Intel closed a $20 billion common stock offering on Wednesday, and the striking part is not the size. It is that the company asked for $15 billion, got more than $100 billion of demand, and walked away with a book roughly five times oversubscribed for capacity that is still years from producing revenue.
The mechanics, per Intel’s own pricing release: 210,526,315 shares at $95 each, gross proceeds of $20 billion, net proceeds of about $19.7 billion, with a 30-day underwriter option on a further 31,578,947 shares. J.P. Morgan, Goldman Sachs, Morgan Stanley and Citigroup ran the books. Stated use of proceeds is the deliberately vague “general corporate purposes, which may include capital expenditures and working capital.” Everyone involved knows what that means: fabs.
The Order Book Behind the Raise
Here is the tension nobody at the roadshow needed to say out loud. Intel is raising equity to build 14A capacity, and the publicly confirmed customer list for 14A is thin.
Elon Musk confirmed in April that Tesla would use Intel’s 14A process for silicon destined for the Terafab complex in Austin, and SpaceX is widely expected to follow. Beyond that, Intel has described ten long-term supply agreements covering CPUs and XPUs, running three to five years, some carrying both volume and price commitments. Those are real contracts. They are also mostly for products Intel already makes, not for the next-generation foundry node the capital is meant to serve. 14A is not scheduled to reach risk production for internal products until the second half of 2027, with high-volume manufacturing in 2028.
So the sequence is: raise the money, build the plant, then sign the customers. That is the opposite of how a merchant foundry normally de-risks a node, and it is precisely the chicken-and-egg problem that has defined Intel Foundry since its creation. No large fabless customer commits to a node that does not yet have proven yield at volume, and the node cannot get proven yield at volume without the capital to build it. Somebody has to move first, and Intel has now decided that somebody is Intel, using other people’s money.
To be fair to the bull case, the business underneath is genuinely healthier than it was. Intel posted second-quarter revenue of $16.1 billion, its fastest growth in nearly fifteen years, with the data center segment up 59% on AI demand. Chief financial officer David Zinsner raised 2026 capital expenditure guidance above $20 billion and told investors to expect 2027 spending significantly above that, with the vast majority going into the US network. We covered the customer-list question when the 18A roster drove the rally earlier this week, and the same caveat applies at 14A, one node further out.
What a Government Stake Does to the Cost of Capital
The reason a company with an unproven foundry order book can print $20 billion of equity into a five-times-covered book is not really about yield curves. It is about who else is on the register.
The Trump administration took a stake of nearly 10% in Intel roughly a year ago, converting about $9 billion of CHIPS Act grants awarded under the Biden administration into equity, which made the federal government Intel’s largest shareholder. That position has worked out extraordinarily well for the Treasury, with reporting putting its current value in the region of $40 billion after a year in which Intel stock rose about 175% and roughly quintupled from its lows.
That stake does something to Intel’s cost of capital that no balance sheet metric captures. An investor buying this offering is not underwriting only a semiconductor turnaround. They are buying into an asset that the United States government has already declared strategically indispensable, has already put taxpayer equity behind, and would be politically incapable of allowing to fail. It is not a formal guarantee. It functions like one anyway, and it is why the book cleared at a price that a standalone foundry with one named 14A customer could never have supported.
Worth noting for existing holders: issuing 210.5 million new shares dilutes roughly 4% of the company away from everyone already on the register, the government’s position included. Senator Elizabeth Warren has already pressed the Commerce Secretary on the terms of the original stake, and the question of whether taxpayers are being treated as strategic investors or as a subordinated backstop is going to get louder as Intel keeps issuing paper.
The Bet, Stated Plainly
Strip out the national-champion framing and Intel has made a straightforward wager: that AI compute demand through 2028 will be large enough, and supply constrained enough, that anyone with leading-edge capacity in the United States will be able to sell it regardless of who signed up in advance.
That is not obviously wrong. The capital flooding into AI infrastructure is enormous and increasingly creative about how it gets financed, as Nvidia’s $500 billion Wall Street financing structure demonstrated this week. If the buildout continues at anything close to its current pace, domestic leading-edge capacity will be scarce and Intel will have been early rather than reckless.
The risk is the mirror image. Equity raised at $95 is permanent, non-callable capital, and it was raised on a stock that has quintupled in a year. If AI capex normalizes before 14A reaches high-volume manufacturing in 2028, Intel will own a lot of very expensive concrete, a diluted shareholder base, and a government stakeholder with a political interest in the outcome. Fabs do not have a resale market.
What to watch over the next four quarters is narrow and specific: named external 14A customers. Not letters of intent, not framework agreements, not another supply deal for Intel’s own CPUs. A signed, disclosed, third-party commitment to buy 14A wafers at volume is the only datapoint that converts this raise from a bet into a plan.
Until one lands, Intel has $19.7 billion in the bank and a very expensive argument to prove.