Intel

Intel Corporation is an American multinational semiconductor company headquartered in Santa Clara, California. Founded on 18 July 1968 by Robert Noyce and Gordon Moore, it is the inventor of the microprocessor and was for more than three decades the largest semiconductor manufacturer in the world by revenue. Its x86 processor architecture underlies the great majority of personal computers and servers in use.

Intel Corporation
Type	Public
Traded as	Nasdaq: INTC; Dow Jones Industrial Average component until 2024; S&P 500 component
Industry	Semiconductors
Founded	18 July 1968, Mountain View, California
Founder	Robert Noyce, Gordon Moore
Headquarters	Santa Clara, California, United States
Area served	Worldwide
Key people	Lip-Bu Tan (chief executive), Frank Yeary (chairman), David Zinsner (chief financial officer)
Products	Central processing units, graphics processors, chipsets, network and communications products, field-programmable gate arrays, foundry services
Revenue	Approximately US$53 billion (2024)
Employees	Approximately 100,000
Website	intel.com

Intel's history divides into three parts. It began as a memory company, inventing the dynamic random access memory chip and the erasable programmable read-only memory, and was driven out of memory by Japanese competition in 1985. It then became the microprocessor company, supplying the processors for the IBM Personal Computer and its successors and building, with Microsoft, the arrangement known as Wintel that dominated computing for twenty-five years. Since approximately 2015 it has lost its manufacturing lead, its position in mobile and in artificial intelligence accelerators, and a substantial part of its share in personal computers and servers to AMD, and has been engaged in an attempt to recover the first of these that is among the largest and most uncertain industrial undertakings in progress anywhere.

That attempt has drawn in the American state. The CHIPS and Science Act of 2022 was passed substantially with Intel in mind, and in August 2025 the United States government took an equity stake of approximately ten per cent in the company, an intervention without precedent in the modern American technology industry.

History

Founding and the memory business, 1968 to 1985

Robert Noyce and Gordon Moore left Fairchild Semiconductor in July 1968, the second such departure in the sequence that populated Silicon Valley with companies descended from William Shockley's laboratory. Noyce had co-invented the integrated circuit at Fairchild, independently of Jack Kilby at Texas Instruments, and had devised the planar process that made it manufacturable; Moore had written the 1965 article observing that the number of components on an integrated circuit had doubled annually and would continue to, a prediction that became Moore's law and functioned for six decades as the industry's timetable.

Arthur Rock raised US$2.5 million in a single afternoon on the strength of their names. Andrew Grove, a Hungarian refugee who had arrived in the United States in 1957 speaking little English, joined as the third employee and became the company's operational architect.

The name is a contraction of integrated electronics, chosen after the rights to Intelco were bought from a hotel chain.

Intel's first products were memory. The 3101 static RAM of 1969 was its first product; the 1103 dynamic RAM of 1970 was the first commercially successful semiconductor memory and displaced magnetic core memory within three years, and by 1972 it was the best-selling semiconductor device in the world. The erasable programmable read-only memory, invented by Dov Frohman in 1971, could be erased with ultraviolet light through a quartz window in the package and reprogrammed, which made it possible to develop software for embedded systems iteratively rather than committing a program permanently at manufacture.

The microprocessor

In 1969 the Japanese calculator company Busicom commissioned a set of custom chips for a printing calculator. Ted Hoff proposed instead a general-purpose programmable processor that could be configured by software for different products, Stanley Mazor developed the architecture, and Federico Faggin, who joined from Fairchild in 1970, designed and implemented it using silicon gate technology he had developed there.

The Intel 4004, delivered in 1971, contained 2,300 transistors on a chip measuring about three by four millimetres and performed roughly the work of the 1946 ENIAC, which had occupied a room. Faggin's initials are etched on the die. Intel bought back the exclusive rights from Busicom for US$60,000 when that company sought a price reduction, a transaction generally counted among the most profitable in commercial history.

The 8008 of 1972 and the 8080 of 1974 followed, the latter becoming the processor of the Altair 8800 and therefore of the first generation of personal computers, and the object for which Bill Gates and Paul Allen wrote the BASIC interpreter with which Microsoft began.

The Intel 8086 of 1978 introduced the x86 instruction set. Its cut-down variant the 8088, with an eight-bit external bus that allowed cheaper supporting components, was selected by IBM for the Personal Computer announced in August 1981. IBM required a second source, and Intel licensed the design to AMD under an arrangement whose subsequent interpretation produced two decades of litigation.

The selection was the most consequential commercial event in the company's history and was not regarded as such at the time. IBM expected to sell perhaps a quarter of a million machines; the open architecture produced an industry of compatible manufacturers, all of whom needed x86 processors, and the instruction set became a standard that has now survived for more than four decades through successive complete redesigns of the underlying hardware.

Leaving memory, 1984 to 1986

Japanese manufacturers, subsidised and organised through a national research programme, took the memory market during the early 1980s by producing chips of higher quality at lower cost. Intel's memory business lost money through 1984 and 1985 while consuming the greater part of its research spending.

Andrew Grove and Gordon Moore resolved the question in a conversation Grove described afterwards in Only the Paranoid Survive. He asked Moore what a new management brought in by the board would do; Moore replied that they would get out of memory; and Grove suggested that they walk out of the door, come back in and do it themselves. The exit was announced in October 1985, cost 7,200 jobs and made Intel a microprocessor company.

The decision is among the most frequently taught in business schools, both as an example of abandoning the business a company was founded on and as an illustration of the strategic inflection point, Grove's term for the moment when the fundamentals of a business change.

The x86 monopoly and Intel Inside, 1985 to 2000

The 80386 of 1985 was the decisive product. It was a 32-bit processor, it was fast, and Intel declined to license it to a second source, ending the arrangement that had obliged it to share the market. Compaq's decision to build a 386 machine before IBM did, in 1986, also ended IBM's control of the standard it had created; after that point the personal computer industry followed Intel and Microsoft rather than IBM.

The 486 of 1989 integrated a floating point unit and a cache. The Pentium of 1993 introduced a superscalar design executing more than one instruction per cycle, and the name was adopted because a number could not be trademarked and AMD and Cyrix had been selling processors named 386 and 486.

Intel Inside, launched in 1991, is among the most successful marketing programmes ever conducted and was unusual in being aimed at the buyers of a component they would never see. Intel paid computer manufacturers a rebate on processors in exchange for placing a sticker on the machine and the five-note audio signature in advertising, and within a few years a component supplier had achieved brand recognition comparable to the manufacturers of the products containing it. The rebates also became central to the antitrust findings against the company, because their terms depended on the proportion of a manufacturer's output using Intel processors.

The Pentium FDIV bug of 1994, a flaw in a lookup table that produced incorrect results for certain floating point divisions, became the company's first consumer crisis. Intel's initial position, that the error would affect a typical user once in 27,000 years and that it would replace chips only for those who could demonstrate a need, was received badly, and after IBM suspended shipments of Pentium machines the company offered unconditional replacement at a cost of US$475 million. Grove described the episode as the moment Intel discovered it was a consumer brand and was judged accordingly.

The company's revenue and profit grew through the 1990s at a rate few companies of its size have matched. It introduced the Celeron for low-cost machines in 1998, in response to AMD's success at the bottom of the market, and the Xeon for servers, which became an enormously profitable franchise as data centres displaced mainframes and minicomputers.

Grove, chief executive from 1987 to 1998, is generally regarded as among the most effective managers in American business history. His methods included constructive confrontation, a practice of open and aggressive argument about problems regardless of rank; the objectives and key results system, which Google and a large part of the technology industry subsequently adopted through John Doerr; and Copy Exactly, a manufacturing discipline requiring that a process proven in one factory be replicated in another down to details of no apparent significance, which allowed Intel to bring new fabrication plants to full yield far faster than competitors.

Itanium, the Pentium 4 and the AMD challenge, 2000 to 2006

Intel's attempt to replace x86 rather than extend it failed expensively. Itanium, developed with Hewlett-Packard from 1994 and released in 2001, used a very long instruction word architecture that depended on the compiler to identify parallelism, and it was late, slow at running existing software and extremely difficult to write compilers for. It was intended to supersede x86 in servers and to give Intel an architecture it did not license to anyone; instead AMD extended x86 to 64 bits with the Opteron in 2003, customers preferred an architecture that ran their existing software, and Intel was obliged to adopt AMD's extension, which it did as Intel 64. Itanium was finally discontinued in 2021 after a development cost estimated in the tens of billions of dollars.

The Pentium 4, introduced in 2000 with the NetBurst architecture, pursued clock frequency at the expense of work done per cycle, on the assumption that frequency would continue to rise. It did not, because power consumption rises faster than frequency and the heat became unmanageable, and the planned 10 GHz parts were never built. AMD's Athlon and Athlon 64 outperformed Intel's products for several years and took a substantial share of the market.

Intel's recovery came from its Israeli design centre at Haifa, which had developed the Pentium M for laptops from the older Pentium III architecture rather than NetBurst, and which produced far more performance per watt. The Core architecture derived from it, introduced in 2006, restored Intel's lead decisively and established performance per watt rather than frequency as the industry's measure.

The same team produced Centrino in 2003, a platform bundling processor, chipset and wireless networking, which did more than any other single product to make Wi-Fi standard in portable computers.

Tick-tock and the peak, 2006 to 2015

Paul Otellini, chief executive from 2005 to 2013 and the first non-engineer to hold the post, presided over the most commercially successful period in the company's history.

The tick-tock model, adopted in 2007, alternated a shrink of an existing design to a new manufacturing process, the tick, with a new microarchitecture on the established process, the tock, on a two-year cycle. It separated the two largest sources of risk in processor development and gave the industry a predictable cadence, and Intel executed it successfully through 65, 45, 32 and 22 nanometre processes. Its 22 nanometre process of 2011 introduced the FinFET, a three-dimensional transistor structure that Intel brought to production approximately three years before any competitor, and its manufacturing lead at that point was the largest it had ever been.

Intel also acquired McAfee in 2011 for US$7.7 billion, an attempt to build security into hardware that produced little and was substantially divested in 2016; and Altera in 2015 for US$16.7 billion, entering field-programmable gate arrays.

Its principal strategic failure of the period was mobile. Otellini later acknowledged publicly that Intel had declined to supply the processor for the original iPhone because the price Apple offered was below what Intel's cost forecasts justified, and that the forecast had been wrong by a very large margin. Its subsequent attempt to enter with the Atom processor, subsidised heavily through what the company called contra revenue, failed against the Arm designs that had been developed for low power from the beginning, and Intel abandoned the mobile processor business in 2016 having spent, by some estimates, more than US$10 billion.

The 10 nanometre failure, 2015 to 2021

Intel's decline began with a manufacturing process. The 14 nanometre node, due in 2014, arrived late and with poor yields. The 10 nanometre node, originally scheduled for 2015 and 2016, did not reach volume production until 2019 and did not reach acceptable yields on high-performance parts until 2021.

The cause was an attempt to achieve too large a density improvement in one step without using extreme ultraviolet lithography, which Intel had judged not ready. The process depended on multiple patterning with existing lithography, cobalt interconnects and contact over active gate, and the combination produced defect rates that could not be reduced. TSMC, which adopted extreme ultraviolet at 7 nanometres and moved in smaller increments, passed Intel in transistor density and performance during the same period, and Samsung followed.

The consequences accumulated. Intel shipped successive refreshes of the same 14 nanometre architecture, labelled by observers with increasing numbers of plus signs, from 2015 until 2021. AMD, which had separated its own manufacturing as GlobalFoundries in 2009 and subsequently moved to TSMC, introduced the Zen architecture in 2017 and, using TSMC's 7 nanometre process, produced processors that outperformed Intel's in servers and desktops for the first time in fifteen years, taking a very large share of the data centre market. Apple announced in 2020 that it would replace Intel processors in its computers with its own Arm designs, completing the transition in two years and demonstrating publicly that the alternative was faster and far more efficient.

Brian Krzanich, chief executive from 2013, resigned in June 2018 after the disclosure of a past relationship with an employee in breach of company policy. Bob Swan, the chief financial officer, succeeded him, the second non-engineer to lead the company, and the period is generally described as one in which the company's engineering problems were managed financially rather than solved.

Two further difficulties compounded the position. The Spectre and Meltdown vulnerabilities, disclosed in January 2018, exploited speculative execution, a technique used by essentially all high-performance processors, to read memory the program should not have access to. Meltdown affected Intel processors specifically and severely, and the operating system mitigations reduced performance on some workloads by double-digit percentages, a substantial loss to data centre operators. A sequence of further speculative execution vulnerabilities followed over the next six years, including Foreshadow, Zombieload, Downfall and Reptar, each requiring mitigation.

Gelsinger and IDM 2.0, 2021 to 2024

Pat Gelsinger, who had joined Intel at eighteen, had been its first chief technology officer, had led the 80486 design and had left in 2009 to run EMC and then VMware, returned as chief executive in February 2021 to considerable relief within the company and the industry.

His strategy, IDM 2.0, had three parts. Intel would continue to design and manufacture its own products; it would use external foundries including TSMC where appropriate, abandoning the principle that Intel products are made in Intel factories; and it would open its factories to external customers as Intel Foundry, competing with TSMC and Samsung for the business of designing companies including, potentially, its own competitors.

He also committed to five nodes in four years, a schedule taking the company from Intel 7, the renamed 10 nanometre process, through Intel 4, Intel 3, Intel 20A and Intel 18A by 2025, intended to restore process leadership. Intel 4 and Intel 3 introduced extreme ultraviolet lithography, which the company had previously deferred; 20A and 18A introduce gate-all-around transistors, which Intel calls RibbonFET, and backside power delivery, which routes power from beneath the wafer and which Intel called PowerVia and brought to production ahead of its competitors.

Very large capital commitments followed: new fabrication plants at Ohio, announced in 2022 at an initial US$20 billion and repeatedly delayed; expansion in Arizona, New Mexico and Oregon; a plant at Magdeburg in Germany, announced with a substantial German subsidy and subsequently suspended; and further sites in Ireland, Israel, Poland and Malaysia.

The CHIPS and Science Act, signed in August 2022, provided US$52.7 billion for American semiconductor manufacturing and was advocated for extensively by Gelsinger. Intel was awarded up to US$8.5 billion in direct funding and US$11 billion in loans in March 2024.

The financial position deteriorated nonetheless. The company reported a loss for 2024, suspended its dividend for the first time since 1992, announced the reduction of approximately 15,000 positions in August 2024, and was removed from the Dow Jones Industrial Average in November 2024 after twenty-five years, replaced by Nvidia. Its market capitalisation fell below that of companies it had dwarfed a decade earlier.

Two product difficulties added to the damage. The 13th and 14th generation Core desktop processors, known as Raptor Lake, developed instability in a substantial number of units from 2023, traced to a microcode fault causing excessive voltage requests that degraded the processors irreversibly; Intel extended warranties by two years and issued fixes, but the handling of the matter over more than a year damaged its standing among the enthusiast users who had remained loyal. Its Arc discrete graphics processors, introduced in 2022, arrived late and with poor drivers, and its Gaudi accelerators for artificial intelligence, acquired with Habana Labs in 2019 for US$2 billion, did not take meaningful share from Nvidia, whose position in that market Intel had failed entirely to anticipate.

Gelsinger was removed by the board in December 2024, three and a half years into a plan he had said would take five.

Lip-Bu Tan and state ownership, 2025 to 2026

Lip-Bu Tan, a Malaysian-born venture capitalist who had led the electronic design automation company Cadence Design Systems for thirteen years and had served briefly on Intel's board, became chief executive in March 2025.

He has reduced the company further, cutting headcount toward approximately 75,000 core employees, cancelling the German and Polish projects, slowing Ohio, flattening the management structure and requiring engineers to spend more of their time on engineering. He has stated that Intel will build 14A, the node after 18A, for external customers only if it secures committed volume, an acknowledgement that the foundry business cannot be sustained on internal demand alone.

The company has also been recapitalised from unusual sources. In August 2025 the United States government converted CHIPS Act grants into an equity stake of approximately ten per cent, making the federal government Intel's largest shareholder, an arrangement without modern precedent and criticised from both the left, as a subsidy to a failing firm, and the right, as industrial policy. SoftBank Group invested US$2 billion, and Nvidia invested US$5 billion in September 2025 alongside an agreement to co-develop products combining Intel's x86 processors with Nvidia's graphics and networking technology, a partnership between companies that had been the industry's principal rivals.

Intel completed the sale of a majority of Altera to Silver Lake in 2025 and has reduced its holding in Mobileye, the autonomous driving company it acquired in 2017 for US$15.3 billion and listed separately in 2022.

The question the company faces is whether 18A and 14A restore competitive manufacturing, and whether external customers can be persuaded to entrust their designs to a company that is also their competitor. Neither has been resolved.

Products

Client processors

Intel's consumer processors are sold as Core, a brand introduced in 2006 and restructured in 2023 into Core Ultra for the newer designs and Core for others, replacing the i3, i5, i7 and i9 numbering used since 2008. Pentium and Celeron served the lower end until they were retired in 2023 in favour of Intel Processor.

Recent generations have adopted a hybrid arrangement, introduced with Alder Lake in 2021, combining performance cores with efficiency cores on the same die, an approach taken from mobile processors and requiring operating system support to schedule work appropriately. Core Ultra processors from Meteor Lake in 2023 use a disaggregated design in which the compute, graphics, input-output and system-on-chip functions are separate silicon tiles bonded together, allowing each to be made on the process that suits it, including tiles fabricated by TSMC. They also include a neural processing unit for machine learning, the basis of the AI PC category Intel has promoted with Microsoft.

Data centre

Xeon processors serve servers and workstations and were for two decades among the most profitable products in the technology industry, at one point holding more than 95 per cent of the server processor market. AMD's EPYC has since taken a very substantial share, and Arm-based processors designed by Amazon Web Services, Google, Microsoft and Ampere have taken more, particularly for cloud providers' own workloads.

Intel's response has been to separate Xeon into performance-core products, sold as Xeon 6 P-core and formerly Granite Rapids, and efficiency-core products with very high core counts for cloud density workloads.

Gaudi accelerators, from the Habana Labs acquisition, compete with Nvidia's products for artificial intelligence training and inference and have not achieved significant share, and Intel has announced a successor architecture, Falcon Shores, combining graphics and accelerator designs.

Graphics and other products

Arc discrete graphics processors, launched in 2022, compete with Nvidia and AMD at the lower end of the market and have improved substantially as drivers have matured, though the business is small. Intel's integrated graphics, present in the majority of personal computers, give it the largest unit share of any graphics supplier.

Other products include chipsets, wireless and wired networking silicon, where Intel has long been the standard supplier of Ethernet and Wi-Fi controllers in personal computers; solid state drives, sold to SK Hynix in 2020 and 2025 in a transaction worth approximately US$9 billion; and Optane persistent memory, based on the 3D XPoint technology developed with Micron, which occupied a position between memory and storage, failed to find a market and was discontinued in 2022.

Mobileye, acquired in 2017, supplies camera-based driver assistance systems fitted to a very large proportion of the world's new vehicles and is developing autonomous driving systems. It was listed separately in 2022 with Intel retaining a majority.

Manufacturing and process technology

Intel is one of only three companies, with TSMC and Samsung, capable of manufacturing leading-edge logic semiconductors, and the only one headquartered in the West.

Its process naming was changed in 2021 to align with competitors' figures, since the nanometre numbers had long ceased to correspond to any physical dimension: the former 10 nanometre Enhanced SuperFin became Intel 7, and Intel 4, Intel 3, Intel 20A and Intel 18A follow. Intel 4 and 3 introduced extreme ultraviolet lithography using machines supplied by ASML, which are the most complex instruments manufactured in volume anywhere and of which ASML is the sole producer.

RibbonFET, Intel's implementation of the gate-all-around transistor, surrounds the channel with gate material on all sides rather than three, improving control of leakage at small dimensions. PowerVia delivers power from the reverse side of the wafer, separating power and signal routing and reducing interference and resistance; Intel brought it to production before its competitors, one of the few areas in which it has led in the past decade.

Its principal fabrication sites are in Oregon, where development is conducted at Ronler Acres; Arizona, where Fab 52 and 62 are being built; New Mexico, for advanced packaging; Ireland, at Leixlip; and Israel, at Kiryat Gat. Assembly and test operations are in Malaysia, Vietnam, China and Costa Rica.

Advanced packaging, combining several silicon tiles in a single package using Intel's EMIB and Foveros technologies, has become a principal area of competition as the benefits of shrinking transistors have diminished, and is one in which Intel's position is comparatively strong.

Graphics

Intel has attempted to enter discrete graphics three times over thirty years and has succeeded commercially in none of them, while simultaneously shipping more graphics processors than any other company because they are integrated into its processors.

The i740, released in 1998 with the Real3D division of Lockheed Martin, was uncompetitive and was withdrawn, though its architecture became the basis of the integrated graphics Intel has shipped ever since.

Larrabee, announced in 2008, proposed a graphics processor built from a large number of simplified x86 cores, which would have been programmable in ordinary software rather than through graphics-specific languages. It was cancelled as a graphics product in 2010 after failing to reach competitive performance, and the architecture became the Xeon Phi accelerator, which was itself discontinued in 2020. Its ambition was correct in one respect: the general-purpose programmability of graphics hardware became the basis of the artificial intelligence industry, which Nvidia rather than Intel captured.

Arc, developed under Raja Koduri, who joined from AMD in 2017, was announced in 2020 and released in 2022. The hardware was competent, the drivers at launch were not, particularly for older games, and the products reached a market in which Nvidia and AMD had established positions. Subsequent generations improved substantially and the business remains small, and Koduri left in 2023.

The graphics architecture nonetheless supports the Xe integrated graphics fitted to the majority of personal computers, the Intel Max data centre processors used in the Aurora supercomputer, and the media acceleration engines used in video transcoding, where Intel's hardware holds a substantial position in streaming infrastructure.

Automotive and Mobileye

Mobileye, founded in Jerusalem in 1999 by Amnon Shashua and Ziv Aviram, developed computer vision systems that interpret road scenes from a single camera. Its EyeQ system-on-chip is fitted to a very large number of vehicles, more than 175 million by the mid-2020s, supplying the lane departure, automatic emergency braking and adaptive cruise control that regulators in Europe and elsewhere now require.

Intel acquired the company in 2017 for US$15.3 billion, its largest acquisition, and listed a minority on the Nasdaq in October 2022 at a valuation considerably below the purchase price. Mobileye has continued to develop autonomous driving systems, operating test fleets in several cities and supplying its Chauffeur and SuperVision systems to manufacturers including Polestar, Zeekr and Volkswagen, while the timetable for fully autonomous consumer vehicles has extended repeatedly across the industry.

The company's approach has been distinctive in its reliance on cameras and on formally specified safety rules, the Responsibility-Sensitive Safety model Shashua proposed, which attempts to define mathematically what constitutes safe driving behaviour rather than learning it statistically, and which has been influential in the regulatory discussion even where it has not been adopted.

Philanthropy and public programmes

The Intel Foundation has funded science and mathematics education, and the company sponsored the Intel International Science and Engineering Fair, the largest pre-university science competition in the world, from 1997 to 2019, and the Intel Science Talent Search, which had begun as the Westinghouse competition in 1942 and which is often described as the junior Nobel Prize, from 1998 to 2016. Both sponsorships were withdrawn during periods of cost reduction and were taken over by other funders, and their loss attracted criticism disproportionate to the sums involved because of what the competitions represented.

The company announced a US$300 million diversity initiative in 2015 with the objective of achieving full representation of women and under-represented minorities in its American workforce by 2020, publishing detailed data annually against the target, and reported reaching it in 2018, two years early. The programme was among the most specific undertaken by any large company and was scaled back in subsequent restructurings.

Intel has also committed to responsible minerals sourcing, achieving conflict-free status for its microprocessors in 2014 and for its entire product range subsequently; to accessibility technology, including the speech system built for Stephen Hawking, which Intel developed, maintained and released as open source; and to supplier diversity spending targets.

Executives and notable people

Beyond the founders and chief executives, figures of consequence have included Ted Hoff, who proposed the microprocessor; Federico Faggin, who designed the 4004 and led silicon gate technology, and who subsequently founded Zilog and Synaptics; Stanley Mazor and Masatoshi Shima, who completed the 4004 team; Dov Frohman, inventor of the EPROM and founder of Intel's Israeli operations; Craig Barrett, who established the manufacturing discipline including Copy Exactly; Pat Gelsinger, who led the 80486 design at 25 and returned as chief executive thirty years later; Renée James, president from 2013 to 2015 and later chief executive of Ampere Computing, an Arm server processor company competing directly with Intel; Jim Keller, who designed AMD's K8 and Zen architectures and Apple's A4 and A5, worked at Intel from 2018 to 2020 and left over disagreement about the use of external manufacturing, a position the company subsequently adopted; and Ann Kelleher, who has led technology development through the recovery programme.

Arthur Rock, who financed the founding, and John Doerr, who worked as an Intel engineer before becoming a venture capitalist, carried the company's practices into a very large part of the industry that followed.

Intel in the artificial intelligence era

The rise of machine learning has been the most damaging technological change Intel has faced, because it shifted the workload that determines data centre purchasing from serial computation, at which its processors excel, to dense parallel arithmetic, at which they do not.

Its responses have been numerous and none has succeeded. Xeon Phi was discontinued; Nervana, acquired in 2016 for a reported US$400 million, was cancelled in favour of Habana Labs, acquired in 2019 for US$2 billion, whose Gaudi accelerators have taken negligible share; Movidius supplies low-power vision processors; the Arc and Max graphics processors address the market from a different direction; and the announced Falcon Shores design has been repositioned repeatedly.

The company's remaining arguments are that inference, the running of trained models, is a much larger long-term market than training and is more cost-sensitive and more amenable to processors already installed; that its Xeon processors host the accelerators in every data centre and benefit from their growth; that local execution on personal computers will require the neural processing units it is shipping; and that if Intel Foundry succeeds, it will manufacture for the companies whose designs have displaced its own.

Each is plausible and none has yet produced revenue comparable to what Nvidia earns, which by 2024 exceeded Intel's total revenue by a considerable margin from a business that did not exist in its current form a decade earlier.

Moore's law

Gordon Moore's observation, published in Electronics in April 1965 and revised in 1975, that the number of transistors on an integrated circuit doubles at a regular interval at minimum cost, is the most frequently cited generalisation in the history of technology and was, for most of its life, less a prediction than a plan. The semiconductor industry organised its research, its capital spending and its product roadmaps around the interval, and Intel in particular treated maintaining it as an obligation.

Its economic content, sometimes forgotten, concerns cost: the number of transistors that can be made for a given price. That relationship has weakened considerably. The cost per transistor has fallen slowly or not at all at the most recent nodes, the cost of a leading-edge fabrication plant now exceeds US$20 billion, and the number of companies able to build one has fallen from more than twenty in 2001 to three.

The industry's response has been to pursue density through three-dimensional stacking, through advanced packaging of separately manufactured tiles, and through specialisation, designing processors for particular workloads rather than improving general-purpose performance. Whether this constitutes a continuation of Moore's law or its replacement is a matter of definition, and Moore himself observed before his death in 2023 that no exponential is forever.

Processor history

Early microprocessors, 1971 to 1978

The 4004 (1971) was a 4-bit processor with 2,300 transistors intended for calculators. The 8008 (1972) was an 8-bit design commissioned by Computer Terminal Corporation, which declined to take it. The 8080 (1974) was the first microprocessor capable of serving as the basis of a general-purpose computer and powered the Altair 8800, the MITS machine whose January 1975 Popular Electronics cover began the personal computer industry. Federico Faggin left in 1974 to found Zilog, whose Z80, an enhanced and compatible 8080, outsold Intel's product and powered a very large number of machines including the Sinclair ZX Spectrum and the TRS-80.

The 8085 (1976) simplified the power supply requirements, and the 8048 microcontroller family found very large volume in embedded applications including, in a derivative form, the keyboard controller of the IBM PC.

The x86 line, 1978 to 1995

The 8086 (1978) established x86 with a 16-bit architecture and segmented memory addressing that permitted one megabyte of memory, a decision that constrained personal computing for a decade and produced the 640 kilobyte limit familiar to users of DOS. The 8088 (1979), with an 8-bit external bus, was chosen for the IBM PC.

The 80286 (1982) introduced protected mode, permitting memory protection and multitasking, in a form so awkward that Bill Gates described it as brain-dead. The 80386 (1985) was the first 32-bit x86 processor and introduced virtual 8086 mode, allowing multiple DOS programs to run simultaneously, and it is the processor on which the modern x86 architecture is built; code written for it still runs on current processors. The 80486 (1989) integrated the floating point unit and a cache and was the first x86 processor to execute one instruction per clock cycle for common operations.

The Pentium (1993) was superscalar, executing two instructions per cycle, and was the subject of the FDIV recall. The Pentium Pro (1995) introduced out-of-order execution and register renaming through the P6 microarchitecture, which was so successful that its descendants, through Pentium II, III, the Pentium M and Core, remained in production in modified form for more than a decade.

Netburst and Core, 2000 to 2015

The Pentium 4 (2000) used the NetBurst architecture with a very long pipeline designed for high clock frequency, reaching 3.8 GHz and no further. It introduced hyper-threading, presenting one physical core as two logical ones. The Pentium D added a second core inelegantly by placing two dies in one package.

Core (2006) and Core 2 descended from the Pentium M and restored Intel's lead. Nehalem (2008) integrated the memory controller, which AMD had done five years earlier, and introduced the Core i3, i5 and i7 naming. Sandy Bridge (2011) integrated graphics on the same die and is remembered as an unusually large generational improvement. Ivy Bridge (2012) introduced FinFET transistors at 22 nanometres. Haswell (2013) introduced the AVX2 vector instructions. Broadwell (2014) was the first 14 nanometre part and was delayed substantially, the first public sign of the manufacturing difficulty to come.

The 14 nanometre years and after, 2015 to 2026

Skylake (2015) was intended to be succeeded by a 10 nanometre design within two years. Instead it was followed by Kaby Lake, Coffee Lake, Whiskey Lake, Amber Lake, Comet Lake and Rocket Lake, all on refinements of 14 nanometres, over six years. Ice Lake (2019) and Tiger Lake (2020) brought 10 nanometre parts to mobile computers only.

Alder Lake (2021) introduced the hybrid arrangement of performance and efficiency cores and restored competitiveness against AMD. Raptor Lake (2022 and 2023) increased core counts and was the subject of the stability failures. Meteor Lake (2023) introduced the disaggregated tile design and the neural processing unit. Arrow Lake (2024) brought the arrangement to desktops, with mixed reception, and Lunar Lake (2024) prioritised efficiency in portable machines and was generally well received. Panther Lake, from 2025, is the first high-volume product on the Intel 18A process and is the test of the manufacturing recovery.

Marketing

The Intel Inside programme, begun in 1991 under Dennis Carter, was the first sustained attempt to build a consumer brand for a component. It rested on cooperative advertising funds: a manufacturer using Intel processors received a rebate that could be spent on advertising carrying the Intel Inside logo, which in practice meant that Intel paid for a substantial proportion of the personal computer industry's advertising.

The five-note audio signature, composed by Walter Werzowa in 1994, is among the most frequently heard pieces of music in the world, and the logo achieved recognition among consumers comparable to that of the computer manufacturers whose products carried it.

Its commercial effect was that consumers asked for Intel processors specifically, which gave the company pricing power over its customers and, critically, made it difficult for a manufacturer to switch to AMD without explaining to buyers why the sticker was missing. The same arrangement was found by competition authorities to have operated as an exclusionary payment when the rebate depended on the proportion of a manufacturer's output using Intel parts.

Intel's later campaigns, including the leap ahead positioning of 2006 and the AI PC promotion of 2024, have not had comparable effect, and the company reduced the Intel Inside programme's scope substantially in the 2010s.

Research and development

Intel Labs conducts research in circuits, architecture, interconnect, security, quantum computing and neuromorphic computing. Its Loihi neuromorphic processors, which simulate spiking neurons in silicon and consume very little power, are supplied to academic researchers and represent a line of enquiry distinct from the neural network accelerators the industry has otherwise pursued. Its quantum programme uses silicon spin qubits, which have the advantage of being manufacturable on the company's existing transistor processes, and is at an earlier stage than the superconducting and trapped-ion approaches of IBM and others.

The company has funded university research extensively, including through the Intel Science and Technology Centers, and sponsored the Intel International Science and Engineering Fair, the largest pre-university science competition in the world, from 1997 until 2019, and the Intel Science Talent Search, formerly the Westinghouse competition, from 1998 to 2016; its withdrawal from both was widely regretted and was among the more visible signs of the company's retrenchment.

Environment

Intel has committed to net-zero greenhouse gas emissions in its global operations by 2040 and to using only renewable electricity. Semiconductor manufacturing is unusually demanding environmentally: it consumes very large quantities of ultrapure water, uses perfluorinated compounds that are extremely potent greenhouse gases, and requires enormous and uninterruptible electrical supply.

The company reports restoring more fresh water than it consumes in several jurisdictions through watershed projects, has achieved high rates of waste diversion, and has been among the industry leaders in reporting. It has nonetheless been the subject of local objection over water use in Arizona and New Mexico, both water-stressed, and over emissions at its Oregon campus, and the growth of semiconductor manufacturing generally has made the industry's environmental footprint a matter of increasing attention.

Intel was also among the first companies to address conflict minerals, announcing in 2014 that its microprocessors were manufactured without tantalum, tin, tungsten or gold from sources funding armed groups in the Democratic Republic of the Congo, having traced its supply chain over several years, and the work informed the wider industry's approach.

Intel Foundry

Intel Foundry, established in its current form in 2021 and separated into a distinct operating and reporting unit in 2024, manufactures semiconductors designed by other companies. It is the most consequential and most uncertain element of the company's strategy.

The commercial logic is that leading-edge manufacturing requires volume. A fabrication plant costing US$20 billion must be run continuously at high utilisation to be economic, and Intel's own product volumes are no longer sufficient to fill the capacity that competing with TSMC requires. Selling capacity to others spreads the fixed cost, and a manufacturer with external customers has an incentive to serve them well that an internal-only operation does not.

The difficulties are substantial and were understood from the outset. A foundry must support customers' design tools, libraries and intellectual property, an ecosystem TSMC has built over thirty years and which Intel, having manufactured only its own designs, did not have; it must meet delivery commitments to customers who can go elsewhere, a discipline unfamiliar to an organisation whose internal customers could not; and it must persuade companies including AMD, Nvidia, Qualcomm and Apple to place their designs with a manufacturer that competes with them. Intel's response to the last of these has been to report the foundry separately, with its own profit and loss account and its own governance, and to state that its product divisions are treated as customers on the same terms as any other.

Announced customers have included Microsoft, for a custom chip on the 18A process, Amazon Web Services for an artificial intelligence fabric and a Xeon variant, and the United States Department of Defense under the Rapid Assured Microelectronics Prototypes programme. Intel has also agreed to manufacture on behalf of Arm-based designers and has partnered with UMC of Taiwan for mature nodes and with Tower Semiconductor, whose acquisition it abandoned in 2023 after failing to obtain Chinese approval.

The foundry reported losses in the billions of dollars annually through 2024 and 2025, which the company presents as investment in capacity that is not yet filled. Lip-Bu Tan's statement in 2025 that Intel would proceed with the 14A node for external customers only if committed volume were secured was the first public acknowledgement that the strategy might not continue.

Manufacturing

Semiconductor fabrication is conducted in cleanrooms in which particle counts are a minute fraction of those in an operating theatre, using several hundred process steps over approximately three months to build transistors and their interconnections layer by layer on a silicon wafer. The most advanced step, lithography, projects the circuit pattern onto light-sensitive resist using machines from ASML costing upwards of US$200 million each, and extreme ultraviolet systems generate light at 13.5 nanometres by firing a laser at droplets of molten tin fifty thousand times a second.

Yield, the proportion of dies on a wafer that function, determines profitability entirely, and a process that yields poorly at the start of production and improves over months is normal; one that does not improve, as Intel's 10 nanometre process did not, is ruinous.

Copy Exactly, introduced under Craig Barrett in the 1980s, requires that a process transferred from the development fab to a production fab be replicated in every detail, including details with no known effect, because the interactions in a process of several hundred steps are not fully understood and small differences produce large yield variations. The discipline allowed Intel to bring new factories to volume much faster than competitors and is among the practices the industry has copied.

Intel's development fab is at Ronler Acres in Hillsboro, Oregon, its largest site anywhere with more than 20,000 employees, and process technology is developed there before transfer. Volume production is conducted at Chandler and Ocotillo in Arizona, Rio Rancho in New Mexico, Leixlip in Ireland and Kiryat Gat in Israel, with assembly and test in Malaysia, Vietnam, China and Costa Rica. The Ohio site at New Albany, announced in 2022 as the largest single private investment in that state's history, has been delayed repeatedly and is now expected toward the end of the decade.

Acquisitions and divestments

Intel has made more than two hundred acquisitions. The largest have been Mobileye in 2017 for US$15.3 billion, Altera in 2015 for US$16.7 billion, McAfee in 2011 for US$7.7 billion, Habana Labs in 2019 for US$2 billion, Tower Semiconductor, agreed in 2022 for US$5.4 billion and abandoned in 2023, and the wireless modem business of Infineon in 2011.

Its record is generally regarded as poor. McAfee was substantially divested within five years. The modem business was sold to Apple in 2019 for US$1 billion after Intel had failed to produce a competitive 5G part, the immediate cause being Apple's settlement with Qualcomm, and Intel exited an adjacent business having spent far more than it recovered. Altera was substantially divested to Silver Lake in 2025. The memory and storage business was sold to SK Hynix. Mobileye was listed separately and Intel's holding reduced.

The pattern is of a company that acquired businesses adjacent to its own and disposed of them within a decade, generally at a loss, while the acquisitions it might have made, in graphics processors and in Arm designs, it did not.

Notable divestments and failures also include the attempted acquisition of Tower Semiconductor, blocked by the absence of Chinese regulatory approval, which would have given Intel an established foundry business and customers; and the withdrawal from smartphone modems, drones, wearables, smart glasses and several other consumer categories entered during the Krzanich period.

The Wintel arrangement

The combination of Intel processors and Microsoft Windows, named Wintel by the industry in the 1990s, was the most durable commercial arrangement in the history of computing, and neither company was contractually bound to the other.

Its mechanism was mutual reinforcement. Each new generation of Windows required more processing power, which sold Intel processors and obliged users to replace machines; each new Intel processor provided the performance that made new Windows features practicable; and the very large body of software written for the combination made it extremely costly for a customer to change either half. Manufacturers of personal computers were reduced to assembling components they did not control, with margins to match, a position from which several of them never escaped.

The arrangement began to loosen in the 2000s, as the internet reduced the importance of the operating system, and ended as a structural fact when Apple moved to its own processors, when Arm servers took share in cloud computing, and when Microsoft began supporting Arm in Windows and designing its own silicon. Microsoft's own position is now substantially independent of Intel, and the relationship between the two is commercial rather than symbiotic.

The period's significance is that it established, and then demonstrated the limits of, a business model in which control of a standard rather than of a product produces enduring profit.

Standards and platform work

Intel's influence on computing extends well beyond processors, because a processor manufacturer must ensure that the rest of the machine does not constrain it, and Intel repeatedly designed and gave away the interfaces that connected everything else.

USB, the Universal Serial Bus, was developed by a group led by Ajay Bhatt at Intel and published in 1996. It replaced the serial, parallel, PS/2, game and SCSI ports that had proliferated on personal computers with a single connector, supplied power as well as data, and permitted devices to be connected without restarting. It is now fitted to essentially every electronic device manufactured, and Intel licensed it without royalty because its purpose was to make computers easier to use rather than to earn fees.

PCI, introduced in 1992, and PCI Express, introduced in 2003, are the buses by which expansion cards, graphics processors, storage and network adapters connect to a processor. PCI Express remains the standard and has passed through successive generations doubling bandwidth, and its specification is now maintained by an industry group rather than by Intel, as PCI was.

Thunderbolt, developed with Apple and released in 2011, combines PCI Express and display data over one connector at very high bandwidth, and from its third version uses the USB-C connector; Intel contributed the specification to the USB Implementers Forum, so that USB4 is substantially Thunderbolt.

Other Intel specifications in general use include the ATX motherboard and power supply form factor of 1995, which determined the physical arrangement of desktop computers for thirty years; the AGP graphics bus; AC'97 and HD Audio; the Wireless MMX and later AVX vector instruction extensions; the Extensible Firmware Interface, which became UEFI and replaced the BIOS that had descended from IBM's 1981 design; and the promotion of Wi-Fi through Centrino, which did more to make wireless networking standard than the standard itself.

The pattern is consistent: Intel spent money to remove obstacles to the sale of processors and did not charge for the results, and the industry's physical and logical interfaces are consequently its work to a degree that is rarely noticed.

Software

Intel is a substantial software organisation, employing more software engineers than many companies whose only product is software. Its compilers have historically produced the fastest code for x86 processors and have been used as the benchmark reference, and were the subject of criticism and of an FTC settlement provision after it was established that they produced slower code paths when run on AMD processors regardless of the capabilities those processors reported.

oneAPI, introduced in 2019, is a programming model intended to allow the same code to run across processors, graphics processors, field-programmable gate arrays and accelerators, an explicit answer to Nvidia's CUDA, whose proprietary nature is that company's principal competitive protection. Its adoption has been limited, for the reason that a developer who has written code for CUDA has no reason to rewrite it.

Intel has been among the largest corporate contributors to the Linux kernel for two decades, has maintained graphics drivers as open source, has contributed to the development of the Yocto embedded Linux project, and released the Clear Linux distribution as a demonstration of performance tuning. It owned and later divested the Wind River embedded software company, and it develops OpenVINO for machine learning inference.

Intel Capital and investments

Intel Capital, founded in 1991, is among the oldest and largest corporate venture capital organisations, having invested more than US$20 billion in well over a thousand companies. Its purpose has been strategic as much as financial: investments were directed at companies whose products would increase demand for processors, including early investments in broadband access, video, wireless and web infrastructure, on the principle that anything requiring more computation was good for Intel.

Its returns have been substantial, and it has at times been among the most active investors in the world by number of transactions. Intel announced in 2025 that Intel Capital would be separated as an independent firm, part of the general reduction of activities not essential to manufacturing and processors.

Geopolitics and semiconductors

Intel is at the centre of a change in how governments regard semiconductor manufacturing. The recognition that the most advanced logic chips are made almost entirely in Taiwan, by TSMC, within range of Chinese military action, has produced industrial policy in the United States, the European Union, Japan, South Korea, India and China of a scale not attempted in any other industry.

The CHIPS and Science Act of 2022 provided US$52.7 billion in subsidy and a 25 per cent investment tax credit, and Intel was its largest intended beneficiary, having lobbied for it extensively. The European Chips Act followed with a comparable objective, and Japan funded both TSMC's Kumamoto plant and the Rapidus venture, in which Intel and IBM have technical involvement.

The United States has also restricted exports of advanced semiconductors and of the equipment to make them to China, from October 2022 and with successive extensions, which has reduced Intel's sales there, historically around a quarter of its revenue, and has accelerated Chinese efforts to build domestic capability. Intel's Dalian memory plant was sold to SK Hynix, and its Chinese test and assembly operations have continued while its ability to sell advanced products there has not.

The American government's acquisition of approximately ten per cent of Intel in August 2025 converted the relationship from subsidy to ownership. The transaction was defended on the ground that a country cannot be dependent on foreign manufacture for the components on which its economy and its defence rest, and criticised on the grounds that it rewards a company whose difficulties are self-inflicted, that state shareholding invites political direction of commercial decisions, and that the same objective could have been achieved by purchasing capacity rather than equity.

Competition

AMD is Intel's oldest competitor, founded in 1969 by another group from Fairchild, and has been variously a licensed second source, a litigant, a near-bankrupt and, since 2017, the principal threat. Its Zen architecture, designed under Jim Keller and manufactured by TSMC, took Intel's performance lead in 2019 and has taken roughly a quarter of the desktop market and a larger share of servers by revenue.

Nvidia was not a competitor at all for most of Intel's history and is now the most valuable semiconductor company in the world, because the parallel processors it built for graphics proved to be the right architecture for training neural networks and because CUDA locked the software ecosystem to its hardware. Intel's failure to anticipate this is the single largest strategic error of the past decade, and the two companies' 2025 partnership is an acknowledgement of the position.

TSMC does not design chips and is therefore Intel's competitor only in manufacturing, where it is dominant, holding roughly 90 per cent of the leading-edge foundry market and manufacturing for AMD, Nvidia, Apple, Qualcomm and, increasingly, for Intel itself.

Arm licenses an instruction set rather than making chips, and its designs, developed for low power, now run essentially all mobile devices, Apple's computers, and a growing proportion of cloud servers through Amazon's Graviton, Google's Axion, Microsoft's Cobalt and Nvidia's Grace. The migration of computing away from x86 is the most fundamental of the threats Intel faces, because it removes the architectural position on which everything else rested.

Others include Samsung in foundry and memory, Qualcomm in mobile and increasingly in personal computers, Apple, whose silicon is used only in its own products but which removed a substantial customer, and the Chinese manufacturers including SMIC, Loongson and Zhaoxin, constrained by export controls and advancing nonetheless.

Memory and storage

Intel's relationship with memory has been circular. It was founded as a memory company, left the business in 1985, returned to flash and solid state storage in the 2000s, and left again between 2020 and 2025.

The 1103 DRAM of 1970 displaced magnetic core memory and was for a period the best-selling semiconductor in the world. EPROM, invented by Dov Frohman in 1971, was erasable by ultraviolet light and was for many years among the company's most profitable products, since it allowed firmware to be developed iteratively. Flash memory was invented at Toshiba but Intel commercialised NOR flash on a large scale from 1988, supplying the memory in mobile telephones through the 1990s, and separated that business into Numonyx in 2008, subsequently sold to Micron.

Intel and Micron formed IM Flash Technologies in 2006 to produce NAND flash and later announced 3D XPoint in 2015, a non-volatile memory technology positioned between DRAM and flash in speed, density and cost, sold by Intel as Optane. Its performance was genuine and its position in the market was not: it was considerably more expensive than flash and slower than DRAM, its advantages required software written specifically for it, and the cost of manufacturing it at a scale that would have made it cheap could not be justified by the demand. Intel discontinued the business in 2022 at a write-off of several hundred million dollars, and Micron had already exited.

The NAND flash and solid state drive business, including the Dalian fabrication plant in China, was sold to SK Hynix in a transaction announced in 2020 and completed in stages through 2025 for approximately US$9 billion.

Networking and communications

Intel has supplied networking silicon since the 1980s and its Ethernet controllers have been the default in personal computers and servers for most of that period, to the extent that the driver names for its parts are familiar to anyone who has configured a server. Its Wi-Fi adapters, developed for Centrino and continued since, hold a similar position in portable computers.

Its ambitions in communications were larger and less successful. The company entered cellular modems through the acquisition of Infineon's wireless business in 2011, supplied modems for certain Apple iPhones from 2016 after Apple sought an alternative to Qualcomm, and withdrew in 2019, selling the business to Apple for US$1 billion after Apple and Qualcomm settled their litigation and Intel concluded that it could not produce a competitive 5G part. It also entered and left the markets for network processors, WiMAX, home gateway silicon and set-top box chips.

Its data centre networking business, based on the Tofino programmable switching silicon acquired with Barefoot Networks in 2019, was discontinued in 2023.

Supercomputing

Intel processors have powered a substantial proportion of the world's largest computers. ASCI Red, built at Sandia National Laboratories in 1996 with 9,298 Pentium Pro processors, was the first machine to exceed one teraflop and held the top position on the TOP500 list for three and a half years, longer than any other system. Its significance was that it demonstrated that very large numbers of ordinary commercial processors could outperform purpose-built supercomputer architectures, which ended the market for the latter.

Intel's later efforts have been mixed. The Xeon Phi many-core accelerators, derived from the cancelled Larrabee graphics project, powered the Tianhe-2 machine in China and were discontinued in 2020 having failed to compete with Nvidia's graphics processors for scientific computing.

Aurora, at Argonne National Laboratory, was contracted in 2015 for delivery in 2018 using Xeon Phi, redesigned after that product's cancellation, delayed repeatedly by the 7 nanometre process difficulties, and finally delivered in 2023 and 2024 using Xeon and Intel Max graphics processors. It exceeded one exaflop, becoming the second such machine after Frontier, and its history is a compact illustration of the company's decade.

Security technologies

Intel has built security features into its processors over three decades, with results that have been genuinely useful and, in several cases, undermined by the same speculative execution problems described above.

Software Guard Extensions, introduced in 2015, create enclaves whose memory is encrypted and inaccessible even to the operating system, intended to allow computation on data a cloud provider cannot read. The technology was adopted for confidential computing and digital rights management, including 4K playback on Windows, and was repeatedly compromised, by Foreshadow in 2018, by Plundervolt, by SGAxe and by others; Intel removed it from consumer processors in 2021 and retained it on server parts.

Trusted Execution Technology, the Management Engine, which is a small processor running independently of the main one and present in all Intel platforms, and Active Management Technology, which permits remote management of a machine even when powered off, have each attracted security criticism because they run code the owner of the machine cannot inspect and, in the Management Engine's case, cannot disable. A critical vulnerability in Active Management Technology disclosed in 2017 permitted remote takeover of affected systems, and researchers have described the Management Engine as the most privileged code on any personal computer.

vPro bundles these capabilities for corporate customers, and Intel has more recently added memory encryption and, on server parts, Trust Domain Extensions for confidential virtual machines.

Financial history

Intel's revenue grew from nothing in 1968 to approximately US$79 billion in 2021 and has since fallen to around US$53 billion. Its gross margin exceeded 60 per cent for most of the period from 1993 to 2021 and has fallen below 40 per cent, a change more damaging than the revenue decline because a manufacturer's profit is the difference between price and a cost base that is largely fixed.

The company's capital allocation in the decade to 2021 is the subject of continuing criticism. It repurchased more than US$100 billion of its own shares and paid substantial dividends during a period in which its principal competitor in manufacturing was building the capacity that would displace it, and it subsequently required government subsidy, asset sales and outside equity to fund the investment it had deferred. The counter-argument is that the manufacturing failure was technical rather than financial and that additional capital spending on a process that did not work would have made matters worse.

Its dividend, paid continuously from 1992, was cut in 2023 and suspended in 2024. Its shares, which had been a component of the Dow Jones Industrial Average since 1999, were removed in November 2024 and replaced by Nvidia, a substitution that summarised the decade more economically than any commentary.

Intel and Silicon Valley

Intel's founding is the central event in the formation of Silicon Valley as an industrial region rather than a place where some electronics companies happened to be.

William Shockley, co-inventor of the transistor, established Shockley Semiconductor Laboratory at Mountain View in 1956 and recruited an exceptional group of young researchers, whom he then managed so badly that eight of them left in 1957 to found Fairchild Semiconductor with financing arranged by Arthur Rock. Shockley called them the traitorous eight, a phrase that has survived him. Robert Noyce and Gordon Moore were among them.

Fairchild became the most important semiconductor company of the 1960s and the most prolific source of others: by one count more than sixty companies were founded by people who had worked there, and the term Fairchildren describes them. AMD, founded in 1969 by Jerry Sanders and seven colleagues from Fairchild, National Semiconductor under Charles Sporck, and Intel itself are the most significant.

The practices that characterise the region were established in this period: equity compensation extended to engineers rather than confined to founders, which Noyce insisted upon at Intel and which made technical staff owners; the acceptance of employees leaving to found competitors, assisted by California's refusal to enforce non-compete agreements; venture capital as an organised profession, which Arthur Rock and subsequently Kleiner Perkins and Sequoia developed around these companies; and the absence of the formal hierarchy and dress of eastern corporations, which Noyce practised deliberately.

Noyce himself, called the Mayor of Silicon Valley, was the model of the technical founder, and his influence on Steve Jobs and on a generation of entrepreneurs is documented extensively. He left Intel's day-to-day management in 1975, served as the first head of SEMATECH, the government-industry consortium formed in 1987 to restore American semiconductor manufacturing against Japanese competition, and died in 1990.

Chipsets and platforms

For most of its history Intel sold not a processor but a platform, comprising the processor, the chipset that connects it to memory, storage and peripherals, and the reference designs and validation work that allow a manufacturer to build a machine quickly.

The chipset business was itself very large and was strategically more important than its revenue suggested, because a competitor selling a processor without a chipset ecosystem could not reach the market. Intel's 440BX of 1998, the 815, 865 and the successive generations of the Z, H and B series determined what a personal computer could do, and the company used chipset features, including the support of particular memory types and the number of PCI Express lanes, to differentiate its processor segments.

Its attempts to use the position aggressively produced difficulties. The decision in 1999 to support only Rambus memory, in which Intel had an interest, in its highest-performance chipsets, against an industry that preferred the cheaper SDRAM, was abandoned after manufacturers and customers resisted and after a recall of the 820 chipset. Its refusal to license the front-side bus to Nvidia for chipset development, and the litigation that followed in 2009, ended that company's chipset business and contributed to its concentration on graphics, with consequences neither party anticipated.

The chipset has since been progressively absorbed into the processor package, and the platform business in its old form no longer exists.

Sites and geography

Intel's presence in particular places is disproportionate to its size. In Oregon, its Hillsboro campuses employ more than 20,000 people and constitute the largest industrial employment in the state; process development for every Intel node is conducted there, and the region is known locally as the Silicon Forest. In Arizona, the Chandler campus and the Ocotillo fabs are among the largest private employers, and the 2021 announcement of two further fabs was the largest private investment in the state's history until it was superseded by TSMC's.

In Israel, Intel is the largest private employer and the largest exporter, with design centres at Haifa and Petah Tikva and a fabrication plant at Kiryat Gat; the Haifa centre designed the Pentium M, the Core architecture, Centrino and several subsequent generations, and its contribution to the company's recovery in 2006 is difficult to overstate. In Ireland, the Leixlip campus has been in operation since 1989 and has been expanded repeatedly with extreme ultraviolet capability. In Costa Rica, Intel's assembly and test operation was for a period responsible for a very large share of that country's exports, and its partial closure in 2014 was a national economic event.

The company also operates design centres in India at Bangalore and Hyderabad, in Malaysia at Penang and Kulim, in China at Shanghai and Chengdu, in Poland at Gdansk, and in Germany at Munich, the last inherited with the Infineon wireless acquisition.

Personal computer market

Intel's fortunes are tied to a market that is mature and cyclical. Personal computer shipments peaked at approximately 365 million units in 2011, declined for most of the following decade, rose sharply during the pandemic as households and employers equipped people to work at home, and fell again as that demand was satisfied.

The company's share of processor units has remained the majority throughout, but the composition has changed against it: AMD has taken roughly a quarter of desktop and a larger share of high-performance segments; Apple has removed its own machines entirely; and Qualcomm's Arm-based Snapdragon X processors, introduced in 2024 with Microsoft's support, have established a further alternative, though their share remains small and the emulation of x86 software on them remains imperfect.

The AI PC category, promoted from 2024 by Intel, Qualcomm, AMD and Microsoft, proposes that machines with neural processing units will run artificial intelligence workloads locally rather than in the cloud, preserving privacy and reducing cost. Whether consumers will replace working machines for the capability is unproven, and the hardware has so far preceded the software that would justify it.

Developer and ecosystem work

Intel has maintained an unusually large investment in the software and tooling around its processors, on the reasoning that a processor is worth what the software written for it can extract from it.

Its compilers, performance libraries including the Math Kernel Library, VTune profiler and threading tools have been used across scientific and high-performance computing for decades. It has funded and staffed work on operating systems, virtualisation, compilers and language runtimes, and its engineers have been among the largest contributors to the Linux kernel and to the GCC and LLVM compilers.

The company has also run very large developer programmes, including the Intel Developer Forum, held from 1997 to 2016, which functioned as the industry's principal technical conference for platform architecture and whose discontinuation was read as a sign of the company's reduced centrality. Its successor events have been smaller and more product-focused.

Corporate affairs

Leadership

Intel has had nine chief executives. Robert Noyce led from 1968 to 1975, Gordon Moore from 1975 to 1987, Andrew Grove from 1987 to 1998, Craig Barrett from 1998 to 2005, Paul Otellini from 2005 to 2013, Brian Krzanich from 2013 to 2018, Bob Swan from 2019 to 2021, Pat Gelsinger from 2021 to 2024 and Lip-Bu Tan since March 2025.

The sequence divides between engineers and others, and the company's own diagnosis of its difficulties has generally attributed them to periods in which financial rather than technical judgement prevailed, an account that is self-serving and also broadly supported by the record: the decisions to defer extreme ultraviolet lithography, to prioritise margin over volume, to decline the iPhone contract and to repurchase shares rather than build capacity were each defensible financially and each proved damaging.

Grove's tenure is the most studied. His books, High Output Management and Only the Paranoid Survive, are among the most widely read in their field, and his practices, including the requirement that everyone including the chief executive work in an identical cubicle, were deliberate statements about the primacy of argument over rank.

Employees and culture

Intel employs approximately 100,000 people, reduced from a peak above 130,000, with the largest concentrations in Oregon, which is the company's largest site anywhere and the centre of its process development; Arizona; California; Israel, where the Haifa design centre produced the Pentium M, Core and several subsequent architectures and where Intel is the largest private employer and exporter; Ireland; India; Malaysia; and Vietnam.

Its culture under Grove emphasised data over hierarchy, disagreement conducted openly, and discipline in execution, and the Copy Exactly manufacturing methodology remains in use. Successive restructurings have strained the relationship between the company and its workforce considerably, and the reductions of 2024 and 2025 were conducted with an abruptness that drew comment even by the standards of the industry.

Finances

Intel reported revenue of approximately US$53 billion in 2024, against a peak of US$79 billion in 2021, with a substantial net loss. Its gross margin, historically above 60 per cent and at times above 65, fell below 40 per cent, a change that reflects both the loss of pricing power and the cost of underutilised factories, which for a manufacturer with very high fixed costs is the most damaging condition possible.

The company suspended its dividend in 2024 for the first time in more than three decades, having previously paid one continuously since 1992, and has funded its capital programme through debt, asset sales including Altera and the memory business, government grants, and the equity investments by the United States government, SoftBank and Nvidia.

Its capital expenditure requirement is the central financial problem. A leading-edge fabrication plant costs in the region of US$20 billion and requires several years to build and qualify, the spending must be committed before the demand it serves is known, and a manufacturer that falls behind must spend as much as the leader while earning less.

The semiconductor shortage, 2020 to 2023

The disruption that began with the COVID-19 pandemic produced the most severe semiconductor shortage in the industry's history and placed Intel in an unusual position, since the company was simultaneously unable to meet demand for its own products and publicly advocating that the world's manufacturing capacity was dangerously concentrated.

The shortage arose from a combination of causes. Vehicle manufacturers cancelled orders in early 2020 expecting a collapse in demand, and their capacity at foundries was reallocated to consumer electronics, for which demand rose sharply as people equipped themselves to work and study at home. When vehicle demand recovered the capacity was not available, and because a modern vehicle requires several hundred semiconductors and cannot be completed without any of them, production lines stopped worldwide. Several million vehicles were not built.

The components in shortest supply were not advanced processors but inexpensive microcontrollers and power management parts made on mature processes of 28 nanometres and above, in factories nobody had built for years because the margins did not justify it. The episode demonstrated that supply security is determined by the least glamorous part of the chain.

Intel's own difficulties were different. Its 10 nanometre capacity was constrained by yield rather than by equipment, and it was simultaneously short of the substrate materials on which packaged chips are mounted, a component made principally in Japan and Taiwan and in similarly short supply. The company used the period to argue for the CHIPS Act, for European subsidy and for the general proposition that leading-edge manufacturing should not be concentrated in one place, and the argument was accepted by governments across the developed world.

Process technology history

Intel's manufacturing cadence was for two decades the most reliable schedule in technology. Its nodes ran 180 nanometres in 1999, 130 in 2001, 90 in 2003, 65 in 2005, 45 in 2007, 32 in 2009, 22 in 2011 and 14 in 2014, approximately every two years, each delivering a density improvement of roughly double.

Several introduced techniques the industry then adopted. Strained silicon at 90 nanometres improved carrier mobility by stretching the crystal lattice. High-k metal gate at 45 nanometres replaced the silicon dioxide gate insulator, which had become so thin that electrons tunnelled through it, with hafnium-based materials, a change Gordon Moore described as the most significant in transistor technology since the late 1960s. The FinFET at 22 nanometres raised the channel into a fin surrounded by the gate on three sides, giving control that a planar transistor could no longer provide; Intel brought it to volume roughly three years before any competitor.

The sequence then broke. 14 nanometres was a year late. 10 nanometres, planned for 2016, targeted a density improvement of approximately 2.7 times rather than the customary 2, without extreme ultraviolet lithography, using self-aligned quadruple patterning, cobalt rather than copper in the lowest interconnect layers and contact over active gate. Each was individually defensible and the combination produced defect densities that would not fall. The company continued to insist publicly that the process was on track for several years after it was not, a communications posture that damaged its credibility when the position became undeniable.

The recovery programme announced in 2021 reset the naming, adopted extreme ultraviolet at Intel 4, and introduced RibbonFET and PowerVia at 20A and 18A. Intel 18A entered production in 2025 with the Panther Lake processors, and whether its yields and performance are competitive with TSMC's equivalent node is the question on which the company's independence rests.

Market position

Intel's share of x86 processor units remains the majority, in the region of 60 to 75 per cent depending on segment and quarter, having been above 80 per cent for most of the period from 2006 to 2019. Its share of the server processor market by revenue has fallen from above 95 per cent to roughly 60 per cent, with AMD taking most of the remainder and Arm designs taking a growing share of cloud providers' internal workloads, which do not appear in the conventional figures at all.

Its share of the total semiconductor market has fallen from first place, held for most of the period from 1992 to 2017, to fourth or fifth behind Nvidia, Samsung, TSMC by revenue measures and, at times, SK Hynix, a decline that reflects both its own difficulties and the extraordinary growth of memory and of accelerators.

The company retains assets that are difficult to replicate: an installed base of software written for x86 that is measured in decades of accumulated investment, relationships with every computer manufacturer and every large data centre operator, a patent portfolio that constrains what competitors can do, advanced packaging capability, and the only leading-edge logic manufacturing outside East Asia. Whether those are sufficient depends on execution that has not been demonstrated since 2014.

Antitrust and litigation

Intel's conduct toward AMD and toward its customers produced the largest competition penalties imposed on any technology company before the actions against Google.

The European Commission fined Intel €1.06 billion in May 2009, then the largest such penalty ever imposed, finding that it had paid manufacturers including Dell, Hewlett-Packard, NEC and Lenovo for using its processors exclusively or almost exclusively, and had paid the retailer Media Saturn not to stock AMD machines, and had paid manufacturers to delay or cancel AMD product launches. The General Court upheld the decision in 2014; the Court of Justice set it aside in 2017 for failure to examine Intel's economic evidence; the General Court annulled the fine in 2022; and the Commission reimposed a reduced penalty of €376 million in 2023 in respect of the naked restrictions, the payments to delay competitors' products, which had not been disturbed.

AMD's private action was settled in November 2009 for US$1.25 billion together with a new patent cross-licence. The Federal Trade Commission settled with Intel in 2010 on terms restricting its conduct, and the New York attorney general and a number of class actions produced further settlements. Japanese and Korean authorities made comparable findings.

The company has also been the subject of extensive patent litigation, most prominently with VLSI Technology, which obtained a US$2.18 billion jury award in Texas in 2021, subsequently overturned, and it has paid settlements over the disclosure practices surrounding the speculative execution vulnerabilities.

Security vulnerabilities

The disclosure of Meltdown and Spectre in January 2018 was among the most significant events in computer security. Both exploited speculative execution, in which a processor performs work before knowing whether it will be needed, and both allowed a program to infer the contents of memory it should not be able to read by measuring the timing effects of that speculative work on the cache.

Meltdown affected almost all Intel processors made since the mid-1990s and permitted reading of kernel memory from user space; it was mitigated by separating the kernel's page tables from the user's, at a performance cost that fell most heavily on workloads making many system calls, which is to say on servers. Spectre affected processors from every manufacturer and has proved considerably harder to eliminate.

The pattern continued. Foreshadow and the L1 Terminal Fault in 2018 affected Intel's secure enclave technology; Zombieload and the microarchitectural data sampling family in 2019; Downfall in 2023, affecting a vector instruction and requiring a mitigation with a substantial performance cost; and Reptar in the same year. Each required microcode updates and, in several cases, performance reductions, and the cumulative effect on the performance of Intel processors delivered before 2020 has been estimated at double-digit percentages for some workloads.

The episode changed the industry's understanding of processor design: performance techniques that had been considered purely architectural were shown to have security consequences, and the assumption that isolation between processes could be guaranteed by the hardware was undermined at a moment when the entire cloud computing industry depended on it.

Culture and management

Intel's management practices have been more widely copied than those of any technology company except Toyota in manufacturing, and almost all of them derive from Andrew Grove.

Constructive confrontation required that disagreements be argued openly and vigorously, with the expectation that the better-supported argument would prevail regardless of who advanced it. Meetings were conducted with a directness that visitors found aggressive, and the practice was taught formally in internal courses. Its purpose was to ensure that bad news reached decision-makers, which Grove regarded as the principal failure mode of large organisations.

Objectives and key results, which Grove developed from Peter Drucker's management by objectives, sets a small number of objectives each quarter with measurable results attached, published across the organisation. John Doerr, who had been an Intel engineer, introduced the system to Google in 1999 and subsequently to a large part of the technology industry.

Disagree and commit required that once a decision was taken, those who had argued against it execute it fully. Knowledge power versus position power distinguished authority derived from expertise from authority derived from rank and held that the former should prevail in technical questions. The late list, a sign-in sheet for employees arriving after eight in the morning, applied to everyone including Grove, and cubicles of identical size were used by every employee including the chief executive.

Grove's High Output Management (1983) and Only the Paranoid Survive (1996) remain in print and are among the most frequently recommended management books in the technology industry; the second introduced the strategic inflection point, the moment at which the assumptions underlying a business cease to hold, using Intel's exit from memory as the principal example.

The culture weakened as the company grew and as its position became secure, and successive chief executives have attempted to restore it. Gelsinger's return in 2021 was accompanied by an explicit appeal to Grove's practices, and Lip-Bu Tan's early actions, including the removal of management layers and the requirement that engineers spend a greater proportion of their time on engineering, were presented in similar terms.

Governance

Intel is incorporated in Delaware and listed on the Nasdaq, with no controlling shareholder until the United States government's acquisition of approximately ten per cent in 2025, which makes the federal government the largest holder.

The board's conduct has attracted criticism at several points: for the selection of chief executives from finance rather than engineering in 2013 and 2019; for the approval of very large share repurchases, exceeding US$100 billion in the decade to 2021, during a period in which the company was falling behind in manufacturing and would shortly require external capital; and for the removal of Gelsinger in December 2024, three and a half years into a plan the board had approved and which he had said would require five.

The arrangements attached to the government's shareholding are unusual. The stake is stated to be passive, without board representation or governance rights, and was created by converting CHIPS Act grants into equity at a discount to the market price. Commentators have observed that a passive government shareholding of that size in a company whose subsidies, export licences and defence contracts are determined by the same government is passive only in form.

Intel and Apple

The relationship between Intel and Apple is among the more instructive in the industry's history.

Apple used IBM and Motorola PowerPC processors until 2005, when Steve Jobs announced a transition to Intel, driven principally by IBM's failure to produce a PowerPC part suitable for portable computers. The Intel Macs that followed were successful, and for fifteen years Apple was among Intel's most prominent customers, its products used in Intel's own marketing.

Intel declined to supply the processor for the original iPhone. Paul Otellini stated afterwards that the price Apple would pay was below Intel's forecast cost and that the forecast had been wrong by a factor that made the decision the most consequential of his career, since the volumes Apple went on to achieve would have made the business enormously profitable and would have given Intel a position in mobile it never subsequently obtained.

Apple announced in June 2020 that it would replace Intel processors in the Mac with its own Arm-based designs, completing the transition within two years. The M1 and its successors, manufactured by TSMC, delivered performance and battery life that Intel's parts did not approach, and the comparison was made publicly and repeatedly by a company that had been Intel's customer. The transition removed a substantial and high-margin customer and, more damagingly, demonstrated to the market that the x86 architecture and Intel's manufacturing were not necessary to high-performance computing.

Intel's response included an advertising campaign in 2021 featuring the actor Justin Long, who had played the Mac in Apple's own earlier advertisements, a choice widely judged to have drawn attention to the comparison rather than away from it.

In popular culture

Intel's visibility as a component manufacturer is a consequence of its own marketing. The Intel Inside logo and its audio signature are recognised by very large numbers of people who could not describe what the company makes, and the sticker on a laptop is among the more successful pieces of industrial marketing ever conducted.

The company has featured in accounts of Silicon Valley's formation, including The Chip, Crystal Fire and Michael Malone's The Intel Trinity, and the founders' departure from Fairchild is the central episode in the genealogy that produced the region's companies, described by the term Fairchildren. Grove's own life, from Budapest under German and then Soviet occupation to the chief executive's office in Santa Clara, was set out in his memoir Swimming Across, and Time named him Man of the Year in 1997.

The Intel Museum at Santa Clara, open to the public, presents the history of the company and of semiconductor manufacture, and the company's role in the standard account of Moore's law has given it a position in popular understanding of technological progress that few manufacturers hold.

Brand architecture

Intel's product naming has changed repeatedly and has become a recurring subject of complaint among the people who buy its products.

The numeric progression from 8086 to 80486 ended when the courts held that a number could not be trademarked, which allowed AMD and Cyrix to sell processors named 386 and 486. Intel responded with Pentium in 1993, a coined name derived from the Greek for five with a Latin suffix, and extended it through Pentium Pro, II, III, 4, D, M and Dual-Core, by which point it signified nothing in particular.

Celeron from 1998 denoted low-cost parts and Xeon from the same year denoted server parts, both names retained for more than two decades. Core arrived in 2006 and the i3, i5, i7 hierarchy in 2008, with i9 added in 2017, alongside generation numbers, suffixes indicating power envelope and manufacturing variant, and a separate scheme for mobile parts. The result was model numbers that required a reference table to interpret, and comparisons between generations that were frequently misleading because a higher number did not reliably indicate better performance.

The 2023 restructuring replaced the scheme with Core Ultra for the newer designs and Core for others, with series numbers rather than generations, and retired Pentium and Celeron in favour of the undifferentiated Intel Processor. It has not obviously improved matters, and the difficulty is structural: a company selling several hundred variants of a product that differ in a dozen dimensions cannot name them in a way that is both precise and comprehensible.

Reception

Intel's standing among its own customers and among technically informed observers has moved further than its financial results. For most of the period from 1995 to 2015 its products were the default choice and its manufacturing was regarded as the best in the world; enthusiast communities argued about the merits of particular parts rather than about whether to buy Intel at all.

The change came through a sequence of specific failures that were visible to individuals rather than only to analysts: six years of processors that were only marginally faster than their predecessors; the performance losses imposed by the Spectre and Meltdown mitigations, which reduced the speed of machines people had already bought; the Raptor Lake instability, whose handling over more than a year was widely regarded as evasive; and the direct comparison with Apple's processors, which the company's former customer made in its own advertising.

The response to Pat Gelsinger's return in 2021 was correspondingly warm, and his removal in December 2024 was received by much of the technical community as evidence that the board did not understand the timescale on which semiconductor manufacturing operates. The subsequent investments by the United States government and by Nvidia, and the prospect of Intel manufacturing chips designed by the company that displaced it, have been treated as a statement about the industry's structure rather than about either company.

The counter-argument, made by analysts who are less sentimental, is that Intel earned very large profits for a very long time from a position it obtained by supplying a component to IBM in 1981, that it defended that position by means competition authorities repeatedly found unlawful, and that its difficulties are the ordinary consequence of a manufacturer losing its process lead rather than a tragedy.

Assessment

Intel is the company that made the microprocessor and organised the semiconductor industry around a schedule of improvement it largely set. Its products are in the great majority of the world's personal computers and, until recently, in almost all of its servers, and the interfaces it designed and gave away connect essentially all consumer electronics.

Its decline has a single proximate cause and several underlying ones. The proximate cause is the failure of the 10 nanometre process between 2015 and 2021, which cost the company the manufacturing lead it had held since the 1990s and on which every other advantage rested. The underlying causes are more contested: an integrated model that made the company slow to adopt external manufacturing when its own faltered; a margin discipline that deterred it from low-price volume businesses including mobile and graphics, both of which became very large; a board and management that preferred financial returns to capital investment during the period in which the investment was required; and an organisational confidence, acquired in two decades of dominance, that made bad news difficult to transmit, which is precisely the failure Grove had designed the culture to prevent.

Its recovery, if it occurs, will be attributed to the 18A and 14A processes and to whether external customers can be obtained for them, and the outcome will be known within a small number of years. The stakes extend beyond the company: it is the only manufacturer of leading-edge logic semiconductors headquartered outside East Asia, and a considerable part of American and European industrial policy is now committed to the proposition that this matters.

Timeline

Key dates include the founding on 18 July 1968; the 3101 memory in 1969 and the 1103 DRAM in 1970; the 4004 microprocessor in November 1971; the EPROM in 1971; the 8080 in 1974; the 8086 in 1978; the IBM PC contract in 1981; the exit from memory in October 1985 and the 80386 in the same year; Grove's appointment in 1987; Intel Inside in 1991; the Pentium in 1993 and the FDIV recall in 1994; the Pentium Pro in 1995; the Celeron and Xeon in 1998; the Pentium 4 in 2000; Itanium in 2001; Centrino in 2003; AMD's x86-64 and the adoption of it; the Core architecture in 2006 and the Apple transition in 2005; the European Commission fine and the AMD settlement in 2009; the Altera acquisition in 2015 and the first 14 nanometre delays; the abandonment of mobile in 2016; the Mobileye acquisition in 2017; Spectre and Meltdown in January 2018 and Krzanich's resignation in June; 10 nanometre volume production in 2019; Apple's announcement of the Arm transition in June 2020; Gelsinger's return in February 2021 and the IDM 2.0 strategy; the CHIPS Act in August 2022; the Mobileye listing in October 2022; the Tower acquisition abandoned in 2023; the foundry separation in 2024; the removal from the Dow and the 15,000 redundancies in 2024; Gelsinger's departure in December 2024; Lip-Bu Tan's appointment in March 2025; and the United States government's equity investment and the Nvidia and SoftBank investments in August and September 2025.

See also

• AMD
• TSMC
• Nvidia
• Gordon Moore
• Andrew Grove
• Moore's law