3.36 billion pre-IPO convertible financing led by Third Point, with $2.36 billion funded at closing and a further $1 billion commitment from NVIDIA expected to fund in mid-November 2026. The notes convert into ordinary shares automatically on completion of the IPO, with NVIDIA receiving non-voting shares. Apollo, Citadel, Hudson Bay Capital, the Abu Dhabi Investment Council and Wellington Management are among the participants. Nscale solved the funding problem with paper that only prices at the IPO.
- The structure removes the rating problem and the strike problem at once: with no public price to anchor a conversion premium against, the conversion happens at whatever the IPO produces. An unpriceable credit becomes a priced equity claim.
- NVIDIA's $1 billion tranche is due to fund in mid-November 2026, so the first live test of the template arrives with the listing itself. A soft IPO print resets the reference price for every pre-IPO AI convertible that follows.
- Coupon, discount and conversion terms were not disclosed. The template is verifiable in structure but not yet in price, which is the single most important gap to watch.
- Nscale filed for its initial public offering on September 18, 2026, targeting a valuation of about $30 billion. The conversion price of the notes becomes a function of a listing that has not happened yet.
That template extends well beyond one company. With AI equity down making straight debt unviable for mid-tier borrowers, and pre-IPO borrowers unable to sell stock at all, the pre-IPO convertible becomes the default instrument for the next wave of AI infrastructure. The dilution stays invisible until the IPO, when noteholders and venture holders claim the same equity — which is precisely why the market has not yet priced what the next trillion of AI capex is being sold with.
Upstream
The Money Lands With the Suppliers, the Dilution Lands With the Builder
The supply chain reveals where the cheap money actually goes. Capital-intensive suppliers of AI hardware and power self-fund from operating cash flow: NVIDIA runs capital spending at 2.4% of trailing revenue with net debt at 0.07x EBITDA, SK Hynix at 13.4% with a net cash balance sheet, Vertiv at 4.0% also with net cash. The businesses that own the assets and rent them out run capital spending at multiples of revenue and raise paper instead.
| Issuer | Capex / revenue (TTM) | Net debt / EBITDA (TTM) | Funding route | Trailing revenue |
|---|---|---|---|---|
| CoreWeave | 279% | 17.6x | Converts at 1.75% and 2.875% | $9.9B |
| Nebius | 822% | 2.1x | Convert at 0.50% | $1.4B |
| Oracle | 105% | 3.2x | $25B straight IG plus a 6.50% mandatory convert | $71.8B |
| Alphabet | 34% | 0.27x | $84.75B equity plus a 6.25% mandatory convert | $460.0B |
| Meta | 43% | 0.95x | Roughly $95B equity sale | $227.3B |
| NVIDIA | 2% | 0.07x | Operating cash flow | $301.8B |
| SK Hynix | 13% | Net cash | Operating cash flow | ₩269.3T |
| Vertiv | 4% | Net cash | Operating cash flow | $13.1B |
- NVIDIA committed $1 billion to Nscale's pre-IPO convertible, with funding expected in mid-November 2026. The chip supplier now co-funds its customer's data-centre wall through equity-linked paper.
- SK Hynix posted ₩104.6 trillion of revenue in the third quarter of 2026, more than triple the ₩32.8 trillion of the fourth quarter of 2025, while holding capital spending to 13.4% of revenue. Memory suppliers expand capacity on operating cash flow, so funding stress never reaches them first.
- Vertiv grew third-quarter revenue 61% year over year to $4.3 billion and runs a 20.3% trailing operating margin with net cash. Power and cooling vendors collect cash from a build-out financed with paper.
- CoreWeave alone plans $37–39 billion of capital spending in 2026. That single order book is why cheap convertible funding matters to the whole upstream.
Downstream
Only the Megacaps Can Still Buy Compute With Plain Investment-Grade Debt
At the other end of the chain, the buyers of AI compute are the only participants who can still fund it with ordinary investment-grade debt. Hyperscaler investment-grade issuance reached $218.1 billion by early October 2026, against $107.3 billion for the whole of 2025, and total index-eligible hyperscaler debt stood at $500 billion at the end of August.
| Issuer | Index-eligible IG debt (Aug 31, 2026) | Five-year CDS (Oct 21, 2026) | Implied five-year default probability |
|---|---|---|---|
| Amazon | $144.8B | 80 bp | 7% |
| Alphabet | $120.4B | Not disclosed | Not disclosed |
| Oracle | $117.2B | 331 bp | 25% |
| Meta | $81.2B | Not disclosed | Not disclosed |
| Microsoft | $36.5B | 56 bp | 5% |
- Oracle's option-adjusted spread of 210 basis points on August 31 sat against a 119 basis point hyperscaler average, and its 331 basis point five-year default swap priced a 25% chance of default. Oracle is priced as a credit, not as a technology issuer.
- S&P cut Oracle to BBB- on July 9, 2026, citing elevated AI-related capital spending. The lowest investment-grade rating now sits on the largest AI capex commitment in the market.
- Oracle's balance sheet shows $155.9 billion of total debt and $119.6 billion of net debt against $66.8 billion of shareholder equity; Alphabet carries $141.9 billion of debt against $677.5 billion of equity. Oracle carries more debt than Alphabet on a fraction of the equity.
- Hyperscalers now make up 4.7% of the FTSE US BIG Corporate index by market value weight, up from 2.7% two years earlier, and 1.2% of the euro investment-grade index, up from 0.08%. AI is now a structural block of the benchmark, not a satellite.
That single comparison is the whole funding curve. Straight debt for the megacaps still prices at 56 and 80 basis points; straight debt for a neocloud prices at 331. For the neocloud, the equity-linked alternative prices at a 2.875% coupon the market already discounts, because the strike is out of reach. Straight debt for the megacaps still clears at 56 and 80 basis points while a neocloud's prices at 331 — and that gap, not the headline coupon, is what determines who can build.
Horizons
What Moves in Days to Quarters, and What Breaks in 2029 to 2033
The near-term evidence will come from pricing, not from headlines about AI sentiment.
- Nscale's IPO and the mid-November 2026 NVIDIA tranche are the first live test of the pre-IPO convertible template. A weak listing print resets the reference price for every pre-IPO AI convertible that follows.
- CoreWeave guided fourth-quarter 2026 interest expense to $1.05–1.15 billion against $3.66 billion of third-quarter revenue. The cash cost of the build-out keeps compounding even as the coupon headline stays at 2.875%.
- Oracle raised roughly $10 billion of stock and $10 billion of debt on November 2, 2026, months after its downgrade to BBB-. Watch whether the next megacap deal prices as equity-linked paper instead of straight bonds.
- Goldman Sachs Research flagged new-issue concessions of up to 20 basis points on large hyperscaler deals in August 2026, against two to three basis points earlier in the year. Concessions are the cleanest weekly read on whether convertible demand has genuinely broken.
- Goldman Sachs Research expects roughly 35% of hyperscaler capital spending to be debt-financed in 2027, up from about 33% in 2026 and 27% in 2025. The debt-funded share of AI capex keeps rising, which enlarges the eventual refinancing wall.
- The 2026 vintage matures where the equity has to be worth more than it is today: CoreWeave's convert matures April 2033, Nebius's notes mature in 2030 and 2034, and Oracle's mandatory preferred converts in January 2029. If the equity does not recover, January 2029 is the first date the bill arrives.
- J.P. Morgan estimates roughly $279 billion of AI-related issuance in 2026 with another $220–300 billion likely in 2027, on top of about $697 billion of hyperscaler capital spending. The dilution bill scales with the build-out it is financing.
The case against this view
What Would Break This Thesis
- If CoreWeave's shares clear the $97.85 strike, the converts convert and the 2.875% coupon was genuine cheap funding instead of deferred dilution.
- If convertible demand breaks outright, AI capex stops rather than reprices: with straight debt at 16% for CoreWeave and Oracle's spread 91 basis points wider than the hyperscaler average, the marginal builder loses the option to fund at all.
- If the megacaps keep funding with plain equity — Meta's roughly $95 billion sale and Alphabet's $84.75 billion raise — the convertible window stays open only for names that cannot sell stock and the whole shift stays a neocloud story.
- Nscale's coupon, discount and conversion price were not disclosed, so the economics of the pre-IPO template remain unverified even though its structure is documented.
- Goldman Sachs Research tracked nearly $500 billion of AI-related debt issuance in 2026 while the AI leadership basket's spread went from 74 basis points to nearly double that; if spreads keep widening, cheap convertibles become the only funded route and the dilution accelerates.
Who the AI funding curve actually touches
- Straight debt yields above 16% while the September convert still pays 2.875%, a ~1,300 bp two-price curve the equity cannot close. The funding gap widens every quarter the stock stays below the $97.85 strike.
- Interest expense has eaten 22.8% to 25.8% of revenue in each of the last five quarters, with Q4 2026 guided to $1.05–1.15 billion. The cash cost of the build-out keeps compounding while the coupon headline stays at 2.875%.
- $4.2 billion of converts at $97.85 equals 42.9 million shares, 7.7% of the 559 million diluted base, rising to 9.4% at the cap rate. Dilution is the price of the cheap funding.
- $566.2 million of capped calls, 13.5% of gross proceeds, buys no capacity and shrinks the equity cushion. The hedge premium is pure financing cost.
- Five-year CDS of 331 basis points implies a 25% default probability over five years, against 7% for Amazon and 5% for Microsoft. Oracle is priced as a credit, not as a technology issuer.
- $155.9 billion of debt against $66.8 billion of equity is more total debt than Alphabet's $141.9 billion on $677.5 billion of equity. The same AI build-out on a fraction of the balance sheet.
- The 6.50% mandatory convertible pays cash dividends until January 2029 conversion on top of $5.1 billion of trailing annual interest expense. The equity deferral now carries a coupon.
- Watch the November 2, 2026 financing and whether the next tranche prices as equity-linked paper. Another downgrade closes the straight-debt window entirely.
- $84.75 billion of equity plus a 6.25% mandatory convertible preferred funds a capex program now running at 33.9% of revenue. Alphabet buys the AI build-out without diluting anyone meaningfully.
- $265.1 billion of cash and short-term investments against $85.6 billion of net debt. Alphabet is the only hyperscaler that can fund a downturn from cash.
- Nearly a quarter of the convertible benchmark's weight is now AI-linked and Alphabet alone printed about $15 billion of it in June 2026. Alphabet sets the price for AI-linked paper.
- Over one to three years, capped calls struck roughly 50% above the June offering price cap its dilution. The quality of the balance sheet is the hedge.
- Roughly $5.75 billion of gross convert proceeds against $1.36 billion of trailing revenue, with capital spending at 8.2 times revenue. The balance sheet is funded on an option that is not in the money.
- The $313.46 conversion price is roughly double the current share price, so the 0.50% coupon on the 2030 notes will not be paid in shares. The 0.50% headline is a coupon, not a yield.
- The $3.0 billion 2030 tranche alone converts into 9.6 million shares, 4.3% of the 220.4 million outstanding. One tranche already dilutes the base by more than 4%.
- With a 2030 maturity and no public equity currency to refinance with, an equity drawdown becomes a refinancing problem.
- $1 billion committed to Nscale's pre-IPO convertible, funding expected mid-November 2026, puts NVIDIA inside its customers' funding. The supplier now underwrites its customer's wall.
- Capital spending of 2.4% of revenue and net debt of 0.07x EBITDA mean NVIDIA funds the build-out from cash flow while customers fund it from paper. The funding risk sits downstream of NVIDIA, not on its balance sheet.
- CoreWeave alone plans $37–39 billion of 2026 capital spending. The order book is only as good as customers' ability to keep funding it.
- Over one to three years, a genuine funding freeze would hit GPU orders with a lag of quarters. Convertible issuance is the leading indicator to watch.
- A roughly $95 billion stock sale, the largest in its history, opened with a $35 billion tranche on October 26, 2026. Meta funds AI capex with ownership, not with coupons.
- Capital spending of 43.2% of trailing revenue, up from 34.7% in 2025. The equity raised is being spent, not hoarded.
- Net debt of 0.95x EBITDA leaves room to absorb a $5 billion-scale coupon if Meta ever chooses convertibles. Meta has the balance sheet to use the cheap instrument if it wants to.
- Over one to three years, continued plain equity issuance is the strongest signal the build-out can be financed without dilution. Watch the mix, not the headline capex number.
- Third-quarter revenue rose 61% year over year to $4.3 billion, with a 20.3% trailing operating margin and net cash. Vertiv is paid in cash by a build-out financed with paper.
- Capital spending of 4.0% of revenue means no external funding is needed to chase the order book. There is no dilution risk in the power-and-cooling layer.
- CoreWeave's $37–39 billion 2026 plan and Oracle's 105% capex intensity are the demand that pays Vertiv. Supplier revenue is a claim on someone else's funding.
- Over one to three years, if convertible issuance stays at record pace, power and cooling demand stays funded. Funding availability, not order intake, is the binding constraint.
- Third-quarter 2026 revenue of ₩104.6 trillion was more than triple the ₩32.8 trillion of the fourth quarter of 2025. HBM demand is paid for in cash, not in share count.
- Trailing revenue of ₩269.3 trillion with capital spending at 13.4% of revenue and a net cash balance sheet. The memory layer self-funds the AI capacity it supplies.
- Nearly $500 billion of AI-related debt issuance tracked in 2026 barely touches memory suppliers. The funding stress stops at the layer that owns the asset.
- Over one to three years, HBM capacity additions are funded internally while customers dilute. The asymmetry favours the supplier's margin, not the builder's
