How Intel makes money: The chip titan carving itself up to save the factories

For three decades, almost every PC and server on earth ran on an Intel chip. The "Intel Inside" sticker sat on a billion laptops, and the company's processors were the quiet engine of the internet age. Intel did not just design those chips. It built them, in its own factories, and that is exactly where the trouble started.

Intel still made $52.9 billion in 2025. But hidden inside that number are two very different companies. Its chip-design business earned $12.7 billion in operating profit last year, while its factory business lost $10.3 billion. A company that once printed money now barely breaks even, and in 2025 the US government, Nvidia, and SoftBank put in more than $20 billion to help keep it going.

In this breakdown, we'll unpack exactly how Intel makes money: the two businesses stapled together inside it, why the factories bleed cash, who is circling its most profitable markets, and whether the most expensive turnaround in chip history can actually work.

Intel was founded in 1968 by Robert Noyce and Gordon Moore, two engineers who had just left Fairchild Semiconductor. It went on to build the x86 microprocessor, the chip design that ran the personal computer era. For decades, Moore's Law and the "Intel Inside" campaign made it the most important company in computing.

The choice that defined Intel, and now threatens it, is that it makes its own chips. Intel is an integrated device manufacturer, or IDM: it both designs processors and fabricates them in its own plants, known as fabs. That is the opposite of how most of the industry now works.

Rivals like AMD and Nvidia are fabless. They design chips and pay someone else to manufacture them, almost always Taiwan's TSMC, the world's largest foundry. A foundry is a factory that makes chips to order for other companies. Staying fabless means never carrying the cost of a fab, and a leading-edge fab is one of the most expensive things a company can build.

Owning the factories was Intel's moat for forty years. It could tune its designs and its manufacturing together and stay a step ahead. Then the moat became a trap. Intel fell behind on manufacturing technology, missing the shift to the newest production methods that TSMC mastered first.

The misses compounded. Intel was slow to the smartphone chip, which went to designs based on Arm architecture. It lost the artificial intelligence wave to Nvidia, whose graphics chips turned out to be ideal for training AI models. And it steadily gave up PC and server share to AMD, which was suddenly building faster chips at TSMC.

Today Intel sells a handful of core products:

By 2024, the strain broke into the open. Intel lost $18.8 billion for the year, its worst result ever, weighed down by write-offs and the cost of a factory build-out that was not paying off. The board pushed out chief executive Pat Gelsinger in December 2024.

In March 2025, Intel brought in Lip-Bu Tan, a semiconductor veteran and prolific chip investor, to stop the bleeding. His plan was blunt: split Intel into two businesses, the designs and the factories, and make each stand on its own. He cut roughly 30,000 jobs, slowed the factory spending, and told the company to build only against demand it could actually see.

The turnaround also drew in Washington. In August 2025, the US government took a roughly 10% stake in Intel, an $8.9 billion investment funded largely by converting the company's unpaid CHIPS Act manufacturing grants into equity. For the federal government, keeping a domestic maker of leading-edge chips alive had become a matter of national security, not just industrial policy.

Intel now reports as two engines. Intel Products holds the chip designs that customers know: the Core PC processors and the Xeon server processors. Intel Foundry holds the factories. A small "All Other" bucket holds Mobileye and a few leftovers.

Only one of these engines makes money. Intel Products earned $12.7 billion in operating profit in 2025. Intel Foundry, though it books $17.8 billion in revenue, mostly makes Intel's own chips, and it lost money at a staggering rate. One quirk to note once: most of Foundry's revenue is internal, Intel's factories charging Intel's product teams, so about $17.7 billion is stripped out to reach the $52.9 billion the company actually reports.

*Foundry revenue is mostly internal. Around $17.7 billion of intersegment sales is eliminated to reach the $52.9 billion consolidated total.

This is the "Intel Inside" business, and it remains the company's cash cow. It sells the Core processors that go into laptops and desktops, bought by PC makers like Dell, HP, and Lenovo and, through them, by hundreds of millions of consumers.

It is comfortably Intel's most profitable segment, earning a roughly 29% operating margin on its sales. That single business threw off more operating profit than the rest of Intel combined, and it is what pays for everything else the company is trying to do.

The pressure is share, not profit. AMD's Ryzen chips have taken a growing slice of the PC market, and a newer threat has arrived from processors based on Arm designs, which trade raw power for battery life. Qualcomm now sells Arm-based Snapdragon chips for Windows laptops, and Apple's move to its own silicon showed the whole industry it could be done.

Intel's answer is Panther Lake, sold as the Core Ultra Series 3. It is the first high-volume chip built on Intel's new 18A manufacturing process, and it matters for more than its speed. If it sells well and yields well, it is the proof that Intel's factories are finally competitive again.

This segment sells Xeon server processors to cloud providers and large enterprises, plus Gaudi accelerators aimed at AI. After years of decline, it is the clearest sign of a recovery. Operating profit here more than doubled in 2025, and in the second quarter of 2026 the segment grew 59% from a year earlier.

The demand behind that jump is the same AI boom that hurt Intel elsewhere. As companies build AI systems, they need not just accelerator chips but also large numbers of conventional server CPUs to run them, and Intel still sells a lot of those.

The concession is real, though. Intel largely lost the market for AI training chips to Nvidia, whose accelerators became the default hardware for building large models. Gaudi, Intel's competing product, never gained meaningful traction. Tan has since reframed Intel's AI pitch around inference, the cheaper day-to-day running of AI models, and the CPUs that increasingly autonomous AI software still leans on.

Intel Foundry is the manufacturing arm, now run as its own business and, for the first time, opened up to make chips for outside customers. It is also the reason Intel nearly breaks even overall. The segment lost $10.3 billion in 2025, an improvement only in the sense that it lost $13.3 billion the year before.

The losses come from physics and economics. Leading-edge fabs are the most expensive factories humans build, and they only earn their keep when run near full capacity. Intel is spending heavily to catch TSMC on process technology, and until the newest factories are full, every one of them is a drain. The company has 18A in high-volume production now and is developing a more advanced node, 14A, next.

The whole bet rides on winning customers who are not Intel. So far it has design wins with Microsoft and Amazon, work for the US Department of Defense, and a deal in which Nvidia will co-develop chips that pair Intel processors with Nvidia graphics. What it does not yet have is a blockbuster anchor customer. Tan has said that 14A will only go ahead if outside customers commit to it, and has warned that Intel could stop making the most advanced chips itself if they do not.

The last bucket is small and easy to overlook. Its main asset is Mobileye, the self-driving car business Intel controls but lists as a separate public company, along with a few residual operations.

It brought in $3.6 billion in 2025 and turned a small profit. Mobileye sells the cameras, chips, and software that let cars see the road, and Intel keeps a majority stake while letting it trade on its own.

The more telling detail is what left this bucket. In September 2025, Intel sold 51% of Altera, its programmable-chip unit, part of Tan's push to shed anything that is not core and raise cash while doing it.

Intel spent more than it earned at the operating line in 2025, and the reason is the factories. The costs that matter here are physical: silicon, machines, and buildings, not salaries or advertising. That is what separates Intel from its fabless rivals, who carry almost none of this weight.

The largest cost by far is the cost of sales, which reached $34.5 billion, about 65% of revenue. This is what it takes to actually make chips: silicon wafers, raw materials, and the depreciation of fabs in Arizona, Ohio, Oregon, and Ireland.

It also includes the tools. The lithography machines from ASML that print the finest chip features cost hundreds of millions of dollars each, and the most advanced ones cost even more. Every fab needs several.

Because Intel owns its factories, any line running below full still has to be paid for, and that unused capacity crushes margins. A fabless competitor simply pays TSMC for the chips it needs and nothing more. This is the core reason Intel's gross margin has fallen from well above 60% in its prime to around 35% today.

R&D was Intel's second-biggest cost at $13.8 billion, though that is down sharply from $16.5 billion in 2024 as Tan trimmed spending. The money funds two things at once: the process technology behind nodes like 18A and 14A, and the design of the chips themselves.

The efficiency of that spending is a fair question. Intel spends roughly a quarter of its revenue on research, while Nvidia spends closer to a tenth of its own and earns far more. Spending more to earn less is the mark of a company still trying to buy its way back to the frontier.

Separate from its day-to-day costs, Intel spent about $14.6 billion on capital expenditure, chiefly new fabs. That is a large number, but a restrained one by Intel's recent standards, down from well over $20 billion as Tan imposed discipline on the build-out.

This is the spending the outside investors are effectively underwriting. It is partly offset by grants from the US CHIPS Act, which were designed to bring chipmaking back to American soil, and by deals that bring in co-investors to share the cost of specific factories.

Cutting roughly 30,000 jobs is not free. Intel booked $2.2 billion in restructuring charges in 2025, mostly severance, though even that was a relief after nearly $6.9 billion of such charges in 2024. Marketing, general, and administrative costs added another $4.6 billion.

Taken together, these lines tell the story of the turnaround. Intel is trying to shrink its way back to strength, spending real money now to lower its costs later.

Intel is unusual in that it fights on two fronts at once. As a chip designer it competes with AMD, Nvidia, and the wider Arm ecosystem. As a foundry it competes with TSMC. Almost all of these rivals are fabless or pure-play manufacturers, which means they do not carry the factory burden that is sinking Intel's margins.

Nvidia is the company that won the market Intel most wanted. Its accelerators power almost the entire AI build-out, giving it around 90% of that market and roughly $130 billion in annual revenue, while Intel's Gaudi never found buyers at scale.

The relationship is stranger than a simple rivalry, though. In 2025, Nvidia invested $5 billion in Intel and agreed to co-develop chips that fuse Intel processors with Nvidia graphics for PCs and data centers.

That makes Nvidia a competitor, an investor, and a potential foundry customer all at once. For Intel, the deal is validation and dependence in equal measure: a lifeline from the very company that beat it to the AI era. Whether the partnership deepens or stays symbolic is one of the more interesting open questions hanging over both firms.

TSMC is the benchmark Intel Foundry has to reach. The Taiwanese company earned around $122.5 billion in 2025 and holds close to 70% of the global foundry market, up from 64% a year earlier. It manufactures the chips for Apple, Nvidia, AMD, and Qualcomm.

Its advantage is trust as much as technology. Customers hand TSMC their most valuable designs knowing it will not compete with them, hit its yield targets, and ship on time. That reputation took decades to build.

Intel Foundry is asking those same customers to trust a company that also designs chips of its own, and that is still proving its newest factories work. Matching TSMC on cost, yield, and reliability is the entire justification for the $10 billion-a-year bet, and it is a steep climb.

AMD is the rival that has quietly done to Intel what Intel once did to everyone else. It designs x86 chips that compete directly with Intel's, in the same PCs and the same servers, and it has been taking share for years. In 2025 it grew revenue 34% to $34.6 billion, with data center sales alone reaching $16.6 billion.

The crucial difference is that AMD is fabless. It builds nothing itself, relying on TSMC for manufacturing, so it carries none of the factory costs weighing on Intel.

That lets AMD pour its money into design and pricing rather than plants. Its Ryzen chips press Intel in laptops, its EPYC server chips press it in data centers, and its Instinct accelerators give it a foothold in AI that Intel lacks. AMD is the clearest proof that a smaller, focused designer can outrun an integrated giant.

The most serious threat is not a single company but an architecture. Chips based on designs from Arm, long dominant in phones, are moving onto Intel's home turf by trading peak performance for far better energy efficiency.

The examples are piling up. Qualcomm sells Arm-based Snapdragon chips for Windows laptops, Apple has moved its entire Mac line to its own Arm-based silicon, and Amazon's Graviton processors now run a large share of its cloud. Arm itself makes no chips; it licenses the underlying instruction set and collects royalties, earning around $4 billion a year doing it.

For Intel, this is the slow erosion beneath everything else. Every laptop and every server that ships with an Arm-based chip is one that does not need x86, the design at the heart of Intel's business. Holding that ground while the factory turnaround plays out is one of the company's hardest tasks.

Intel's future rests on two bets, and the bigger one is the factories. Tan wants to turn Intel Foundry into a genuine contract manufacturer that makes chips for other companies, restoring advanced chipmaking on American soil in the process. The 18A process is now in high-volume production, and the more advanced 14A node is the next test. Tan has been clear that 14A will only proceed if outside customers commit to it, and has said Intel could exit leading-edge manufacturing altogether if they do not.

The second bet is the data center. Rather than chase Nvidia in AI training, Intel is aiming at the CPU demand that comes with running AI, especially inference and the more autonomous software that still leans on conventional processors. The 59% jump in that segment in early 2026 suggests the demand is real.

Underneath both bets is a discipline that was missing before. Tan has cut roughly 30,000 jobs, slashed factory spending, sold most of Altera, and kept only what he sees as core. The message to investors is that Intel will spend against demand it can prove, not demand it hopes for.

Then there are the outside investors themselves, which are both a support and a risk. The US government holds about 10% of Intel after an $8.9 billion investment, Nvidia put in $5 billion, and SoftBank added $2 billion. That capital keeps the factory build-out funded, but it also ties Intel's fortunes to Washington and blurs the line between a private company and an instrument of industrial policy.

The financial picture is finally pointing the right way. After the $18.8 billion loss in 2024, Intel nearly broke even in 2025, and in the second quarter of 2026 revenue grew 25% from a year earlier, its fastest growth since around 2011, alongside $7 billion in operating cash flow. Tan calls 2026 an execution year and points to 2027 as the moment the turnaround should show up in full.

Whether it gets there depends on two things Intel has to prove. The first is whether it can land a marquee outside customer for its 14A factories, the anchor that would turn Intel Foundry from a cash drain into a business. The second is whether it can hold its share of PCs and servers against AMD and the Arm ecosystem long enough for the factory bet to pay off. Both will be far clearer by 2027, and until then Intel is a company that makes real money designing chips, and spends nearly all of it trying to keep making them itself.