The AI industry has found a way to borrow money for nothing. AI promoters are raising billions in zero-interest convertible debt backed by nothing but stock volatility — $54 billion in U.S.-listed convertible issuance in the first half of 2026, up 43 percent from the same period last year and the heaviest year-to-date volume since the early-pandemic panic, according to Dealogic data going back to 1995. On its face, the story is cheap capital meeting hot demand. In the register that matters, it is an entire infrastructure build-out funding itself with money that costs nothing now and will cost plenty later, provided there is a later. The figure is the fixed-income flank of a broader capital torrent that has been funneling record sums into AI across equity and debt markets.
It is worth being precise about the mechanism, because the phrase “borrowing for nothing” does real work only when the reader can see the engineering. A convertible bond is a debt instrument with an embedded equity call option. The issuer borrows money and promises to repay it, but the lender also receives the right to convert that debt into shares at a specified price. When the issuer’s stock is volatile and climbing — as AI stocks have been doing with some consistency this year — the conversion option becomes valuable enough that the issuer can pay zero or near-zero interest. Akamai Technologies, the cybersecurity and cloud-computing firm, issued $3.5 billion in zero-coupon convertible notes in May, split between 2030 and 2032 maturities, with conversion premiums set at 35 and 42.5 percent above the share price on the day of pricing. Zero coupon means the bond pays no interest at all over its entire life. CoreWeave, the AI cloud provider, sold $4 billion in 1.75 percent convertible notes shortly before, well below what conventional debt of comparable risk would command.
“Convertibles are growth capital for growth issuers,” said Joe Wysocki, a senior portfolio manager at Calamos Investments, “and I don’t think you can think of a better growth opportunity than AI.” CoreWeave’s vice president of corporate development, Nick Robbins, was more direct: “High-growth AI businesses are perfect for the convertible market because the volatility that comes with all that growth makes pricing very attractive to issuers.” This is true in the way that a card game is attractive to the person dealing. Volatility is what makes the conversion option valuable; the conversion option is what the issuer sells in exchange for not paying interest. The company gets free money. The investor gets a bond that pays nothing and an option that pays only if the stock keeps rising. The risk asymmetry is complete.
The market conditions supporting this arrangement are, by the account of the people profiting from them, unusually favorable. The ICE BofA US Convertible index has gained more than 20 percent year-to-date, outpacing the S&P 500 and the Nasdaq. Credit spreads — the premium investors demand for holding riskier corporate debt over ultrasafe Treasurys — are near decade lows. Stock prices are elevated. Share volatility is high. Michael Youngworth, head of global convertibles strategy at BofA Securities, confirmed the trifecta: “What benefits convertible bond issuers: high stock prices, tight credit spreads, and well-supported stock volatility. We have all three of those things right now.” A reasonable observer might note that the same conditions making it cheap to borrow against stock-price promises are the conditions that make it dangerous — because the cheapness depends on those promises holding.
The structural precedent is one the convertible market has followed before. In 2024, crypto firms — most conspicuously Strategy, the bitcoin-accumulation vehicle founded by Michael Saylor — flooded the market with converts. When crypto prices sank, issuance collapsed. The converts did not, on the whole, convert into wealth. They converted into losses, quietly absorbed by yield-hungry investors who had mistaken volatility for a business model. The AI convertible boom differs from the crypto boom mainly in narrative density: the infrastructure is physical — data centers, fiber, power-purchase agreements for nuclear baseload, GPU clusters — and the companies raising the money sometimes have actual revenue, though rarely actual profit. CoreWeave, for all its growth, is a cloud provider whose customers are themselves AI startups burning cash. Akamai is a mature firm that happens to be riding the same updraft. The money is real. The physical build-out is real. The question is whether the revenue that is supposed to materialize on the far side will be real enough to repay the capital that built it.
A zero-coupon convertible is a pure bet on stock-price appreciation. The bondholder receives no interest payments and only profits if the shares clear the conversion premium, which in Akamai’s case means a 42.5 percent rise above a 26-year high. The bet is not that the company will generate free cash flow, pay a dividend, or even survive as a going concern. The bet is that the stock price will keep rising. The stock price, in turn, is rising because investors are betting the AI build-out will produce returns that, as a base-rate analysis of AI revenue projections indicates, are historically without precedent. One wager is stacked on another, and the zero coupon sits at the bottom, extracting the cost of capital from the future while returning nothing to the present.
This is the mechanism for converting hype into cash before the hype converts into revenue — if it ever does.
John Kenneth Galbraith, writing about the 1929 crash in The Great Crash, gave the pattern a name. He called it “the bezzle” — the amount of undiscovered embezzlement in the system at any given moment, which rises during a boom because everyone feels richer than they are. The embezzler has the money; the victim has not yet discovered the loss. The interval between the two is the bezzle, and it is always larger than anyone expects, because the feeling of prosperity it generates is real even though the underlying wealth is not. The AI convertible boom is not embezzlement in the criminal sense — no one is stealing — but the financing structure produces a temporal illusion with the same shape. The issuing company has the cash; the investor holds a bond paying zero percent and the belief that conversion will be profitable; both parties feel they are getting a good deal.
Cory Doctorow has spent years applying Galbraith’s bezzle to the technology sector — Uber, crypto, platform monopolies — using it to mark the interval when a scheme is already broken but its participants have not yet noticed. The image that sticks is Wile E. Coyote running on air, legs still pumping, the fall not yet begun. The AI convertible market is running in that interval. The institutions buying these bonds — hedge funds, convertible-arbitrage desks, dedicated funds like Calamos — hedge their equity exposure through short positions in the underlying stock; their risk is managed at the portfolio level. But the bonds themselves flow into funds that retail investors hold: pension funds, mutual funds, target-date retirement vehicles. The bezzle is being stretched across an entire capital market, and the circle is closed. Investors are handing billions of dollars to companies that pay them next to nothing — or, in the case of the zero-coupon issues, literally nothing — on the promise that stock prices will one day rise enough to turn the bonds into equity. The stock prices are rising because investors keep handing over the money. The only thing that can break the circle is the discovery — the moment the market notices that the revenue is not arriving, that the AI build-out has overshot, that the conversion premium was the only premium that ever existed.
There is a public-relations language for all of this, and the participants speak it fluently. Wysocki and Robbins offer the standard grammar: convertibles are growth capital for growth issuers; high-growth AI businesses are perfect for the convertible market because their volatility makes pricing attractive. The statements are not false. They are incomplete in the way that all statements designed to keep the bezzle going are incomplete: they omit the part where the volatility stops.
Manoj Shivdasani of GSR Research supplied, with a candor that doubles as a description of the problem, the structural frame: “The risk is we are getting a little overexposed to AI. But that’s part of the convert market. This market finances high-growth names.” It does finance high-growth names. It financed them in 2000, before the dot-com bust. It financed them in 2024, before crypto pulled back. The market’s institutional memory is shorter than its marketing cycle.
Ed McGowan, Akamai’s chief financial officer, described the convertible market as “very good to us” and “the most efficient, cheapest alternative for capital.” He is right. It is the cheapest alternative, and it is cheap because the investor is bearing the risk that the stock price might not reach the conversion threshold — in which case the bond matures paying nothing over its entire life — or the company cannot repay, in which case the investor is a creditor in a bankruptcy proceeding holding a zero-coupon note against an asset base of depreciating GPUs.
There is a Polish saying my grandfather used, one that translates badly but means something close to the work doesn’t care how you feel about it. The work these bonds are financing is real — the data centers, the fiber, the nuclear baseload, the GPU clusters that make large-language-model inference possible. The work is happening. The question, as it always is in the interval Galbraith described, is who is paying for it, and in what form the bill arrives.
Right now the bill arrives as a zero-coupon convertible note. The interest is nothing. The principal depends on the stock price. The stock price depends on the story holding. The convertible market is a machine for turning volatility into cash. Nobody has yet built a machine that runs on revenue.