Ninety-Five Dollars: How Intel Raised $20 Billion at the Exact Top of the AI Rally
Intel priced a $20 billion upsized stock offering at $95 a share on August 11, 2026 — and the stock closed below $93 nine days later. The deal terms, the dilution math, and what $19.7 billion buys in the foundry arms race.
By MyAudioBooks.ai ·
Listen free: Ninety-Five Dollars: How Intel Raised $20 Billion at the Exact Top of the AI Rally
On the evening of August eleventh, twenty twenty-six, a price crossed the wire in one of the largest stock offerings in the history of the American semiconductor industry: ninety-five dollars a share, two hundred ten point five million shares of Intel, just under twenty billion dollars raised in a single night. Three months earlier, the same stock had traded in the low forties. The offering had been announced at fifteen billion dollars and upsized by a third before dinner. The book, Wall Street's word for the pile of institutional orders behind a deal like this, was full before the banks finished building it. Nine days later, the stock closed below ninety-three dollars, and every one of those institutions was under water. This is the story of the fastest twenty billion dollars ever raised in chips — and the nine days it took the market to change its mind about it.
At My Audio Books dot A I, you can create your own audiobooks from prompts, turn your documents into audio, all with one subscription, and store your items in your own personal library.
The mechanics matter, so start with them. Intel sold two hundred ten point five million new shares at ninety-five dollars each, for gross proceeds of just over twenty billion and net proceeds of roughly nineteen point seven billion after fees. The banks that ran the deal held a thirty-day option to buy another thirty-one and a half million shares at the same price, which would push the total toward twenty-three billion. The deal was priced on a Monday and settled that Wednesday. General corporate purposes, the prospectus language says, including capital expenditures. In semiconductor accounting, capital expenditures is a word that means factories.
Section One. What Nineteen Point Seven Billion Dollars Buys.
To understand why Intel sold stock at the top of a rally, you have to understand what a modern chip factory costs. A leading-edge fabrication plant — the buildings, the cleanrooms, the lithography machines, the utility plants that consume as much power as a small city — runs twenty to thirty billion dollars before it produces its first sellable wafer. The machines inside a single fab can cost one hundred fifty million dollars apiece. A fab takes three to five years to build and ramp. And a leading-edge node is only leading-edge for about two years before the next one arrives, which means the spending never stops: the moment one factory is finished, the capital plan for its replacement begins. Nineteen point seven billion dollars, the net proceeds of this offering, is roughly one factory. That is the unit of account in the foundry business, and it explains everything else.
The foundry industry is a scale game with almost no players left. Building chips for other companies — the business Taiwan Semiconductor Manufacturing Company turned into the most important industry on Earth — requires riding successive generations of technology, each one more expensive than the last, with no guarantee the customers come. Most companies quit. GlobalFoundries stopped at the fourteen-nanometer generation in twenty eighteen rather than spend the tens of billions the next node required. Intel is one of three companies on the planet still trying to reach the leading edge, and the only one headquartered in the United States. Washington has spent the current era of industrial policy trying to make sure that number does not fall to two — and Intel's foundry ambitions are the centerpiece of that project. The offering, in that light, was not just a balance-sheet move. It was the recapitalization of a national bet.
Consider the economics that make this business so strange. A wafer fabricated at the leading edge sells for thousands of dollars more than the same wafer one generation behind, which is why the last company standing at each node earns margins that look like a software company's while spending capital like a highway department. But the pricing power only exists while you are actually at the leading edge — miss one generation, as Intel did through the twenty-tens, and your factories depreciate faster than they earn. Taiwan Semiconductor spends more on capital expenditures in a single year — upwards of forty billion dollars in the current cycle — than the gross domestic product of several European Union members. That is the scoreboard Intel is playing against, and it is why a company that already spends tens of billions a year on factories needed twenty billion more in one night. In the foundry race, pausing is the same thing as losing; the only question is whether you fund the marathon with equity, with debt, or by dropping out.
The economics of being second are worth pausing on, because they explain the desperation underneath the rally. A foundry that is one node behind can still fill its factories — the world needs far more chips than the leading edge can produce — but it earns commodity prices for capacity that cost leading-edge money to build. A foundry that is one node ahead, by contrast, books margins that would embarrass a luxury brand, because the customers with the most valuable designs will pay almost anything for the best transistors. That cliff-edge difference between first and second place is why the market rewards any hint of a technology lead with a doubling stock, and why it punishes any hint of a slip with a halving. The ninety-five dollar print was the market pricing in the possibility that Intel had clawed back to the edge. The nine-day fade was the market remembering how many times that possibility has been priced in before.
Why stock instead of bonds? Intel already carries tens of billions of dollars of debt on its balance sheet, a legacy of the borrow-and-build years. A company with that load, facing a ratings downgrade if it borrows tens of billions more, sells equity instead — accepting dilution as the price of keeping the fortress standing. Two hundred ten point five million new shares against roughly four and a half billion shares outstanding is dilution of about four and a half percent: every existing shareholder's slice of the company shrank by roughly a twentieth overnight, in exchange for a factory-sized pile of cash that owes no interest. When the stock is rallying, that trade is cheap. Which brings us to the timing.
There is one more piece of context that makes the equity choice unusual. In the old semiconductor playbook, the customers paid for the factories. The companies that buy chips would prepay, co-invest, or sign take-or-pay contracts that underwrote the capital plan — their demand was the collateral. Intel's foundry customers have begun doing versions of exactly that, and every such contract signed makes the equity less necessary. The fact that Intel still chose to sell four and a half percent of itself into the open market tells you the contract pipeline, however promising, does not yet cover a twenty-billion-dollar factory program. The offering is what you do between the promises and the payments.
Section Two. Selling Into the Rally.
The offering was announced as a fifteen billion dollar deal, and demand let the banks upsize it by a third in a day. That demand existed because Intel's stock had spent the summer doing something it had not done in years: going up, violently. From the low-to-mid forties in April, the shares roughly doubled by August, carried by a market that had decided the artificial-intelligence boom needed American foundry capacity and Intel was the only credible supplier of it. Every data-center order, every sovereign-A I headline, every capacity-constraint warning from the industry added fuel. By the pricing night, the stock had become the market's purest expression of the foundry renaissance thesis.
Selling stock into that kind of rally is what finance calls selling into strength, and it is unambiguously the rational thing for a board to do. The company's own officers would say, and did say, that they simply funded the plan at attractive prices. But timing analysis is unforgiving, and the tape has since rendered its verdict: ninety-five dollars, the price at which Intel sold two hundred ten million shares of new equity, is looking very much like the local top of the artificial-intelligence foundry mania. The stock that the market bid to ninety-five in the fever of August could not hold ninety-three by the twenty-first. The offering was struck at the exact peak of enthusiasm — the moment when the last marginal buyer was matched with two hundred ten million shares of new supply.
There is an older precedent worth knowing. In the summer of twenty twenty, an electric-truck maker with no revenue sold shares into a retail mania at prices its officers could not have dreamed of a month earlier, and was widely mocked for it — until the cash from that sale, raised at the top, turned out to be the only thing that kept the company alive through the years that followed. Markets forgive the sold-into-strength management eventually, because the alternative is selling into weakness. Intel did not need the money the way that company did. But the principle is identical: capital raised at the top of enthusiasm is the cheapest capital a company will ever see, and the officers who decline to take it are making a market call with the balance sheet. Intel's officers took it. The nine-day fade that followed is the price of taking it — and it is a far smaller price than the alternative.
Follow the money to see who is losing right now. The buyers of the deal — the mutual funds, the sovereign vehicles, the index-adjacent institutions that take down offerings like this — paid ninety-five. Nine days later the stock traded below ninety-three, a paper loss of more than four hundred million dollars across the offering, before the underwriters' option even had a chance to put more shares into losing hands. In the bond-market phrase, the deal is underwater. Sometimes that means nothing; hot offerings stumble and recover all the time. Sometimes it means the market, having absorbed the new supply, has quietly decided it was the last trade at the top. The difference only shows in hindsight — but the speed of this one, nine days, is itself information.
Section Three. The Dilution Ledger.
Run the arithmetic the way a prospective shareholder should. Two hundred ten point five million new shares against roughly four and a half to four point seven billion existing shares: about four and a half percent of the company sold for nineteen point seven billion dollars in cash. That cash enters the balance sheet and immediately belongs to everyone — including the buyers who just paid ninety-five. If the foundry buildout earns returns above its cost of capital, the trade is accretive: four and a half percent of dilution bought a factory that prints cash for decades. If the foundry continues to burn cash the way Intel's has, the trade merely slows the bleed while spreading it over more shares.
The offering's quiet genius, from management's perspective, is what it does to the debt math. Tens of billions in new equity pushes the leverage ratios down, protects the credit rating, and buys years of investment runway without a single coupon payment. It converts a balance-sheet problem into a share-count problem — and share-count problems, unlike debt problems, never come due. The cost is borne diffusely, by everyone who held the stock, in the currency of ownership rather than interest. That is the entire theory of equity capital markets expressed in one deal: when your story is hot, sell the story; when it is not, you will wish you had.
There is also the signal to read. Companies that sell stock at the top of rallies are run by people who understand that windows close. Companies that buy back stock at the top are run by people who believe the window is permanent. Intel has done both in the last decade — the buybacks of the twenty-twenties boom at prices far above today's, and now this offering into a doubling rally. The market tends to forgive the second category of management and forget the first. The twenty twenty-six offering will be remembered as either brilliant market timing or the moment the company's own officers told you what they thought the stock was worth — and both readings are the same reading.
At My Audio Books dot A I, you can listen to this story and thousands of others that explore the hidden science and mechanics behind the headlines.
Section Four. What to Watch.
First, the underwriters' option. If the banks exercise their thirty-one and a half million share option in the coming weeks, the raise grows toward twenty-three billion — and the overhang of new supply grows with it. Second, the foundry pipeline. Every announcement of a new customer for Intel's manufacturing services — the hyperscalers, the sovereign funds, the chip designers without factories of their own — is a data point on whether the nineteen point seven billion is matched by revenue contracts or merely by hope. Third, the capital expenditure schedule in the next earnings report: the cash was raised for factories, and the market will now watch whether the factories are actually ordered, sited, and built on the timeline the story promised. Fourth, the rating agencies: the whole point of equity over debt was the balance sheet, and the agencies' next reviews will grade whether the trade worked. Fifth, the stock itself. Every dollar below ninety-five widens the under-water position of the offering's buyers, and the closer the calendar gets to the thirty-day option expiry, the more that mathematics matters. Sixth — and largest — the artificial-intelligence capex cycle itself. This offering is a leveraged bet on that cycle's duration, and no document filed with the S E C will move the stock a fraction as much as one soft quarter of data-center demand.
Section Five. The Broader Pattern and the Open Question.
Zoom out and the offering is a specimen of the new shape of the semiconductor industry: a national-champion company, mid-turnaround, funding a multi-decade factory program by selling stock into an artificial-intelligence rally at the exact moment the rally peaked. The pattern echoes every capital cycle in industrial history — railroads, fiber optics, shale — where the last and largest tranche of money is raised just as the thesis reaches its maximum audience. The fiber-optic builders of nineteen ninety-nine sold bonds into a bandwidth mania and went bankrupt when demand failed to arrive on schedule; the shale drillers of twenty fourteen raised equity into an oil plateau and spent a decade earning it back. Both of those cycles, it is worth remembering, eventually saw the demand arrive — just later, and at lower prices, than the financiers of the top had underwritten. The difference this time, if there is one, is that the underlying asset is not a commodity or a network but the physical substrate of computation itself, and demand for that substrate has so far broken every forecast ever made about its limits.
There is also a geopolitical ledger to the twenty billion. Every dollar of it was raised from private investors in a public market — but the factory it funds sits inside an American industrial-policy architecture of subsidies, export controls, and customer arm-twisting that has no peacetime precedent. The offering is, in effect, the private tranche of a public project: Washington de-risks the demand side with policy, and the equity market funds the supply side with mania. That hybrid is how the West now builds infrastructure it no longer trusts governments to fund alone — and Intel has become the single largest test of whether the hybrid can actually produce leading-edge chips at scale. If it works, the ninety-five dollar offering will read in retrospect like the financing moment of the re-shored semiconductor age. If it fails, it will read like the fiber bonds: real money, real factories, and a demand curve that arrived five years too late to save the financiers.
Which leaves the question the pricing tape cannot answer. Intel sold four and a half percent of itself for one factory, at ninety-five dollars a share, to investors who were under water nine days later. Was that the top of the foundry mania — or the last cheap financing of the most important industrial project in the American economy? The institutions that bought the deal have already voted with their marks. The factories, when they come online at the end of the decade, will cast the only vote that matters. Between now and then, the tape will oscillate, the option will expire, and twenty billion dollars will slowly turn into concrete, cleanrooms, and machines. That is either the shrewdest trade of the chip cycle or its most expensive act of faith — and unlike most such questions, this one has a physical answer, being built right now, one fab at a time.
At My Audio Books dot A I, you can create fiction, non-fiction, and turn your documents into audio, all stored in one place with a single subscription — plus get instant access to thousands of audiobooks and deep-dive investigations. Learn more today at My Audio Books dot A I.