Intel went from $18 to $92. What that price now assumes

On 8 April 2025, Intel closed at $18.13. The company that had defined American semiconductors for forty years was worth about $79 billion — roughly a fifth less than the $99.8 billion of book equity on its own balance sheet. The market was saying, in the plainest way it can, that Intel’s assets were worth more dead than working.

Fourteen months later the stock touched $140.94. On 20 August 2026 it closed at $92.13.

Something changed. This article is about what it was, because the answer is narrower and more specific than “AI,” and because the same filings that explain the recovery also show what has not been fixed.

Why it was cheap

Intel’s problem was not complicated. It had stopped making money.

Operating income went negative in the second quarter of 2022 and stayed negative, with two brief exceptions, for three years. The bottom was the September 2024 quarter, when Intel lost $9,057 million at the operating line and quarterly gross profit fell to $1,997 million — less than a sixth of what the same line had produced in 2018.

Two things were happening at once, and only one of them was visible in the headline. Intel’s products still sold. What consumed the money was the foundry — the manufacturing arm Intel had bet the company on, which was spending tens of billions to build capacity for customers it did not yet have.

That is the setup. A profitable products business, a manufacturing business losing billions a quarter, and a share price that had given up on the arithmetic ever working.

What changed, in one table

Intel publishes segment results with its earnings, and the June quarter’s table is the clearest explanation of the last year that exists.

Data Centre and AI operating income nearly quadrupled. Revenue in that segment went from $3,939 million to $6,262 million, and because Intel was already carrying the cost of the factories, most of the extra revenue fell through to profit. That single line contributed $1,841 million of the $4,972 million swing.

The foundry improved too, losing $1.1 billion less than a year earlier. But read the number rather than the direction: Intel Foundry still lost $2,089 million in three months. The thing that broke Intel is not fixed. It is losing less.

So the consolidated profit is not a story about the foundry turning around. It is a story about the products business finally earning more than the foundry loses — $4,817 million against $2,089 million, with roughly $900 million of corporate costs in between.

That distinction matters, because the two versions imply very different futures, and only one of them is in the filing.

The government took a stake, and it cost Intel $12.5 billion

On 22 August 2025, with the stock at $24.80, Intel signed an agreement with the US Department of Commerce. Under the CHIPS Act Secure Enclave programme, Intel put a block of its own shares into escrow, to be released to the government as Intel performs and receives cash.

The terms are in Intel’s own glossary, and they are unusual:

If Escrowed Shares are not released from escrow at the end of the performance period, half of the shares will be released to the DOC with no consideration and the other half will be forfeited and cancelled.

Here is the part that produced an $11 billion loss. The obligation is denominated in shares, not in dollars. Intel accounts for it as a derivative liability carried at fair value, so every time the share price rises, the obligation gets more expensive and the increase runs through the income statement.

Intel’s shares rose 211.5% during the June quarter alone, from $41.19 to $128.32.

27 December 2025        $2.7B
27 June 2026           $15.6B
───────
Charged to income      $12.9B

Of that, $12.5 billion hit the June quarter, inside a line called “interest and other, net” that came to −$12,576 million against −$95 million a year earlier. It turned a $1,796 million operating profit into a net loss attributable to Intel of $11,033 million.

Intel’s own reconciliation puts the same item at $2.45 per share, against a GAAP loss of $2.16 and non-GAAP earnings of $0.42.

Nothing left the company. No cash moved. But the obligation is real, 143 million shares are still sitting in escrow, and the liability will be marked again at the end of every quarter — up if the stock rises, back down if it falls. Intel also issued the government warrants over 241 million shares at $20.00, exercisable only if Intel ceases to own a majority of its foundry. A shareholder sued in Delaware in March 2026 to have the whole agreement voided; motions to dismiss were filed in May and the case is unresolved.

Then Intel sold $20 billion of stock

On 10 August 2026, three weeks after results, Intel announced a $15 billion share sale. Two days later it priced — at $20.0 billion.

210,526,315 shares at $95.00
───────────────
Gross proceeds                $19,999,999,925
Underwriting discount            −$329,999,999
───────────────
Proceeds to Intel             $19,669,999,926

A deal a third larger than announced, at 2.6% below the last trade, is what a company does when the demand is there and it wants the money. Intel says the proceeds are for “general corporate purposes, which may include, but are not limited to, capital expenditures and working capital.”

What makes it interesting is the timing. In the same half-year, Intel’s capital expenditure fell 29.1%, from $8,733 million to $6,192 million, and its free cash flow went from −$5,870 million to +$1,910 million. A company whose cash generation had just turned positive, and whose capital spending was coming down, raised $20 billion of equity anyway.

There is a reading in which that is obvious — a company rebuilding fabs wants years of funding secured, and the cheapest moment to sell equity is when your stock has just gone up five times. The filings do not say which reading is right. They do say the money was raised.

Intel was not alone in that three-week window. Four companies raised $52.35 billion for capacity between 27 July and 19 August, using five different instruments — and only two of them create interest expense. Equity, the one Intel chose, is not among the two.

What the price assumes now

Which brings us to the number most coverage skips.

Intel’s share price is up 5.1 times from the April 2025 low. Its market value is up 6.1 times. The gap between those two figures is dilution: shares outstanding went from 4,362 million to 5,253 million, a rise of 20.4%, as the government escrow released, employees vested, and finally 210 million new shares were sold in August.

At $18.13 the market paid 0.79 times Intel’s book equity. At $92.13 it pays 4.51 times. On the June quarter’s non-GAAP earnings annualised, it pays about 55 times.

We are not going to tell you whether that is too much. But it is worth being precise about what has to be true for it to work, because the filing tells you: the products business has to keep growing into AI demand, and the foundry has to stop losing $2 billion a quarter. At the low, the market was pricing the second thing as impossible. It is now pricing it as solved. The June quarter shows it half-solved, improving, and still deeply loss-making.

Intel at least gives you the filings to argue with. Not every repricing does — Moderna added $44 billion in a session on a trial result that appears in no SEC filing at all, which leaves a reader with the last quarterly report and nothing else.

What would make this wrong

• One quarter is one quarter. Intel lost $3,136 million at the operating line in the March 2026 quarter, three months before this one. The series is not a clean line and we have not smoothed it.
• Intel Foundry’s revenue is mostly internal. Of $5,765 million in the June quarter, the large majority is sales to Intel’s own product divisions, eliminated on consolidation. Growth in that segment is not the same as winning outside customers.
• The escrow charge reverses. If Intel’s shares fall, the liability is written down and the income statement gets a gain of the same character. Neither the loss nor a future gain is cash.
• Book value is a weak yardstick for a company like this and we use it only because it is the measure the market itself was using at the low. A semiconductor manufacturer’s assets are not worth their carrying value in a bad scenario, and are worth far more than it in a good one.
• The annualised earnings multiple annualises one quarter, which is the crudest possible method. It is here to give the order of magnitude, not a valuation.
• These are unaudited quarterly figures, and the share prices are market data, not filings.

Check it yourself

Every figure above is in two documents. The segment table is on page 9 of the Q2 2026 earnings release, under “Supplemental Operating Segment Results,” and the same table for the prior-year quarter is on the page after. The Escrowed Shares note, the 143 million share count, the warrants and the Delaware case are in the notes to the 10-Q. Shares outstanding after the offering are on the cover of the 424B5, and the quarterly operating income series is one API call, linked below.

The only things we added are the ratios, and the arithmetic is in the article.

If a figure here does not match a filing, tell us and it will be corrected on the article and on the corrections log, with the date.