Intel

IThe business

Intel designs processors and, unlike nearly every rival, makes most of them itself. In 2025 it had revenue of US$52.9B, flat on 2024, an operating loss of US$2.2B and a net loss attributable to Intel of US$267M R (reported).⁴² It ended the year with 85,100 employees, from 131,900 at the end of 2022, and at 26 September 2026 it had 81,100.⁹¹²¹⁰ Its head office is in Santa Clara, California. Its customers are mainly computer makers and cloud companies; the 10-Q names Dell, Lenovo, Amazon Web Services and Oracle among those it indemnifies in a patent case.²³ The x86 instruction set it controls is the foundation of most PC and server software.²

I need to say first how much has changed in a year, because the stock has moved with it. In August 2025 the US government agreed to take up to 433.3M new shares in return for US$8.9B of grant money it already owed. SoftBank paid US$2.0B at $23.00 a share and NVIDIA US$5.0B at $23.28.¹³¹⁴¹⁵ Altera, the programmable-chip unit, was 51% sold in September 2025.³ The stock was $36.90 at the end of 2025. It reached $140.94 on 22 June 2026, and on 12 August 2026 Intel itself sold 242M new shares at $95.00 for US$23.0B.⁵¹¹³ On 29 October it reported a good third quarter and guided the fourth quarter to US$15.0B to US$16.0B, below the third. The stock closed $92.07 that day, $65.76 the next and $59.97 on 2 November.¹¹⁰ I can’t say why the price fell 28% in a day; the release and the 10-Q name a PC slowdown that Intel expects to last into 2027, a fourth-quarter gross margin of 37.1% and an accident at the Israel plant, and I can’t say how much each weighed.¹⁰³

Three ways it earns money

Selling processors to PC makers. The Client Computing and Physical AI Group, CCPG, sold US$8.1B in the third quarter, 5% less than a year earlier, and earned US$2.3B at the operating line, a margin of 28%.³ Client volume was down 14% in the quarter and 12% for nine months, while average prices rose 9% and 17%, from richer mix and price rises to cover costs.³ Intel says demand is easing because memory prices have risen, and that this will get worse in the fourth quarter and last through 2027.³ That is a business shrinking in units and growing in price, which does not usually last.

Selling processors and accelerators to data centers. DCAI, the Data Center and AI group, sold US$7.3B in the quarter, up 78%, and earned US$3.0B, a margin of 42%.³ The reason is in the filing. Server prices rose 54% and volume 10%. Most of the price increase came from richer mix, with demand-based pricing and the need to cover higher input costs behind the rest.³ Intel says it cannot meet all of the demand because substrates and memory are short, and expects that to last into next year.³ This is the part of Intel that has changed. A year ago the same segment sold US$4.1B at a 23% margin.³ I treat a part of the 2026 jump as the price of scarcity and not as a new normal E (estimate).

Making chips for itself and, in time, for others. Intel Foundry runs the fabs, develops the process nodes and does the packaging. It sold US$17.8B in 2025, almost all of it to Intel Products at prices Intel sets, and lost US$10.3B at the operating line.² In the first nine months of 2026 it lost US$6.4B on US$17.9B of revenue, of which only US$878M was external, mostly from Altera becoming a customer rather than a sister company.³ In the third quarter, external revenue was US$411M, up US$379M, which Intel attributes to a non-recurring foundry contract.³ The filing says plainly that substantially all of Intel Foundry supports internal manufacturing.³

The process nodes

The nodes are the heart of the foundry question. At the start of 2026 Intel released its first products on Intel 18A, which it calls its most advanced node in volume production, and in June 2026 it began risk production on 18A-P, a version meant for outside customers.³ In the second quarter of 2026 it committed to finish developing the next node, 14A, with a first product designed for it, and it released the 14A 0.9 design kit in October.¹⁶³¹⁰ The 2025 annual report said that if Intel could not win a large outside customer for 14A, it might pause or stop the next-generation nodes.² By the third-quarter filing the wording had moved to “potential significant customers” evaluating 14A and the pace of building “ultimately dictated” by committed demand.³ That is a shift, but it is not a customer. In late September and October 2026, headlines said Apple, Nvidia, Broadcom, Amazon, Google, Microsoft and others were evaluating or had agreed to use 14A. None of these is in an Intel filing, and I treat them as unverified E (estimate).¹⁷³

The factories are where the money is spent. Net property, plant and equipment was US$105.3B at 26 September 2026.³ The company slowed the new Ohio plant and cancelled the planned German and Polish expansions in 2025; it is expanding Arizona, New Mexico and Oregon.²³ Brookfield owns 49% of the joint company that holds two Arizona fabs; in April 2026 Intel bought back Apollo’s 49% of the Ireland fab for US$14.2B.³¹⁸ Commitments for capital spending were US$19.6B at the quarter end, most of it due within a year.³

Where it stands today

Revenue for the first nine months of 2026 was US$46.2B, up 18%. Gross margin was 41.0%, against 34.3%, and it rose to 43.0% in the third quarter. The fourth-quarter outlook is a gross margin of 37.1%, with an earnings per share of $0.10 on a GAAP basis (US$0.25 non-GAAP).³¹⁰ Reported net income for the nine months was a loss of US$11.4B, but only because the value of shares owed to the US government, which Intel must mark to the stock price, went up by US$12.6B as the stock rose. That is an accounting loss, not cash.³ Operating income was US$1.2B after US$4.3B of restructuring and impairment charges, mostly a US$3.9B write-down of goodwill, nearly all on Mobileye.³

Two other facts belong here. In mid-October 2026 an accident at the Kiryat Gat fab in Israel caused employee casualties and injuries and halted production for about two weeks; Intel expects higher costs through the first half of 2027 and could not estimate the effect or the insurance recovery.³ A significant part of its revenue comes from Intel 7 chips made at that plant, and Intel is not insured for business interruption from war.³ The company also says Iran listed it near the top of US firms it would target.³ These are risks I can name but cannot price.

IIThe moat

The test of a durable business is what happens when conditions turn. Intel had one of the best records in the world until about 2020, and the numbers since are the test.

From 2016 to 2021 Intel’s gross margin ran between 55% and 62% and its operating margin between 22% and 33%, and from 2017 to 2021 it kept US$13.1B a year on average of cash after capital spending and share-based pay C (calculated).⁴ By 2025 the gross margin was 34.8% and the operating margin was −4.2%.⁴² Revenue fell 33% from the 2021 peak. Operating income swung from US$19.5B to a loss of US$11.7B in 2024.⁴ A company with a lasting edge does not usually lose a third of its sales and all of its profit in three years, and Intel did, to a rival (AMD) that owns no factory, and to a customer base that went elsewhere for AI.

What does protect it

• The x86 standard. Intel controls the instruction set, AMD depends on it, and a vast software ecosystem has been built for it over decades.²¹⁹ A customer can replace an Intel server processor with an AMD one without rewriting software, which is why AMD took share, but replacing x86 altogether is a decade-long job. Arm-based designs are the threat the filing names.² Data center revenue is now rising again and Intel says demand for its server processors exceeds supply.³
• Being the only leading-edge maker in the United States. The 10-K says Intel is the only company doing both leading-edge logic R&D and high-volume manufacturing in the US.² That has value to a government that now owns up to 8.4% of the company, after converting grants it had promised into shares.¹³⁸
• Packaging. The 10-K calls advanced packaging another key to the strategy, and says the most advanced products of potential outside customers use it.² Packaging is where Intel has most credibly sold to others. I could not find a revenue figure for it.

Where it is thin

Share. The best figures I found are Mercury Research’s as reported by Tom’s Hardware, in the AMD page on this site: AMD’s x86 server unit share was 34.5% in the second quarter of 2026, from 27.3% a year before, and its share of x86 server revenue was 46.2% in the first quarter.²⁰ Headlines in August 2026 put AMD’s client share above 30% and said Intel’s overall x86 share was at a 31-year low. I did not read Mercury’s report and treat the figures as indicative E (estimate).⁶ Intel’s own filing shows what share loss looks like: client volume down 14% in the quarter “due to lower demand and a competitive market environment.”³

The AI accelerator race, lost so far. Intel’s Gaudi accelerators produced US$922M of inventory charges in 2024, and the 10-K now describes a revised strategy of inference GPUs on an annual cadence, starting with Crescent Island, now called the Intel GPU D-series, and later Jaguar Shores.²¹⁰ The 10-K says NVIDIA’s GPU systems have had the highest demand.² The NVIDIA page on this site puts its data center revenue at 13 times AMD’s, and the AMD page says AMD’s data center segment has been larger than Intel’s in every quarter since the third of 2025.²¹²⁰ NVIDIA is now a shareholder and a partner: it agreed in September 2025 to buy US$5.0B of stock at $23.28 and to build x86 chips with Intel that connect to its NVLink fabric.¹⁵ That is a better place to stand than competitor, but it does not make Intel an AI accelerator maker.

The factory. An edge shows in profit, and the factory has none. Intel Foundry lost US$7.1B in 2023, US$13.3B in 2024 and US$10.3B in 2025, US$30.7B in all, then US$6.4B in nine months of 2026.²³ The 2025 10-K warns that Intel has over US$100B of property, plant and equipment, most of it tied to the foundry, on which it could lose money, and the company impaired and sped up depreciation on assets in the second quarter of 2025 after finding that its asset portfolio “exceeded manufacturing capacity requirements.”²³ That is an admission that Intel built ahead of its own needs.

My reading: the x86 franchise is a real but eroding edge; the factory is a strategic asset with no earned edge yet. The edge in the shares today is the option on the factory.

IIIThe competitive advantage

The question is what Intel can do that a rival cannot. It has two rivals in two different races: AMD and Arm for designs, TSMC and Samsung for factories.

The comparison is hard on Intel. TSMC sold US$121.4B in 2025, 2.3 times Intel, from 90,557 people, and earned a 50.8% operating margin.⁷ Intel earned −4.2% from 85,100 people.⁴ Both spent 33.4% of revenue on capital equipment, counting Intel’s equipment bought on extended terms: US$17.7B of US$52.9B for Intel, NT$1,272B of NT$3,809B for TSMC C (calculated).²⁷ The difference is what the spending produced. I go through this at length in the Valuation section.

AMD is the better comparison for the products. It had revenue of US$34.6B, a gross margin of 49.5% and an operating margin of 10.7% in 2025, with no factories and about 31,000 employees.²⁵¹⁹ Its server share is rising at Intel’s expense.²⁰ The AMD page on this site compares data center margins at 39.5% for Intel and 31.3% for AMD, and Intel’s was 42% in the third quarter of 2026, though the two are measured differently.²⁰³ I do not read it as a clean win, because a part of Intel’s margin is a shortage price, and because Intel’s Products segment buys its wafers from Intel Foundry at prices Intel itself sets.

What Intel can offer that others can’t

Three things, and each is a claim from Intel that outside customers have not yet paid for in size.

US-made leading-edge wafers. TSMC makes in Arizona, but the 10-K says Intel is the only company doing leading-edge logic R&D and volume manufacturing in the US, and TSMC lists its Arizona fab as one of many sites.²⁷ A customer worried about supply from Taiwan, or wanting a second source, may pay for that. The 10-Q says the scale and pace of Intel’s expansion will be “dictated by the amount of committed demand” and that it is seeking design wins from “potential significant external customers.”³ I read that as: no big customer has committed to 14A in a filing.

Packaging and chiplets. The 10-Q lists a packaging tie-up with Lens Technology and joint work with ASML, Lam Research and Applied Materials on High-NA EUV readiness.¹⁰ The ASML page on this site shows Intel’s first-half 2026 equipment additions of US$6.2B.²⁶ High-NA is a bet on the 14A timeline. If 14A slips, so do the tools’ returns.

A government and customer base with a stake in its survival. The Commerce Department owns shares, SoftBank and NVIDIA bought at $23, and Tesla and SpaceX have named Intel as a Terafab collaborator.^13141527 Those are signs of interest. They are not purchase orders.

Where rivals can catch up

AMD wins by shipping a new design each year on TSMC’s newest node, which Intel cannot match unless its own factory keeps pace. Intel said in its 10-K that most competitors use TSMC or Samsung, and that it too will keep using TSMC for some tiles.² So the biggest threat to Intel’s factory case is the one Intel carries in its own product plan: it still buys wafers from the best foundry for its best parts. The Intel 10-K says “We expect to continue to use TSMC and other third-party foundry suppliers for various key tiles in a number of current and future products.”² If Intel’s own designs don’t trust Intel’s own nodes, an outside customer will ask why it should.

IVManagement

Who controls it

No one controls Intel in the sense of a founder or a block. At the 2026 proxy date the largest holders were the US government at 8.4% (433.3M shares, assuming the escrow is fully released), Vanguard at 8.1% and BlackRock at 6.8%.⁸ The Commerce Department agreed to vote with the board’s recommendations, with exceptions for anything that would hurt its relationship with Intel or that touches its other rights.¹³ A shareholder sued in March 2026 in the Delaware Court of Chancery to unwind the agreement, and the case is in a motion to dismiss stage.³ If a reader wonders who really controls the company, the answer is that the board does, with a government stockholder that votes with it by contract and can step out of the contract on matters that concern it.

Lip-Bu Tan has been chief executive since 18 March 2025, after Pat Gelsinger left in December 2024 and the board chair, Frank Yeary, served as interim executive chair.⁸ Mr Tan is also the founder and chairman of Walden International, a venture fund, and the proxy discloses deals where Intel Capital and Walden each invested in the same company, such as US$3M each in a convertible note of a packaging-substrate maker in April 2025.⁸ That is disclosed, small and worth watching, because substrates are short and Intel has just signed two agreements to secure them at US$5.3B to US$6.1B in all.⁸³ Craig Barratt became chair when Mr Yeary retired at the 2026 annual meeting.⁸

How they are paid, and what that pays them to do

Mr Tan’s 2025 pay was US$92,990,900, of which US$49.6M was stock awards, US$39.5M options, US$2.4M a cash bonus and US$0.7M salary and other.⁸ His annual bonus target is 200% of a US$1.0M salary and paid out at 118.7%.⁸ The performance shares granted in 2025 pay on one measure only: Intel’s total shareholder return against the S&P 500 over three years, with target at the 55th percentile.⁸ That pays him for a rising share price compared with the index, over three years, and not for profit, cash or returns on the factories. I see that as a clear but narrow signal. It rewards the narrative that has just moved the stock from $18 to $141 and back to $60, and it does not ask him to earn the cost of capital on the US$105B of plant.

His own money is in. The proxy shows 16,471 shares held directly and 1,209,906 held indirectly, a total of 1,226,377 shares worth about US$74M at $59.97, plus unvested awards.⁸ That is about four-fifths of one year’s pay package, and the proxy lists a ten-times-salary ownership guideline with a compliance date of March 2030.⁸ The chief financial officer, David Zinsner, holds 364,918 shares.⁸ A director, Alyssa Henry, adopted a plan in August 2026 to sell up to 15,400 shares; that is small and routine.³

What they did with the money

This is the part of the record that matters most, because it covers a full cycle. From 2016 to 2024 Intel paid US$43.0B of dividends, and from 2016 to 2021 it spent US$47.2B on buybacks, mostly in 2018 to 2020 when the shares closed each year between $46.93 and $59.85.⁴⁵ From 2017 to 2025 it paid out US$82.7B in total, and in the same nine years its owner cash, my measure, was US$9.2B C (calculated).⁴ The rest came from borrowing: debt went from US$25.3B at the end of 2016 to US$46.6B at the end of 2025.⁴

It then spent US$93.2B on capital equipment in 2021 to 2024 to catch up in process technology and “shell ahead” of demand, as the 10-K puts it, and lost the cash flow it needed to do so.²⁴ The dividend was cut in 2023 and 2024 and not paid in 2025.⁴ In 2025 and 2026 it financed itself by selling pieces of the company: Altera, part of Mobileye, 49% of two joint fab companies, and new stock to the government, SoftBank, NVIDIA and the public, as in the table below.³² The shares sold at $20 to $23 in 2025 were worth, at the August 2026 offering price, four times what buyers paid. The company that bought back its own stock at close to $50 had to sell it at less than half that within five years. I take this as a record of capital allocation that was wrong in timing and scale, even with a better strategy now.

The new team has done several things well. It cut headcount by 22% in 2025, trimmed research spending from US$16.5B in 2024 to US$13.8B in 2025, slowed Ohio, closed the planned German and Polish projects, and bought back Apollo’s Ireland stake rather than leave a more expensive claim outstanding.⁹⁴²¹⁸ Spending on research and general costs was US$4.5B in the third quarter of 2026 against US$4.4B a year earlier, up 4%, so most of the saving came in 2025 and has been banked R (reported).¹⁰ The next step is not cost. It is customers.

VFinancials

The ten years show a business that was among the most profitable in the world, ran into a manufacturing problem, and has been paying for the repair out of other people’s money. The second half of this section is what an owner can take from it.

Revenue peaked at US$79.0B in 2021 and was US$52.9B in 2025. Gross margin fell 20 points. The drop in operating profit from US$19.5B in 2021 to a US$11.7B loss in 2024 is a swing of US$31.1B on a revenue fall of US$25.9B, which means more than all of the lost revenue went straight to the loss line, because the factory costs did not fall with it.⁴ Research spending rose from US$13.6B in 2020 to US$17.5B in 2022 and fell to US$13.8B in 2025.⁴

What an owner can really take out

Owners are paid in cash that is left after the business has paid for what it needs to stay in place. For Intel I define that as cash from operations, less all capital spending, plus the incentives that reduce it, less share-based pay, which is a real cost paid in shares.

On that measure the company kept US$13.1B a year on average in 2017 to 2021, and in 2022 to 2025 it lost US$12.5B, US$16.5B, US$18.3B and US$8.8B C (calculated).⁴ In the nine months to September 2026 cash from operations was US$13.2B, but capital spending, counting equipment bought on extended terms, was US$11.8B and share-based pay US$1.9B, so owner cash was about −US$0.3B C (calculated).³ That is better, and it contains US$1.7B of customer deposits and the benefit of being short of supply, which I would not count on to repeat. Intel’s own “adjusted free cash flow” was US$1.6B in the third quarter, but it adds US$0.8B of partner contributions, which are Brookfield’s money and not an earning.¹⁰ It also does not deduct the shares paid to employees.

The gap between reported earnings and owner cash is the point. In 2025 the net loss attributable to Intel was US$267M and owner cash was −US$8.8B. In 2026 the reported nine-month loss is US$11.4B and owner cash about −US$0.3B. Neither reported number is a good guide.⁴³ In 2026 the large gap runs the other way, because the loss comes from the mark-to-market on shares owed to the government.

The row that tells the story is the middle bar. Depreciation of US$10.8B in 2025 is the nearest figure to what it costs just to stand still, and the company spent US$17.7B counting equipment bought on extended terms.⁴² The 2026 plan is bigger: Intel has US$19.6B of capital commitments, which is a figure I take as the floor for 2027 E (estimate).³ Standing still is not free at Intel, and growing is not cheap.

The balance sheet

At 26 September 2026 Intel held US$54.0B of cash and short-term investments against US$50.1B of debt, a net cash position of US$3.9B, after the US$22.6B stock sale.³ Without that offering it would have been net debt of about US$19B. Debt rose from US$46.6B at the end of 2025, because Intel borrowed US$13.0B in 2026 (US$6.5B of long-term notes and a US$6.5B bridge loan that it repaid) and repaid US$9.0B.³ It also holds US$8.8B of equity investments, including its 49% of Altera, and has a non-controlling interest of US$14.4B for Brookfield’s Arizona stake and US$2.1B for the Mobileye minority.³ A credit agency lowered the rating from BBB+ to BBB in August 2025; Intel says it still has an investment-grade rating.²¹⁰

There are other claims. Intel carries a US$13.7B liability for the escrowed shares owed to the government, which settles in shares and not cash.³ It has accrued US$1.0B for the VLSI patent case and US$312M for a European Commission fine.³ It has US$9.8B of other purchase obligations, plus US$5.3B to US$6.1B of newly agreed substrate commitments, of which about US$3.3B is payable by the end of 2027.³ And it has a pending US$1.9B purchase of Classiq, a quantum software company, which it may settle in shares.³

The price in numbers

At $59.97 the equity is worth US$319B on the 5,315M shares on the 10-Q cover, or US$330B if I add the 132M escrowed shares and 63M of dilution from employee awards C (calculated).¹³ After subtracting the net claims I describe in Valuation, the enterprise value is about US$333B. That is 5.4 times my 2026 revenue estimate of US$61.7B, and about 53 times my 2026 operating income of US$6.3B before restructuring E (estimate). At the peak of 2020, when Intel earned US$23.7B, the stock closed the year at $49.82 and the market value was about US$202B, which is 8.5 times that year’s operating income.⁴⁵ The market is paying six times that multiple on about a quarter of that income.

VIValuation

First, how this goes badly wrong

Before any number, I ask what would most likely destroy the business or the investment, and work backwards. For Intel I can see three.

The factory never earns a return. Intel has put over US$100B into plant and loses US$6.4B in nine months making wafers for itself.³ If 14A does not win a large outside customer, Intel said it might stop pursuing the next nodes, and then the existing fabs serve only Intel’s own demand, which is about US$60B of revenue.² That is not a collapse. It is a business that earns a modest return on a very large base. My bear case is that, with a margin of 12% on revenue of US$67B in 2031.

The shortage price goes away. Server prices rose 54% and client prices 9% in the quarter, and Intel says supply of substrates and memory limits what it can sell.³ When supply returns, AMD, Arm-based designs and Intel’s own customers have more choice, and the price premium shrinks. Intel earned a 42% margin on data center sales in the quarter and 23% a year earlier.³ The truth is probably between them, and I have taken a margin on all of Intel of 21% in my base.

The company has to raise money again. Intel sold shares four times in fourteen months, once at $95.^11131415 If 14A needs capital before customers pay for it, the next sale would be priced off a stock that has fallen 57% from its high, and every existing holder would be diluted. Owners are protected only by cash earned, and Intel has not earned any.

Two further things can break it, and I cannot price either: a war that stops the Israel plant, which supplies a large share of Intel 7 output and which Intel does not insure against, and a conflict over Taiwan that changes the whole industry.³ I leave them out of the cases and say so.

What an owner can really take out

The cash is covered in Financials: US$9.2B of owner cash over nine years against US$82.7B paid out, and about −US$0.3B in the first nine months of 2026. For the model I use free cash flow after share-based pay and all capital spending, which is the same idea. In my base it is US$2.4B in 2027, US$14.5B in 2031 and US$18.5B in 2036. The first ten years of base cash have a present value of US$71B, which is 21% of the US$333B enterprise value the market pays C (calculated). The rest is cash after 2036.

What I can’t judge

Whether 14A wins a large outside customer. This is the input that moves the answer most, and nothing in the filings tells me. Whether the third-quarter data center margin of 42% is shortage pricing or new structure. How much of the Foundry loss is real and how much is a transfer price: about 95% of Foundry’s revenue is sold to Intel Products at prices Intel sets, so the Products margin of 35% and the Foundry margin of −28% in the third quarter are partly Intel’s own accounting. I therefore anchor the model on the consolidated numbers, which are market results, and not on a sum of segments.³ Products could not be valued at 35% without repricing its wafers at what an outside foundry would charge. And I cannot judge the effect of the Israel accident, the helium shortage the 10-Q names, or the Iran war. The range I show is the honest answer, not a false precise number.

How the cases are built

I start from 2026: nine months of US$46.2B of revenue plus US$15.5B, the middle of Intel’s fourth-quarter guidance, makes US$61.7B, 17% more than in 2025.³¹⁰ Operating income before restructuring is about US$6.3B, or 10%, which is the nine months (US$1.2B plus US$4.3B of restructuring) plus my reading of the fourth quarter, where the 37.1% gross margin and US$4.9B of operating expenses leave about US$0.9B E (estimate).³¹⁰ Share-based pay is inside it.

Base. Revenue grows 8% in 2027, 9% in 2028 and 2029, then fades to 3% by 2036. The first years carry data center strength and the weak PC market. Margin rises to 21% by 2031 and stays there. That is a Products margin of about 33%, a Foundry loss that narrows to about 10% of its revenue, and about US$4B a year of unallocated costs. Capital spending is 26% to 27% of revenue in 2026 and 2027 and falls to 18%. Cash tax is 12% and rises to 17%, as the valuation allowance and credits are used up. The discount rate is 11%.

Bear. No growth in 2027, then 2% a year; margins of 9% to 12%; capital spending falling to 16%. This is the factory without outside customers and the products without the shortage.

Bull. Revenue of US$123B in 2031 and US$182B in 2036, margins of 32%. That is a company that wins a large foundry customer and earns margins near its best year, on 2.3 times the revenue of 2025 by 2031. It is a generous case.

The discount rate

I use 11.0%. The capital asset pricing model starts with the 4.91% ten-year Treasury yield on 2 November 2026 and Damodaran’s 4.20% implied equity risk premium of 1 October.³⁰³¹ A beta of 1.45 gives 11.0%. Intel’s measured beta on month-end returns against the S&P 500 is 2.4 over five years and 4.3 over three, which would give 15% to 23%, but those numbers are mostly this year’s rise from $37 to $141 and the fall since, which I do not want to treat as risk forever.⁵ I sit between that and a market beta of 1. The sensitivity table shows the other rates.

Here is what I find striking. The price needs more than the bull case. Either margins above anything Intel has earned, or revenue of US$269B, or my bull case with revenue growing a little faster still. My base cash flows earn 7.0% a year at this price, which is 2.1 points more than the Treasury, for an owner who has seen the stock fall 57% in four months and the company sell shares at $95.

The margin of safety, and the price against the range

My margin-of-safety price is $17, 40% below the base of $28 and level with my bear of $17. I use a large gap because the range is wide, the key input is outside the filings, and the dilution of the last fourteen months is a reminder that surprises at Intel have gone one way. At $59.97, the price is 114% above the base and 3% above the bull. The market may be right that the factory wins, because the possible prize is large: a US-based leading-edge foundry with a government and customer base behind it is rare. I cannot show that it is wrong. I can show what it needs.

Quality and price

These are two questions. How good is the business? The data center franchise is good and improving, the PC franchise is mature and shrinking in units, and the foundry is unproven. That is a mixed business with one excellent part. How good is the price? Poor, if my numbers are close, because the market value is about US$330B for a company with 2026 operating income of US$6B. A fair business bought at a high price is a poor purchase, and a good business bought at the wrong price can be one too. This looks like the second case with a hole in it.

Scenario tests on the base

The table runs six stresses through the base. They are tests of the model, not forecasts.

The surprise is how little a one-year stumble costs. The Israel plant outage costs about 20 cents a share in this frame, because the value sits in the long run. What matters is anything permanent: a lasting margin of 17% rather than 21% takes $5, and a lasting share loss takes $6. The risk is not a bad year. It is a worse business.

The next best use of the same dollar

The first alternative is a Treasury at 4.91%.³⁰ Against it, my base cash flows return 7.0% at this price, with far more variation. The second alternative is the closest comparable on this site, TSMC, which makes the leading-edge wafers Intel hopes to sell. My TSMC page values it at NT$2,006 against a price 18% above that, and the company earns 50.8% operating margins on a third of revenue spent on capital.³²⁷ The comparison I draw is only about price against my own value: TSMC’s price is 18% above my base and Intel’s is 114% above. The same page logic applies to the others: NVIDIA is 5% above my base, ASML 25% above, AMD 160% above.²¹²⁶²⁰ By my own numbers, Intel is not the cheapest way to own AI chip demand. I say only that the price asks for more here.

Where this could be in 5 and 10 years

Five years out, my base has Intel earning US$19B at the operating line, which is below its 2020 level, on revenue that is 16% above its 2021 peak. Ten years out, US$24B. For my bull to be right Intel would need to earn two-and-a-half times its best year. The odds, multiplied out: say a 35% chance of a large outside 14A customer by 2028; a 50% chance that data center grows faster than 12% a year for five years; a 25% chance that the factory’s unit cost falls to near TSMC’s level; and a 60% chance of no further large equity raise. If those were independent, the bull needs all four, 2.6% of the time. They are not independent, since a big customer helps the other three, so I put the bull nearer 5% to 8% E (estimate). The price asks for slightly more than the bull.

VIIThe case for the price

What most people miss

Three things cut in Intel’s favour and get less attention than the headlines. First, the government’s shares were sold to it at about $20 and carry a contract that the government votes with the board, which makes the largest shareholder an unusually stable one, and it is also one that benefits from the stock being high.¹³⁸ Second, Intel’s customers pay it deposits: US$1.7B arrived in the second quarter of 2026 under long-term supply arrangements, and US$1.2B sits in long-term liabilities.³ That is a sign that buyers fear a shortage, which is also the reason I do not trust the margin. Third, Intel has US$54.0B of cash and short-term investments against US$50.1B of debt after the August offering, so the next two years of capital spending are funded without another raise, if the plan holds.³

The bull case

In plain words: Intel was the best chip company in the world, lost its way on manufacturing, and has a new chief executive who has cut headcount by a fifth, shipped the first 18A products and posted, in the finance chief’s words, the strongest data center revenue growth in 15 years. The U.S. government, the leading AI chip designer and a large Japanese investor have each put money in at prices well below today’s. The 14A node is the first designed for outside customers. If a handful of the largest chip designers move some of their volume to US fabs for supply security, Intel Foundry fills its plants, and the loss turns into a profit.

The named holders have said this in their own press releases, and I give the dates. On 18 August 2025 SoftBank’s chairman, Masayoshi Son, said Intel would play a critical role as advanced chipmaking and supply expand in the United States.¹⁴ On 18 September 2025 NVIDIA’s chief executive, Jensen Huang, called the collaboration historic and said it joined NVIDIA’s CUDA stack with Intel’s CPUs and the x86 ecosystem; NVIDIA agreed to buy US$5.0B of stock, and Intel agreed to build custom x86 data center chips for NVIDIA, linked through NVLink.¹⁵ Mr Tan, in the 29 October 2026 results, said that demand for its server processors was strong, that its immediate priority was supply, and that near-term headwinds hit the PC industry.¹⁰ NVIDIA’s own holdings filing for 30 June 2026 valued its Intel stake at US$30.0B, according to the NVIDIA page on this site.²¹ Those are all statements of intent. The nearest thing to a commitment is the product work: Intel has agreed to build a custom x86 chip for NVIDIA and a PC chip with NVIDIA graphics.¹⁵ For the foundry, the named support is government and investor money. No filing names a customer for 14A.

Against reality

If Intel keeps data center revenue near US$7B a quarter at a margin above 35%, brings the PC business back to growth, cuts the Foundry loss to about 10% of its revenue by 2031 and holds capital spending to 18% of revenue, then my base applies and the value is $28. That is 114% below the price. To reach $59.97 the model needs either a company that earns US$61B at the operating line in 2036 or a margin of 41% by 2031. Those are possible. They are not what Intel’s filings show today, which are operating income of US$2.5B in the third quarter, a Foundry that lost US$1.9B and a company that, in nine months, kept no cash for its owners R (reported).³

The gap to the price is mostly the factory. At a TSMC-like cost position for its wafers, the Foundry would add up to $28 a share to my base, which gets to about $56. I show the arithmetic in the next section. It is the one route that gets close, and it asks Intel to cut the cost of making the same wafers by more than half.

I multiplied out the odds for the bull case in Valuation: about 2.6% if four conditions were independent, perhaps 5% to 8% with the overlap E (estimate). The price asks for slightly more than the bull. So the stock is priced for success, and the question is whether the filings show success arriving. At the moment they show a strong product quarter, a fourth-quarter guide that is lower, a PC market Intel expects to get worse, and a factory still losing money.

What a 10x would take. Ten times $59.97 is $600, a market value of about US$3.3 trillion on 5,510M shares. At 25 times earnings that needs about US$132B of net income, which is 12 times the annual rate of the US$2.76B non-GAAP net income in the third quarter C (calculated).¹⁰ At a net margin of 25%, a level Intel reached in every year from 2018 to 2021, that is revenue of about US$529B, ten times the US$52.9B of 2025 and 6.7 times its highest in these ten years, US$79.0B. Reaching it in ten years needs revenue to grow about 24% a year from 2026. I know of no company that started at a market value of US$330B and multiplied it tenfold in ten years. It is close to impossible, and I would not build a plan on it.

Where I come down

What moved me was the data center quarter: US$7.3B at a 42% margin from a segment that earned US$964M a year earlier, plus a cash cushion that no longer depends on the next capital raise. What did not move me was the factory, which has no outside customer I can name, and the price, which already assumes one. What would raise my values: a named, committed 14A customer with volume; foundry losses under US$1B a quarter with external revenue above US$2B; and owner cash positive for two years running. Until then my values stay at $17, $28 and $58, and my margin-of-safety price at $17.

VIIIWho could run it better, and could it be taken over?

Scenario analysis, not reporting. Everything below about what might happen to Intel’s ownership or structure is my own analysis of public filings. No filing I read says any of it is planned, and I treat press talk as unverified unless a filing backs it.

The question is fair for a company that lost US$30.7B in its factories in three years and sold new shares at $20 and at $95. Could someone run it better? Could it be bought? I can put facts on both. I can’t give odds that mean much, and I do not try to predict a deal.

What the filings fix in place

Three terms in the government agreement shape every option. The Commerce Department can get up to 433.3M shares in all; it must vote them with the board, except on matters about its own rights or its relationship with Intel.¹³ It also holds warrants for 240,516,150 shares at $20.00, exercisable only if Intel stops owning at least 51% of its foundry, and they expire in five years.¹³ At $59.97 the warrants are worth about US$9.6B in intrinsic value, which is a cost to other shareholders of any move that lets control of the foundry go. Nothing in the agreement, as I read it, stops a buyer from acquiring all of Intel, because the whole company would still own 100% of the foundry. Brookfield owns 49% of the joint company that holds two Arizona fabs, and Mobileye is 76% owned and listed separately.³³⁶ Those are the existing pieces.

A stockholder has also sued the directors and the Commerce Department over the agreement, asking that it be cancelled. Motions to dismiss were filed in May 2026.³ A buyer or a government that wanted to change the terms would have to deal with that case.

Option 1: split into Intel Products and Intel Foundry

The company already reports the two separately, which makes a split look tidy on paper. On the numbers it is not. The foundry sold US$6.3B to Products in the third quarter and US$0.4B to everyone else.³ A standalone foundry with that revenue and a US$1.9B quarterly loss could not borrow on its own and would need Products as an anchor customer on a contract, at a price that must be high enough to cover its costs. A price that high would cut the Products margin of 35%, which is why I won’t value the two as a sum of parts at their current segment margins. Intel’s own segment note says the foundry margin is partly set by intersegment charges.³ Products would also want the option to use TSMC for any part where the Intel node trails, as it does today.² The government’s warrants make the split expensive for existing holders unless Intel keeps 51%. A sensible version is a 51%-owned foundry with outside investors putting in fresh money: the structure Intel has used with Brookfield in Arizona and, until April 2026, with Apollo in Ireland, whose 49% it bought back for US$14.2B.³¹⁸ Whether that capital was cheap depends on the price Apollo paid in 2024, which I did not pull.

Option 2: a national champion with government and customer backing

This is already happening in part. The U.S. government is a holder, NVIDIA has US$5.0B in the stock and a product collaboration, SoftBank US$2.0B, and the company talks of “potential significant” 14A customers.^1315143 Press headlines in September and October 2026 said Apple, Nvidia, Broadcom, Microsoft, Amazon and others were evaluating or had agreed to use 14A. These are reports, not filings, and Intel’s external foundry revenue was US$411M in the quarter. Even a signed customer is not a business. The reason is simple. A customer will commit wafer volume to a factory only if it believes the factory will make its chips at a competitive yield on a schedule it can plan for, and the government can ask but not guarantee that. A customer that is also a stockholder, as NVIDIA is, helps the signal. The most useful version is an advance payment or a purchase obligation, which Intel can show in its filings as deposits. It has received US$1.7B of deposits in 2026, but the 10-Q presents those as supply arrangements for products, not as foundry prepayments.³

Option 3: TSMC or another operator

The press has speculated that TSMC could run or invest in US fabs, and headlines on 7 and 15 October 2026 said Musk denied a TSMC takeover of Terafab and that TSMC confirmed it was considering a Terafab expansion; they differ on who would run it.³⁷³³ I could read only the headlines. I found no filing or statement of any talks between TSMC and Intel. If there were, several rules limit the outcome, and I give my understanding as a non-lawyer, not as legal advice. Antitrust. Intel and TSMC are the two leading-edge logic makers of significance outside Samsung, so a control deal would need merger clearance in the United States, and almost certainly in the EU and China, where Intel and TSMC both sell. The Intel 10-K itself says it holds a position in leading-edge manufacturing that it describes with TSMC and Samsung as the only three companies working on 2-nanometer class transistors.² Export and technology rules. Taiwan treats its leading-edge processes as strategic, and the U.S. restricts what advanced chip technology can go to some destinations. National security. The U.S. government is a holder, and Secure Enclave is a CHIPS Act programme for which Intel says it completed the RAMP-C step.¹⁰³ A foreign operator in control of Intel’s fabs would meet resistance from the very party that now owns shares. A minority technical role, such as a process or equipment partnership, is more plausible and does not need these approvals in the same way.

Option 4: a private-equity or strategic takeover

The arithmetic is simple. The market value is about US$319B on the shares outstanding. A 30% premium, which is common in take-privates, would be about US$414B for the equity, and Intel also has US$50.1B of debt, US$14.4B of Brookfield’s stake in Arizona and other claims. The largest leveraged buyout I know of, Electronic Arts at about US$55B in 2025, which I have not rechecked against a filing for this page, is a sixth of Intel’s market value. Intel’s operating income is US$6B before restructuring, so debt could carry very little of the price, and a buyer would have to put in hundreds of billions of equity. The only buyers that size are large public companies, which raises antitrust problems, or governments and sovereign funds, which raise a second problem: a foreign buyer is subject to review by the Committee on Foreign Investment in the United States, and the U.S. government already owns shares. Private-equity firms have already been partners: Brookfield in Arizona, Apollo in Ireland, Silver Lake in Altera.³ They bought pieces, not the company. A takeover of the whole is therefore not a real option at the price of $59.97, and I assign it no probability.

Option 5: Elon Musk and Terafab, only what I can verify

Here is what the documents say. SpaceX’s final prospectus of 12 June 2026 describes Terafab as a chip manufacturing initiative with a long-term goal of making one terawatt of compute hardware a year, announced with Tesla in March 2026; it says Intel joined in April 2026 and is expected to contribute its expertise in designing, fabricating and packaging chips; and that specific projects, timelines and spending have not been determined.²⁷ It also says Terafab may not succeed, that SpaceX expects to keep buying a significant share of its chips from others, and that neither Tesla nor Intel is obligated to remain part of the project.²⁷ Tesla’s proxy letter of 29 October 2026 calls Terafab a way to secure supply, and does not mention Intel.³⁸ Tom’s Hardware headlines say Intel joined the project (7 April), that Musk said Terafab would use Intel’s 14A process (23 April 2026, in what the headline calls a likely licensing deal), that a filing for a US$55B fab in Texas was made (6 May), and that Intel reaffirmed its 14A role while Musk denied a TSMC takeover (7 October).^39404137 Intel’s Q3 release does not mention Terafab.¹⁰

What I could not verify. I found no filing or company statement of any offer or talks about Musk, SpaceX or Tesla acquiring Intel or its foundry. Headlines from 28 May 2026 are columnists’ predictions that SpaceX will buy Intel, and a 20 October headline says Musk once tried to buy chip plants from Intel, TSMC and Micron and was turned down. I treat both as unverified and do not rely on them.³⁵ Keep these apart: a Terafab collaboration is documented, a purchase of Intel is not.

How Intel’s efficiency compares

The question under all of these is whether someone else would get more out of the same assets. The filings let me compare, though not on the measure that matters most, fab utilisation, which none of the four companies discloses.

Intel has about half the revenue per employee of TSMC ($621,000 against $1,341,000), a gross margin 25 points lower, and an operating margin 55 points lower, while spending the same third of revenue on capital. Its asset base earns 50 cents of revenue per dollar of net plant and equipment, against $1.03 at TSMC. A part of the difference is mix: Intel also designs and sells chips, and its research covers products, not only process. AMD has a similar research share (23.4%) and a 49.5% gross margin, with no factories. Samsung is the weakest comparison: its chip division makes mostly memory and its foundry is reported inside it, so I show only company-wide margins. The Samsung page on this site cites a Kiwoom estimate that the foundry and System LSI unit lost ₩6.74T in 2025.³⁴ That is the other leading-edge alternative, and it is also loss-making. The TSMC advantage has three sources that the filings show: 17 million wafers a year of capacity, 534 customers in 2025 and a 59.9% gross margin that pays for 6.5% of revenue in process research.⁷ Intel has few outside customers.

What a TSMC-like operator could add: a what-if

I take only the foundry. Its cost of making wafers was 158% of its revenue in 2025 and 128% in the third quarter of 2026, against 40.1% of revenue at TSMC. In my base I already assume the foundry loss narrows to 10% of its revenue by 2031. The what-if asks what happens to my base value if the same wafers cost less to make, with the revenue held fixed at US$32B in 2031, a number I derive from a run-rate of about US$24B to US$27B and growth of about 6% a year. A cost cut does not make Intel into TSMC, because the price Intel charges Products is its own number: only a lower real cost, or outside sales at outside prices, is a gain to the owner.

Read the table down. Getting the foundry to break even adds $4.6 a share, 16% of my base. A margin of 15% adds $11.6. Matching TSMC’s 50% margin would add $27.8 and take the value to about $56, near the price. So the market price is, roughly, my base plus a TSMC-like foundry. That is a compact way to state what the shares assume. It would require Intel to make wafers for 45% of what they cost it today at the 2031 volume, with yields, scale and customers like TSMC’s, and to do it inside five years. As a rough check on odds: say a 40% chance of yield parity on 14A, a 40% chance of enough outside volume, a 30% chance of pricing power with customers and a 50% chance of doing this without a slip in time. Multiplied, that is 2.4% E (estimate). Even if I double each, the odds are small. I read the what-if as the size of the prize, not its likelihood. A more efficient operator could add several dollars a share if the foundry reached break-even, and the odds of that are better than the odds of a TSMC-like result.

My read

Better management matters, and the evidence is that this team has done better on costs, cash and product than the last one. The harder problem is structural: Intel is trying to be a customer and a competitor to the same people, and its factories were built for a volume it does not have. A split, a consortium or a partner would help only if they bring wafer orders, not money alone. A takeover is out of reach on size and politics at $59.97. If someone were to buy Intel or its foundry, I would expect them to ask first what an outside customer would pay per wafer, and the answer is not yet in any filing.

IXRisks, and what would change my view

The three ways this goes badly wrong are at the start of Valuation. These are the rest, and a test of each.

• Customer concentration and the shortage. The data center jump rests on hyperscaler demand and on short supply of substrates and memory. Intel expects the shortages to last into next year.³ If hyperscalers build their own processors or buy AMD, DCAI revenue could fall back toward US$4B a quarter. My test: the scenario table shows what prices normalising costs.
• The Israel plant and the Iran war. Intel 7 products come from the Kiryat Gat fab; the plant was halted for about two weeks in October; Intel is not insured against business interruption from war; and a helium shortage from strikes on Qatar has hit the industry.³ I cannot estimate the odds.
• Dilution and the escrowed shares. 132M shares remain in escrow for the government, and US$7.2B remained available under the buyback authorisation, with no shares bought in the third quarter.³ If the stock rises Intel records a loss on the escrow liability, and if it falls it records a gain: a feature of the accounting that makes reported earnings a poor guide.
• Litigation. VLSI (US$1.0B accrued, with an appeals ruling in April 2026 sending one case back), a European Commission fine of €237M on appeal, and patent suits against Intel’s customers that allege over US$2.0B in damages, though the first to reach trial found no infringement.³ None is large against the balance sheet, though the VLSI total has no upper bound in the filing.
• The government as a holder. The agreement is subject to unilateral change by the Commerce Department to follow federal law, and other branches did not commit to support it.¹³ A policy change could affect Secure Enclave payments or the escrow release.
• Mobileye. Its goodwill of US$4.3B has less than 10% headroom, after a US$3.9B write-down in the first quarter, and Intel holds 76.2% of it.³³⁶ It is worth about US$3.7B at $5.65 a share, which I derive from the share count.⁴²³⁶

What a bad year looks like in the model

The scenario table in Valuation reruns six stresses. The model says a one-year stumble is cheap and a permanent change in margin or share is expensive. The signposts below tell me which one I am seeing.