Personal computer

A personal computer (PC) is a general-purpose computer designed for use by one person at a time, usually at a desk or on a lap, and able to run a wide range of software without specialised operator training. Its core is a microprocessor attached to memory, storage, input and output devices, and an operating system. The entry treats the PC as both a piece of engineering and a social and economic phenomenon, and it explains how the term covers a family of machines rather than one product.

The PC matters because it moved computing out of institutional data centres and into homes, schools, offices and small businesses. Its design choices, especially the open hardware standards of the 1980s and the graphical interface that followed, shaped the modern internet economy, the software industry and the way ordinary people create and share information. Industry estimates put annual global shipments in the range of roughly 270 to 280 million units in the mid-2020s, although figures vary considerably between research firms, a point the entry should flag.

The article follows the PC from its roots in mainframes, time-sharing and minicomputers, through the microcomputer boom and the IBM-compatible standard, to the graphical interface, hardware architecture, software ecosystem, device forms and processor competition of later decades. It then covers networking, industry structure, social uses, security, repairability and environmental impact, the rise of AI-capable PCs, and the unresolved debates about the platform's future.

Definition and scope

A personal computer (PC) is a general-purpose computer meant to be operated by one person at a time. Reference works usually define it by use rather than by hardware: its size, capability and original price suit an individual, and the user runs it directly with no operator mediating access. This contrasts with the batch processing or time-sharing models that let larger, more expensive minicomputer and mainframe systems serve many people, usually at the same time. The entry therefore uses three criteria together: a single-user purpose, general-purpose programmability that lets the machine run many kinds of software, and standard components that can be sourced from multiple suppliers. The microprocessor is the component that made this combination affordable, but a processor alone does not make a PC.

Distinguishing PCs from other machines

The main contrast is with mainframes and minicomputers, which were shared among users through batch scheduling or time-sharing. Servers are also multi-user by design, since they answer requests from many clients over a network, and workstations are high-end single-user machines usually aimed at engineering, design or scientific work; the line between a powerful PC and a workstation is a matter of marketing and specification rather than kind. Smartphones and mobile phones are programmable, single-user devices too. What usually places them outside the PC category is their form factor, their mobile operating systems and the controlled way software reaches them, not their basic computing capability.

Three meanings of "PC"

The phrase has three overlapping uses. As a category, it names the whole class of general-purpose single-user machines, including desktops, laptops and small-form-factor systems. As a brand term, it has been attached to specific products: the IBM Personal Computer of 1981 gave the name to a platform that clones later copied. As a colloquial term, "PC" often means a Windows machine, set against a Mac. The abbreviation began as shorthand for "personal computer" but later came to mean mainly the Intel-and-Windows platform. The dictionary record of the longer phrase is older still: Merriam-Webster gives 1959 as the first known use of "personal computer."

Blurred boundaries and a working definition

The boundaries are least clear for portable and hybrid devices. A hybrid laptop is one that can be used as either a laptop or a standalone tablet, and industry usage tends to call the same class 2-in-1s. Chromebooks use Google's ChromeOS instead of Windows or macOS and are optimized for online apps. Whether they count as PCs depends on whether one emphasizes the hardware form, the operating system or the application model. Statistical practice adds further ambiguity: some research firms have included tablets in their PC totals, so published unit figures are not always comparable with one another.

This entry adopts a working definition to keep the discussion consistent. A PC is a general-purpose, single-user computer built from standard components that runs a full general-purpose operating system. On that basis, desktops, notebooks, Chromebooks and 2-in-1 hybrids are included, while phones are treated as a neighbouring category. Tablets are discussed when they overlap with PC use or when they compete directly for the same roles. Where a source uses a narrower or broader meaning, the entry notes it.

Precursors: mainframes, time-sharing and minicomputers

In the 1950s and early 1960s, computing meant mainframes: large, expensive machines that organisations shared among many users. While mainframes cost up to $1,000,000, minicomputers cost well under $100,000. Because access was so scarce, most work ran in batch mode. Programmers prepared jobs on punched cards or tape, operators fed them to the machine in queues, and results came back later, sometimes after hours or days. The model rewarded throughput and institutional scheduling, not the convenience of an individual user, and it set the expectations that later systems would try to overturn.

The first serious challenge was time-sharing. The idea was that a processor could switch rapidly among many users, giving each person the impression of a dedicated machine. At MIT, Fernando Corbató's group built the Compatible Time-Sharing System (CTSS). The increasing number of users needing access to computers in the early 1960s leads to experiments in timesharing computer systems. An experimental version was shown in November 1961 on a modified IBM 709, supporting several terminals at once. CTSS was later moved to the IBM 7090 and 7094, and routine service for MIT users began in 1963. It was designed to run interactive and batch work side by side, and it was operating system software in the full sense. Its text formatter and inter-user messaging, which presaged instant messaging and electronic mail, showed that people would use shared machines for personal work, not only for calculation.

Time-sharing still depended on expensive central hardware, so a second strand of development concerned smaller machines. Digital Equipment Corporation (DEC), founded in 1957 by Ken Olsen and Harlan Anderson, sold transistorized modules before building complete computers. Its PDP-1, introduced around 1960, was used in early experiments with interactive systems, partly because it was cheaper and more flexible than a mainframe. Its successor, the PDP-8, appeared in 1965 at a price of about $18,500, and it became the template for the minicomputer class. Launched in 1965, it was the first minicomputer to sell for under $20,000. Its low cost let laboratories, universities and original equipment manufacturers embed computers in instruments and systems, creating a market for dedicated machines that were no longer run by a central computing department. Prices in secondary sources vary for the early PDP models, so the figures above should be treated as approximate.

The most direct precursor of the personal computer was the Xerox Alto, developed at Xerox PARC from 1972 and released in 1973. It was an experimental single-user machine that combined a bitmapped display, a mouse, overlapping windows, local networking over Ethernet and page-oriented editing. Introductory price: US$32,000, and Xerox built only about 2,000 units. It made two thousand Altos, mainly for use in Xerox offices or affiliated institutions, but it didn't market the Alto as a consumer product. Its graphical ideas had roots in Douglas Engelbart's work at SRI, including a 1968 demonstration of a mouse and windowed display, but the Alto was the first to make them a practical single-user workstation.

These machines rested on a set of ideas about how people should use computers. The first was interactive computing: a person at a terminal should converse with the machine in real time, rather than hand over a job and wait. The second was the notion of a machine that an individual could own and control, which PARC researchers developed in terms of children and non-specialists. Alan Kay's 1972 Dynabook concept, a portable device for personal creative work, is the clearest statement of this vision. Those ideas were ready well before the hardware could deliver them at a price individuals could pay. The gap was closed by the integrated circuit and, above all, the microprocessor, which is the subject of the next section.

The microprocessor and the hobbyist microcomputer (1970s)

Before the 1970s, a computer's central processor was usually built from many separate logic chips. The microprocessor changed that by putting a complete CPU on a single chip. Intel's 4004, released in November 1971 and developed for the Japanese calculator maker Busicom, contained about 2,300 transistors and ran at 740 kHz. Its 4-bit word length was adequate for calculators but limiting for general computing, so 8-bit designs followed as manufacturing improved. Cost was the other constraint. Fewer defective chips per wafer cut the cost of complex designs sharply, which made a chip sold for $25 possible.

Intel's 8080, released in April 1974, ran at about 2 MHz and needed only a single 5-volt supply. It became the core of early hobbyist machines and set a de facto instruction-set standard. MOS Technology, founded by former Motorola designers, introduced the 6502 in 1975 at a price below one-sixth of rival chips such as the Motorola 6800 and the 8080, and that price pressure pushed down prices across the processor market. Zilog, founded by Federico Faggin after he left Intel, released the Z80 in 1976. Rather than undercut its former employer, Zilog aimed for a more capable chip, and the Z80 stayed software-compatible with the 8080, so existing programs carried over. The 6502 and Z80 families went on to power many of the home computers of the late 1970s and early 1980s.

The machine most often called the first commercially successful personal computer was the MITS Altair 8800, which appeared on the January 1975 cover of Popular Electronics. It was sold as a kit for roughly $400 to $450, depending on the source. It used an 8080, started with 256 bytes of memory that could be expanded to 64 KB, and had no keyboard or screen. Users entered programs with front-panel switches and read results from lights. Despite these limits, demand was strong. Two programmers, Bill Gates and Paul Allen, wrote a BASIC interpreter for it without ever having an Altair. They debugged the code on a time-sharing DEC PDP-10 using a simulator. Their company, then called Micro-Soft, survived, while MITS did not. Early on, software proved to be what made a personal computer useful to non-specialists, and this lesson recurred throughout the decade.

The Homebrew Computer Club, formed in Silicon Valley in 1975, gathered hobbyists, engineers and would-be entrepreneurs who swapped designs and parts. Steve Wozniak, who had been attending its meetings since March 1975, designed the Apple I for his own use. He demonstrated the prototype at the club in 1976. The finished single-board machine sold for $666.66 and included 4 KB of RAM, but buyers still had to supply a power supply, keyboard, display and storage. Apple Computer was founded as a partnership on April 1, 1976, and its first production was financed by Wozniak selling his HP-65 calculator and Jobs selling his Volkswagen microbus. Other firms also moved beyond bare boards. Processor Technology's Sol-20 was among the first machines with a keyboard built into the case, a feature many later home computers copied.

The shift from kits to assembled machines came quickly. Hobbyists who were willing to solder were a shrinking share of buyers, and retailers such as the Byte Shop sold both assembled and kit computers. The Apple II, the Commodore PET and the TRS-80 arrived in 1977 as complete, cased systems with keyboards. Commodore's $795 PET 2001 was shown at the January 1977 Consumer Electronics Show, but it did not ship in volume until December. These three machines, sometimes called the "1977 trinity," took over the market. Processor Technology went bankrupt after a run of failed products. Once the market was proven, almost every consumer electronics maker rushed to enter it. Several firms with no earlier computer business, including Mattel, Coleco, Texas Instruments and Timex, launched home computer lines that were short-lived.

The IBM PC and the compatible standard

IBM entered the microcomputer market late. By the time its machine appeared, several companies were already selling small computers, but IBM's own history argues that these earlier machines offered too little software to attract ordinary buyers. The IBM 5150, introduced on 12 August 1981, arrived about a decade after the first commercial microcomputers, yet IBM's reputation gave buyers confidence that personal computers would be here to stay. The base model was built around the Intel 8088, a chip that combined a 16-bit internal architecture with an 8-bit external data bus, which helped keep costs down while delivering strong performance for its time. IBM priced the machine at roughly US$1,500, and the operating system came from outside the company: the machine ran PC-DOS, the IBM-branded version of Microsoft's MS-DOS.

An open design built from outside parts

The feature that mattered most for the industry's future was IBM's choice to depart from its usual practice. Its team sourced much of the hardware from outside component suppliers and designed the machine to run third-party operating systems and software. IBM also published a technical reference covering the circuit designs and source code, so that companies could develop software and peripherals for it. The result was rapid ecosystem growth. Within a year of launch, more than 750 software packages were available for the IBM PC. Expansion cards were also soon widely available, since the machine used standard slots that third parties could fill. The original 8-bit bus was later standardized as ISA, and when IBM moved to the 16-bit AT, early IBM PC compatibles kept the same bus designs as their IBM counterparts, changing to the 16-bit IBM AT bus with the release of the AT. This plug-compatibility meant that a card or program built for one machine could generally work in another, which lowered the risk for both buyers and suppliers.

The BIOS problem and clean-room cloning

The one component IBM kept proprietary was the BIOS, the firmware that starts the machine and provides standard hardware services to the operating system. The BIOS firmware was originally proprietary to the IBM PC, and it was reverse-engineered by several companies, including Compaq, Phoenix Technologies and AMI, to create compatible systems. Copying the code directly would have been copyright infringement, so the approach that worked was clean-room design. Engineers who had studied IBM's listings were kept apart from the team writing the replacement, which was built from documented behaviour rather than IBM's source. Because the programmers who wrote the Phoenix code never read IBM's reference manuals, nothing they wrote could have been copied from IBM's code, however closely the two matched.

Columbia Data Products and Eagle Computer were among the first to release clones in 1982, but IBM sued several rivals for copying the BIOS, and some settled by agreeing to re-implement it. Compaq's approach proved decisive. Its portable, announced in November 1982, licensed MS-DOS from Microsoft and legally reverse-engineered IBM's BIOS software. IBM did not successfully challenge it, and the lawful route it demonstrated was then taken up by others. Phoenix Technologies, for example, created the first commercially available IBM PC compatible ROM BIOS during 1983 and the first part of 1984, and licensed it to other manufacturers, which made it far easier for new firms to build compatible machines without developing their own firmware from scratch.

Consequences for pricing, competition and control

The compatible standard turned the PC into a commodity platform. Because the hardware was assembled from widely available parts and the software ran on any machine with a matching BIOS, manufacturers competed on price, reliability and distribution rather than on proprietary technology. Buyers gained choice and downward price pressure, while IBM's own margins came under strain. The two companies that became the platform's key suppliers were Intel, whose x86 processors remained the central component, and Microsoft, which kept the right to license MS-DOS to other manufacturers. The partnership that emerged between them is often called "Wintel."

IBM's control of the architecture did not last. From 1981 until 1987, IBM controlled the PC architecture. Its PS/2 line, introduced in 1987, did not conform to the PC architecture that IBM itself had created, and IBM thereby lost control of the PC architecture. The clone market had by then made it impossible for IBM to set the rules alone. Even so, the design lineage persisted: PC-architecture CPUs have instruction sets derived from the Intel 8088, the IBM PC's original processor. The 1981 decision to publish specifications and rely on outside suppliers thus created a standard that outlived IBM's ownership of it.

Graphical interfaces and the mouse

Research origins at Xerox PARC

The pointing-and-windows model of interaction was developed well before personal computers were sold. Douglas Engelbart's NLS system, demonstrated in 1968, already combined overlapping windows, hypertext and mouse control, and parts of it were later carried into the Alto. Xerox opened its Palo Alto Research Center in 1970, and it became the place where the graphical user interface, the mouse, the WYSIWYG editor and the laser printer were brought together in working form. The Alto, begun in late 1972 and first running in April 1973, used a bit-mapped display of about 606 by 808 pixels and was controlled with a mouse. It was never sold commercially, but more than 1,200 units were distributed inside Xerox and to outside research sites for testing. Its mouse began as a single-button device at SRI and was reworked at PARC into a three-button design; user testing later showed that three buttons confused people, and that a single button was also a poor choice because it made double-clicking necessary for basic actions.

Xerox's commercial attempt followed with the Star in 1981, priced at $16,595 and aimed at office workers. In 1979 Steve Jobs toured PARC under an arrangement linked to Xerox's investment in Apple, and the ideas he saw shaped Apple's later products. Although the Alto's GUI was technically ahead of its time, it took roughly two decades for graphical interfaces to become the norm on desktop machines.

Apple Lisa and the Macintosh

Apple released the Lisa on January 19, 1983, at a price of $9,995, aimed at business users. It offered a mouse, icons, overlapping windows, menus, drag and drop, copy and paste, and a trash can, but its 5 MHz processor, unreliable floppy drives and high price limited sales, and it sold only around 10,000 units. Apple cut prices and introduced the Lisa 2 models in 1984, but the Lisa never reached a mass market.

The Macintosh carried the same basic ideas into a far cheaper machine. It was promoted with a Super Bowl commercial on January 22, 1984, and went on sale on January 24 at $2,495. It had a 9-inch monochrome display of 512 by 342 pixels and lacked the Lisa's hard drive, multitasking and advanced memory, yet it made the mouse-driven desktop accessible to ordinary buyers. Its success relative to the Lisa is usually attributed to price and marketing as much as to technical merit.

Windows and the desktop metaphor

Microsoft's graphical environment, Windows, arrived in 1985 as a layer over MS-DOS and was followed by versions 2.0 (December 1987), 3.0 (May 1990, adding virtual memory and multitasking) and 3.1 (April 1992). Each release strengthened the same vocabulary of desktop, folders, files, windows and icons. Because the IBM PC and its clones ran DOS-based software, Windows allowed users of existing hardware to adopt the new metaphor without changing machines, and this helped turn the desktop model into a shared convention rather than a single company's design.

Usability was the central argument. Command-line interfaces demanded memorised syntax, and typing errors could be costly; a graphical desktop let users recognise options on screen. The change also widened the audience for PCs beyond specialists, and it made the look and layout of an application a selling point in its own right. Software vendors such as Microsoft, which bundled Word, Excel and PowerPoint into Office in 1990, gained an advantage from applications that shared menus, keyboard shortcuts and window behaviour.

Legal disputes and competition

In March 1988 Apple sued Microsoft and Hewlett-Packard, alleging that the Windows 2.03 and HP NewWave displays infringed copyrights in the Macintosh visual displays; Apple listed 189 contested displays. A 1985 licence had allowed Microsoft to use Macintosh interface elements, and the district court ruled that 179 of the displays were covered by that licence. Most of the remaining ones were held not to infringe, because the court treated them as ideas or functional elements that copyright does not protect. The Ninth Circuit affirmed in 1994. Apple's claims largely failed, though it kept a narrow win concerning the trash can and folder icons from NewWave. Xerox also sued Apple, but those claims were dismissed without a ruling on whether the Mac's interface copied Xerox's work.

The outcome limited the protection of "look and feel" and established that a court must dissect an interface into its separate elements, rather than treat it as a single whole. Because general concepts such as windows, icons and menus could not be owned, rival systems were free to adopt them, and the desktop metaphor became an industry norm. The dispute also strained relations between Apple and Microsoft until a 1997 settlement, in which Microsoft invested in Apple; reports give different amounts for the investment. Concern over similar suits also influenced standards work in Unix, where the OPEN LOOK specification was developed partly to reduce legal exposure. Patent law was not the main tool here, since the desktop metaphor could not be patented in the same sense as a hardware invention.

Basic hardware architecture

Nearly every personal computer follows the stored-program design that John von Neumann and his colleagues described in 1945 in the report on the EDVAC computer. In this design, instructions and data sit in the same memory, and the processor fetches instructions from that memory one after another. Because a program is just data to the machine, software can be loaded, replaced or even generated without changing the hardware. The main parts are the central processing unit (CPU), which contains a control unit that sequences work and an arithmetic logic unit that performs calculations; main memory; input and output devices; and the buses that connect them. A bus is a shared set of wires that carries addresses, data and control signals. The shared path between processor and memory is also the source of the so-called von Neumann bottleneck, which caches and other techniques are designed to soften.

The memory hierarchy

A PC does not rely on one kind of memory. It uses a memory hierarchy in which speed, capacity and cost pull in opposite directions. At the top are the registers, a few dozen to a few hundred tiny storage locations inside the CPU core that hold the values an instruction is working on. Below them come the cache levels (L1, L2 and usually L3), built from fast semiconductor memory on or near the processor die. L1 is the smallest and fastest, typically tens of kilobytes per core, while L3 is shared among cores and measured in megabytes. Main memory, made of DRAM modules, comes next: it offers gigabytes of capacity at much lower cost per byte, but its latency is an order of magnitude or more higher than cache. At the bottom sit solid-state drives and hard disk drives, which hold data after power is removed and offer capacities in the terabyte range, but with access times that can be thousands of times slower than RAM.

The hierarchy works because programs tend to reuse nearby data and recently used data, a pattern usually called locality of reference. Hardware keeps copies of frequently used blocks close to the CPU, so most requests are satisfied at the upper levels. The practical consequence is that adding RAM helps mainly by letting a working set fit in memory and avoiding slow swapping to disk, while cache size and memory latency govern much of the everyday speed that users experience.

Motherboard, chipset and firmware

The motherboard is the printed circuit board that carries the CPU socket, memory slots, expansion slots and the connectors for storage and peripherals. In earlier designs, a chipset of separate integrated circuits handled tasks such as memory control and input/output routing, with a "northbridge" and "southbridge" dividing the work. In modern systems much of the memory controller and high-speed links have moved onto the processor itself, so the chipset mainly manages slower I/O. Firmware stored in a flash chip on the board initializes the hardware at power-on and then hands control to the operating system; the traditional BIOS and its successor UEFI are covered in the software sections of this entry.

Standard form factors make parts interchangeable. The ATX specification, which Intel released in July 1995, replaced IBM's older AT layout as the default for desktop cases, boards and power supplies. A full ATX board measures 305 × 244 mm (12 × 9.6 inches), and smaller variants such as microATX and Mini-ITX keep the same rear I/O layout with fewer expansion slots. Standard mounting holes, I/O panel positions and power connectors allow a builder to swap a board, a supply or a case from different manufacturers. Later, the ATX power connector was expanded from its original 20-pin form, but the basic interface remained compatible across generations.

Power supply and thermal management

A desktop power supply converts mains alternating current into several regulated low-voltage direct-current rails, typically at 12 V, 5 V and 3.3 V, and delivers them to the motherboard and drives. Its efficiency rating indicates how much of the drawn power reaches components rather than being lost as heat. Every switching transistor and every watt of current flowing through resistance generates heat, and that heat must be removed. Heatsinks, fans and, in some systems, liquid cooling transfer heat from the processor and graphics chips into the air. When a processor's temperature exceeds its design limit, it reduces clock speed or voltage in a process called thermal throttling, which trades performance for protection. This is why thermal design often limits sustained performance more than peak capability. Laptops and small form-factor machines face the tightest constraints, because limited space and airflow force designers to balance clock speed, fan noise and battery life.

Input, display and expansion peripherals

Early personal computers were controlled mainly through punched cards, manual switches and typed commands, so the keyboard was almost the only input route. Prior to his inventions, computers relied on keypunch cards or manual switches for control, and data was only viewable in print. The typewriter-style QWERTY keyboard carried over from office machines and became the default layout of the IBM PC and its clones. Its design shaped habits that persisted for decades: touch-typing, a fixed set of modifier keys, and a reliance on keyboard shortcuts that power users still prefer. Pointing devices arrived later. Development of the mouse began in the early 1960s by SRI International's Douglas Engelbart, while he was exploring the interactions between humans and computers. Engelbart's first prototypes were paired with a chord keyset, and his team's demonstration in 1968 showed the mouse as one part of a wider system. The mouse only became ubiquitous after graphical interfaces made pointing the natural way to operate a screen. Apple never even considered including Engelbart's keyset in the early personal computers. The keyset was too costly, clunky, and complicated, and the simpler mouse won out. Its adoption with the Xerox Alto and the Apple Macintosh in the 1980s made pointing and clicking a standard habit.

Touch input followed a separate path. Resistive touchscreens were common in niche and point-of-sale devices for years, and capacitive multi-touch became mainstream only with smartphones and tablets after 2007. On desktops and laptops, touch remained a supplement to the keyboard and trackpad, and it was adopted most readily on convertible and detachable designs.

Display technology changed as well. Early PC monitors used cathode-ray tubes, in which an electron beam scanned a phosphor screen; these were bulky and power-hungry but offered good contrast and fast response. Liquid-crystal displays replaced them in laptops first and then on desktops, with cold-cathode and later LED backlighting. Organic LED (OLED) panels, which emit light per pixel and need no backlight, became common in premium laptops and monitors in the 2010s and 2020s. Graphics adapters evolved from simple text and framebuffer cards into separate graphics processing units (GPUs) with thousands of parallel cores. Those cores render images, and the same parallel structure now speeds up general computation and machine-learning workloads.

Storage peripherals show the same progression from removable media to fixed, faster devices. IBM began developing the floppy disk in 1967 and introduced it in 1971 in an 8-inch format. Floppy disks were an almost universal data format from the 1970s into the 1990s, used for primary data storage as well as for backup and data transfers between computers. The 5.25-inch disk arrived in 1976 and the 3.5-inch micro-floppy later became the last mass-produced format. Hard drives had existed since the 1950s on mainframes, but the hard disk drive (HDD) entered the consumer market in the early 1980s but did not achieve prominence until the 1990s. Solid-state drives, which store data in flash memory without moving parts, first appeared in 1991 in small capacities and became the usual system drive in the 2010s as prices fell. Optical media, such as CD-ROM and DVD, served as software and distribution media from the late 1980s onward, and USB flash drives later replaced many of their roles.

Expansion has always depended on standard connectors. Early PCs used internal expansion slots on the motherboard, such as the ISA bus, into which users fitted sound, video, network and storage cards. An expansion slot is a socket on a computer motherboard that allows you to add additional components to your system. Serial and parallel ports, which connected printers, modems and mice, were gradually replaced by USB, which combined data transfer with power and hot-plugging. Standardised ports let buyers add peripherals from different vendors without opening the case.

That upgrade path has narrowed. Desktop towers still accept standard memory modules, drives and graphics cards, but in thin laptops and all-in-one machines many parts are fixed to the board. The industry's push toward thinner, lighter designs has led manufacturers to solder components directly to motherboards, eliminating upgrade possibilities. Some systems keep one memory slot next to soldered memory, which gives a limited upgrade path. Business-class laptops have generally kept more replaceable parts than consumer models. The shift has made the computer's lifespan depend more on the purchase configuration than on later expansion, a point discussed in the sections on repairability and lifecycle.

Operating systems, firmware and software ecosystem

A PC is useful because of layers of software between the hardware and the user. The lowest layer is firmware, which starts the machine. Above it runs an operating system that schedules programs, manages memory and storage, and gives applications access to devices. Each platform's software library, not only its hardware, determines what users can do with it.

Main PC operating systems

Windows held 62.16% of the global desktop OS market in June 2026, down from 71.68% in March 2025, based on StatCounter traffic data. Other trackers report a higher figure for the same period: one measured Windows at 71.18 percent of page views in July 2026. These numbers measure web traffic, so they reflect active use of a page-view sample rather than installed base, and they differ by method. Windows remains the main platform for consumer, enterprise and gaming PCs, from the Windows line that began as a graphical shell over MS-DOS and later moved to a new kernel.

Apple's macOS, a Unix-based system, is strongest among creative professionals and some developers. In the same StatCounter series, macOS held 14.58% in June 2026. Linux is open source and dominates servers and scientific computing: all of the top 500 most powerful supercomputers since November 2017 have used Linux. On desktops its share is small, a few percent, but it is widely used by developers and in India, where India's 16.21% Linux share reflects cost-sensitive hardware buying, a large developer base, and government digital literacy programs. ChromeOS is built around the web browser and web applications. Its use is concentrated in education: in the US, ChromeOS runs at 8.44%, roughly 4.5 times its global level, almost entirely from K-12 school laptops.

Firmware: from BIOS to UEFI

Firmware is the first code a PC runs. The traditional BIOS was created in the late 1970s to provide load and basic I/O services to operating systems on the original x86 PCs. Its 16-bit design, its reliance on the master boot record and its slow sequential startup became constraints as hardware grew more complex. Intel began work on the Extensible Firmware Interface in 1998, and in 2007 a group of companies including Microsoft, Apple, AMD and Dell agreed to adopt the renamed UEFI as the successor. The UEFI Forum was formed in 2005 to make the standard interface more open, broaden its usage, and allow the system firmware community to participate in maintenance and further development of the standard.

UEFI initializes the hardware, locates a boot loader and passes control to the operating system, but it adds features the BIOS lacked. Modern firmware can support remote diagnostics and repairing of computers, even when there isn't an operating system. It also supports Secure Boot and GPT partition tables, which allow very large disks. Most implementations keep a compatibility module for older boot methods, so the setup screen still often says "BIOS" even on modern machines. Windows 11 depends on this: when the device starts, the firmware interface controls the booting process of the PC, and then passes control to Windows or another operating system.

Application ecosystem and drivers

Applications sit on top of the operating system. Office suites (Microsoft Office, LibreOffice), web browsers, games, creative tools such as photo and video editors, and developer tools such as compilers for C and runtimes for JavaScript make up the everyday software base. Developers also run virtual machines to test other systems, and open-source software supplies many of the components used across platforms.

Hardware is reached through drivers. A driver is a program that translates the operating system's generic requests into the specific commands a device understands. Printers, network cards, sound chips and graphics cards each need one. Drivers are usually written by hardware vendors, and their quality strongly affects stability and performance. Operating systems that support many devices out of the box, or that let vendors ship drivers through a common model, reduce this burden.

Software compatibility and the Windows platform

Windows has kept its position less because of technical superiority than because of its software library. The IBM-compatible PC standard let many vendors build hardware that ran the same applications, and Windows inherited that base. Applications written for Windows encourage users to stay on Windows, and vendors write for the platform with the largest user base, a classic network effect that raises switching costs. In gaming this is especially visible: in February 2026, Steam's hardware survey found Windows at 96.61% of surveyed users, far ahead of Linux at 2.23% and macOS at 1.16%.

Alternatives have struggled for this reason. IBM's OS/2 and several later systems failed to attract enough applications, and Linux desktops have suffered from fragmentation and driver gaps. Apple and Google have succeeded in narrower segments, using their own hardware or education markets. Microsoft's end of mainstream support for Windows 10 in October 2025 pushed many users toward new hardware, which connects to the replacement-cycle issues discussed elsewhere in this entry.

Device forms and categories

Personal computers are built in several physical forms, and the boundaries between them have moved as processors, displays and batteries have become smaller and more efficient. The traditional family is the desktop tower, in which the processing unit sits in a separate case from the display and keyboard. The all-in-one places the computer hardware behind the screen, which removes the tower. The notebook or laptop packs the same functions into a battery-powered clamshell. Around these sit the workstation, a high-end desktop or mobile machine with professional graphics and error-correcting memory for engineering and content creation; the gaming PC, usually built around a powerful graphics card; and the mini PC and small-form-factor machine, which trades expansion room for a compact chassis, often for office, classroom or media use.

Convertibles, detachables and Chromebooks

Several hybrid designs blur the line between a laptop and a tablet. A convertible notebook has a hinge that lets the screen fold back over the keyboard, while a detachable has a display that separates completely from its keyboard. Industry trackers treat the two differently: convertibles are counted as notebooks, whereas detachables are usually grouped with tablets. Chromebooks, which run Google's ChromeOS and are mostly built around cloud applications, are a distinct category within this range. They are concentrated in schools. One market estimate reports that education accounts for about 60% of Chromebook sales, and that the operating system held only about 1.86% of the global desktop OS market in March 2026, compared with 8.44% in the United States.

Smartphones and tablets have absorbed many everyday tasks such as messaging, media consumption and light browsing, and smartphones in particular have displaced the cheapest home computers for many households. Yet the laptop remains the default personal computer for most office, student and professional users because it combines a full keyboard, a conventional operating system and the ability to run demanding software.

From desktop-led to portable-led sales

For much of the PC era desktops were the majority of units. That is no longer true. In the second quarter of 2026, Omdia reported that global shipments of desktops, notebooks and workstations totalled 65.7 million units, of which about 51.7 million were notebooks and 13.9 million were desktops, including desktop workstations. The shift reflects several engineering factors. Portable designs depend on low-power processors and on lithium-ion batteries whose energy density determines how long a machine can run unplugged. They must also dissipate heat within a thin enclosure, which limits sustained performance: a laptop that runs at full clock speed for a few seconds may slow down once its cooling system is saturated. Desktops retain advantages in cooling capacity, upgradability and performance per dollar, and they remain common in businesses and among gamers and professionals who need large displays or multiple GPUs.

Category blur and counting problems

Market statistics depend heavily on how a device is classified, and the categories do not always agree. IDC's personal computing device tracker separates traditional PCs (desktops, notebooks and workstations) from tablets, and it reported a Q2 2026 total of 68.2 million PCs, while Omdia's narrower definition produced 65.7 million for the same quarter. Detachable tablets, which IDC counted separately from notebooks in the first quarter of 2026, illustrate the problem: a device sold as a laptop in one survey can appear as a tablet in another. Chromebook market-size figures show a similar spread. Research firms have valued the market for 2024 and 2025 anywhere between roughly $7.8 billion and $14.7 billion, depending on scope, forecast method and whether they measure revenue or units. Analysts therefore compare published totals with caution and check which devices each firm includes.

Processor architectures: x86, Arm and Apple silicon

The x86 lineage

The CPU of most personal computers has long belonged to one instruction-set family. The x86 instruction set architecture gave rise to Intel's most successful line of products and became a ubiquitous presence in desktop computers. Its roots lie in the microprocessor that Intel released in June 1978, a 16-bit design with a 20-bit address width. The cheaper 8088 variant powered IBM's Personal Computer in 1981, which fixed x86 as the foundation of the IBM-compatible market.

Intel did not control the family alone. IBM wanted a second source for its processors, so Intel gave AMD and Harris Semiconductor a license to make the 8086, 8088 and later the 80286. The relationship later turned adversarial. Intel tried to cut AMD out, which led to a legal battle, and in 1995 AMD won the right to make its own x86 chips. AMD then invented the 64-bit extension of x86, known as AMD64 or x86-64, and Intel later obtained licensing rights to produce x86-64 processors as well. Today the two firms rely on a patent cross-licensing agreement that gives each access to essential patents and improves software interoperability. Market estimates vary by method, but one analyst put x86 systems at 93.96% of PC market share in 2025.

The main reason x86 has stayed entrenched is compatibility. Decades of operating system and application software, especially on Windows, assume the x86 instruction set, so changing it carries large switching costs for users and developers alike. The same logic explains why x86 still dominates servers, high-performance workstations and critical legacy applications, even though newer Intel and AMD designs have narrowed the efficiency gap with Arm.

Arm and the efficiency argument

Arm-based designs grew from embedded and mobile use, where battery life and energy use matter most. Arm already dominates the smartphone market, and its lower power consumption is the main reason it has spread into PCs. Qualcomm's Snapdragon chips brought Arm to Windows laptops, but the early numbers were small: one estimate put Qualcomm's Windows shipments at about 720,000 units in the third quarter of 2024, around 0.8% of the market. Forecasts for the segment have been revised repeatedly. TrendForce expects Windows on Arm AI notebook penetration to rise from 1.2% in 2025 to 3.2% in 2026, while earlier projections from Canalys had counted all Windows, Mac and Chromebook devices as "PCs" and predicted a much larger share. Readers should check how each firm defines an Arm PC before comparing figures.

Apple silicon

Apple's move away from Intel is the largest single shift in PC architecture since the x86 standard was set. Apple announced in June 2020 that it would transition the Mac from Intel to its own custom silicon. The company had long built processors for iPhones and iPads; it started expanding in-house silicon design with the A4 in 2010. The first Apple silicon Mac arrived quickly: the first new computers running Apple silicon came out in November 2020. Apple's chips are built on Arm, and each M1 integrated CPU, GPU, Neural Engine, unified memory and input/output controllers into a single system-on-chip. Unlike Intel's modular design, the unified memory let the CPU and GPU share high-bandwidth memory. Reviewers reported large gains in performance per watt compared with the Intel predecessors.

The transition was not a clean break. Apple shipped Rosetta 2, an emulation layer that let apps built for Intel run on the new chips, and developers were given early hardware through the Developer Transition Kit, which used an A12Z processor in a Mac mini enclosure. Apple had said the change would take "about two years", but the schedule slipped, and the transition was finished in 2023 with the launch of an Apple silicon Mac Pro. Critics argued the rollout was uneven, since older Intel models stayed on sale for longer than the new ones had been available.

Compatibility, emulation and developer choices

Architecture determines which software runs natively. An application compiled for x86 must be translated, emulated or rebuilt before it can run on Arm, and the cost of that work shapes which platforms developers support. Windows on Arm has faced this problem most acutely. Analysts describe compatibility as the make-or-break factor, although the gap is shrinking as more apps ship native Arm versions and emulation improves. Developers therefore face a choice between maintaining separate builds, relying on translation layers that cost performance, or waiting for their users to move. For many users, the decision depends less on raw speed than on whether the tools they need are available natively.

Networking and internet connectivity

Early personal computers were isolated machines, and their first route to other computers ran over the telephone network, which had been built for voice. A modem bridged that gap by converting digital data into signals that could travel over analogue lines. In the late 1950s, AT&T developed the first commercial, mass-produced computer modem, which linked SAGE computers across the United States at 110 bits per second. After the Carterfone decision allowed non-Bell companies to sell equipment, the Hayes 80-103A, a 300-baud device, became one of the first modems made for a personal computer. Hayes's 1981 Smartmodem integrated its own command set, which became an industry standard and let users dial, answer and hang up through an RS-232 serial port.

Dial-up access then spread through the 1990s. In the 1990s, tens of millions of people in the United States alone used dial-up modems for internet access. Commercial providers such as America Online, Prodigy, EarthLink and CompuServe dominated early access in the US. The connection occupied the household telephone line, and it was not until 1996 that the 56k modem let users browse the web at 56,000 bits per second. Broadband arrived in the following decade. By the mid-2000s, DSL and cable broadband were widespread, with Wi-Fi beginning to appear in homes.

Local connectivity developed along a separate track. Ethernet, developed at Xerox PARC in the 1970s and standardised by the IEEE as 802.3 in 1983, became the dominant wired local area network technology in offices and, later, homes. Wi-Fi is the commercial name, licensed by the Wi-Fi Alliance, for wireless local area networks based on the IEEE 802.11 specifications. The technology was originally developed for mobile computing devices such as laptops, but it is now used for the internet, VoIP telephony, gaming and basic connectivity of many consumer electronics. Early versions appeared in the mid-1990s, and Apple's inclusion of the technology in the iBook laptop in 1999 helped it take off.

PCs could interoperate only because they shared common protocols and naming conventions. The Internet's Transmission Control Protocol/Internet Protocol suite (TCP/IP) defines how data is addressed and delivered, while the domain name system solves the problem of finding machines by name. Early networks relied on a central host table. Such a system only works well when the number of devices is small. In 1981, the idea of organising names into domains was first introduced in RFC 799, Internet Name Domains. Its hierarchical structure led to the DNS as it is known today. DNS translates readily memorised domain names into the numerical IP addresses needed to locate services and devices. The initial top-level domains included .com, .net, .org, .edu, .gov, .mil and .arpa. Through this layering, a PC running any web browser or mail client could reach any server on the network without knowing its numeric address.

Connectivity changed the PC from a standalone tool into an access point for remote services. Email, the World Wide Web and, later, cloud services turned the machine into a client of resources located elsewhere, so that software and data could live on remote servers. The home router became the gateway at the centre of this arrangement. DSL routers are the centre of private home networks, serving as the gateway that all network devices use to reach the Internet. Many consumer units combine wireless and wired components with a broadband modem. This device is often overlooked, yet it is also the first line of defence against hackers, malware and viruses. Security research has repeatedly found weaknesses. Research found that the firmware running most broadband routers is years out of date and riddled with potential security holes. In one UK study of 13 ISP-supplied models, two-thirds were found to be flawed, although the providers disputed the findings. A compromised router can be turned against the household: an attacker can position themselves as a middle-person by changing router settings such as the DNS servers, gaining control over unencrypted traffic. Routers controlled by attackers can also be used to build botnets. Connected devices beyond the PC widen the exposure, and compromised IoT devices can join botnets, relay traffic into the rest of the local network, and open outbound tunnels for remote control. Users often lack the expertise to maintain protective tools, and an old router may stop receiving firmware updates that fix security problems, making it easier for intruders to reach the network. These risks are part of the wider concern with cybersecurity that accompanies constant connectivity.

Industry structure and market history

Annual PC unit shipments grew strongly through the 1990s and into the 2000s. The highest annual total in the sources consulted is for 2011, when global PC shipments reached their historical high of 365.4 million units. Analysts later used the phrase "Peak PC" for this point, and they attributed the subsequent decline largely to smartphones and tablets taking over many of the tasks people had once relied on PCs for. The early-2000s slump is visible in the data as a weak 2001, after which the market rebounded.

The decline after 2011 was gradual rather than sudden. The total fell to as low as 260 million in 2018, and IDC's count for that year was 258.5 million units. IDC also recorded 2018 as the seventh consecutive year of year-over-year decline.

The COVID-19 pandemic briefly reversed this trend. Demand rose as people needed computers for work, school, or entertainment while staying at home. According to Gartner, shipments reached 275.1 million units in 2020, up from 262.6 million in 2019, a growth rate of 4.8%, the strongest in a decade. A later compilation reports that shipments fell from approximately 342 million units in 2021 to around 242 million units in 2023. These two series are not directly comparable, because they are not published on the same basis, and the compilation does not state its scope.

Shipments recovered from 2024. IDC reported 262.7 million PCs shipped in 2024, up 1% from 2023. Gartner put full-year 2025 shipments above 270 million units, a 9.1% increase, and attributed much of the growth to robust consumer demand and business demand fueled by the Windows 11 upgrade cycle. The recovery is now being reversed. IDC forecasts that global PC shipments will decline 11.3% for full-year 2026, with the worst declines expected in the fourth quarter, and it links the decline to a persistent memory shortage for which no meaningful relief is expected before end of 2027.

Replacement cycles drive much of this volatility. Medium-to-large companies typically update their PCs at least every three to four years. Operating-system support deadlines can pull purchases forward. Microsoft's end of support for Windows 10 on October 14, 2025 helped lift shipments, and Gartner linked its 8.2% year-over-year increase in global shipments in the third quarter of 2025 to that deadline. Commercial customers can continue receiving paid security updates, with Microsoft's Extended Security Updates program for Windows 10 available in one-year increments until October 2028.

Vendor concentration is high. In the fourth quarter of 2025, Gartner reported that Lenovo held about 27.2% of worldwide shipments, HP Inc. about 21.5%, Dell about 16.5%, and Apple about 9.4%. The top three vendors together accounted for roughly 65% of the market, and the top three vendors, Lenovo, HP Inc. and Dell, increased their market share year-over-year.

Market-size estimates differ because research firms define the category differently. Gartner only counts desktops and notebooks as PCs, excluding workstations, and its figures cover desktop and laptop PCs equipped with Windows, macOS and Chrome OS. IDC's traditional PC count includes desktop, notebook, and workstation units, and IDC also publishes broader personal computing device data that covers tablets. Gartner's figures are also based on shipments selling into channels, which means they measure vendor sales to distributors and retailers rather than end-user purchases. Revenue estimates vary even more: one compilation lists a 2023 market value of about $155 billion and a 2025 expectation of about $223 billion, but these figures are drawn from different research houses with different definitions. Readers should therefore compare series only after checking their scope, timing and measurement basis.

Social uses: work, education, creativity and play

Office work and spreadsheets

The PC's entry into everyday office work is usually traced to one program. VisiCalc was released in 1979 for the Apple II and is generally described as the first electronic spreadsheet, letting users lay out figures in a grid and see dependent cells update when an input changed. Its release made it the first electronic spreadsheet program for personal computers, brought out for the Apple II in 1979. Buyers reportedly bought Apple IIs mainly to run it, which made it an early "killer app" that justifies the purchase of the hardware it runs on. For accountants and small firms, the gain was concrete: in some cases a 20-hour-per-week bookkeeping chore became a few minutes of data entry.

The software went on to sell well, reaching over 700,000 copies in six years. A version for the IBM PC followed in 1981 and extended its reach to the new standard. A research estimate cited by one history site suggests that by 1983, VisiCalc held roughly 85% of the spreadsheet software market on personal computers, though such market-share figures from the early period are rough. Successors such as Lotus 1-2-3 and later Microsoft Excel built on the same idea and became standard tools in finance and planning. The broader pattern followed: word processors, databases and printed reports moved from typing pools and mainframe batch jobs to desks. Many analysts argue that this shift flattened some clerical hierarchies, reduced demand for routine data-handling roles and made analytical skill a larger part of white-collar work, though the net effect on employment is debated.

Education

Schools and universities were early customers. Before the mid-1980s, most educational computing ran on shared terminals connected to mainframes or minicomputers, and the PC brought computing into individual classrooms and laboratories. BASIC, which shipped with most early home and business machines, became the default language for teaching programming to beginners. Through the 1980s and 1990s, schools in many countries bought Apple and later IBM-compatible machines, often through local budgets or government programmes, and the PC gradually became a tool for writing, drill-based learning and, later, research. Home study also grew, as families used PCs for homework, reference material on optical discs and, after the 1990s, online sources. Access was uneven across schools and households, a theme developed in the digital divide section. Higher-education institutions adopted PCs for coursework, theses and campus networks; see education and higher-education.

Creative work

The PC turned into a studio, workshop and editing suite for several creative fields. Sound cards, MIDI interfaces and sequencing software allowed musicians to compose and record on desktop machines, and later digital audio workstations replaced much specialised hardware for many home and professional producers (see music). Desktop publishing emerged in the mid-1980s when page-layout software, laser printers and graphical interfaces combined so that designers could prepare typeset pages without traditional prepress equipment (see printing). Digital photography and image-editing tools later made the PC a central place for managing and retouching photographs (see photography and image-processing). Video editing followed as storage and processing power increased.

Games

PC gaming has roots in the mainframe and minicomputer era. MIT students developed Spacewar! in 1961 on a PDP-1 that had been used for statistical calculations. Text adventures followed; the first text adventure, Adventure, was developed for the PDP-11 by Will Crowther in 1976 and expanded by Don Woods in 1977. Home machines then made games a mass activity. Sierra On-Line released Mystery House in 1980, a title often credited with inventing the graphical adventure genre. The Commodore 64 was marketed heavily at games and is commonly reported to have become the best-selling personal computer model, with one widely cited figure of more than 17 million units.

The 1983 video game crash weakened the console market in North America, and the home computer became a refuge for players and programmers. Home computer games became popular after the 1983 crash, which led to the era of the "bedroom coder". Small teams and individuals could build companies, and shareware distribution let them reach buyers cheaply. Low costs made it possible for individuals and very small teams to build up a game development company. Technical progress also favoured PCs: a mouse and keyboard allowed players to point quickly at places on the screen, and games took advantage of better graphics and sound. In 1992 Wolfenstein 3D showed that PCs could deliver fast, immersive first-person experiences, and Doom followed in 1993. In the 1990s the PC also became a showcase for adventure titles such as Myst. Yet in the 1990s, PC games lost mass market traction to console games, and the PC's gaming role later re-centred on online play, digital distribution and enthusiast hardware. The distinctive habits of PC play, including the desk-bound setting, also shaped the games themselves; see video-game.

Scientific and engineering use

Beyond office and home use, PCs became common instruments for analysis and small-scale modelling. Researchers and engineers used spreadsheets, statistical packages and scripting tools to clean data, fit models and produce figures, while computer-aided design software let designers draw and test parts on desktop machines. As graphics processors became programmable, they were also used for general-purpose numerical work (see graphics-card and cuda), and PCs increasingly served as front ends for jobs run on clusters or supercomputer systems. Detailed treatment of individual scientific fields and of large-scale computation belongs in their own entries, including parallel-computing.

Digital divide and social effects

Unequal ownership and skills

PC access has followed income and schooling for as long as governments have measured it. Late-1990s U.S. analysis treated income and education as among the leading drivers of the digital divide. Race added a separate dimension. After controlling for income, occupation, urban residence and household structure, one study of 1990s data still found computer ownership significantly lower for black non-Hispanic households, while the gap for Hispanic households was smaller.

Later data show that the poorest households remain far behind. A 2019 Pew survey found that more than four in ten adults in lower-income households lacked home broadband or a traditional computer. Lower-income adults were also more likely to be "smartphone-dependent" internet users, rising from 12% in 2013 to 26% in early 2019. The same pattern appears among children: in 2015, 35% of lower-income households with school-age children had no home broadband connection, a gap often called the "homework gap."

Some indicators have improved. An analysis hosted on the Benton website reports that households with no computer at all fell from 12.74 million to 5.06 million between 2016 and 2023, although the same analysis found that smartphone-only households rose by nearly 54 percent. Geography and infrastructure matter as well. In many developing-country schools, a lack of electricity, inadequate computing infrastructure and little IT in homes limits what a PC can do, and a study of young people's computer use found that students from well-resourced schools have far more access to computers and the internet.

Policy responses

Governments have tried to close these gaps through public provision. One proposed Australian scheme would have supplied senior secondary students with computers and helped fund high-speed connections for nearly all homes, aiming at 98% of households. England committed to giving all children home computer and internet access for school and college work. Norway's one-to-one laptop policy, rolled out gradually across high schools, became one of the better-studied examples (see below).

Work patterns, remote and hybrid work

Home computing became a workplace question as well as a domestic one. One estimate cited on the Benton site suggests that 37% of U.S. jobs can be performed remotely, and these jobs account for 46% of all wages. The shift to remote and hybrid arrangements after 2020 made a capable personal computer and reliable broadband close to essential for knowledge-based roles in professional services, finance, administration, information work and education. Construction, retail, food service and care work, by contrast, remained largely on-site, so workers in those sectors gained little from home PC access. This concentration is a general observation rather than a finding drawn from the sources above, and it deepens the divide between workers who can and cannot work from home.

Developing countries, secondhand machines and low-cost designs

Low-cost computing has been a recurring response to the global divide. The One Laptop per Child effort, which was implemented in over 40 countries, was one of several projects that aimed at laptops in the range of $200–$300. Designers have also sought alternatives to the Wintel platform: the Simputer, a pocket device, was sub-$200 and developed in India, and was made sharable rather than personal. Not all such efforts succeeded commercially. AMD's Personal Internet Communicator, sold at $249 for the computer and a 15-inch monitor, was later discontinued as a money-losing line.

Secondhand and refurbished machines are another route. A total-cost-of-ownership model for schools in several developing countries found that ultra-low-cost laptops and secondhand desktops cost under US$300 per unit, but their shorter lifespans add replacement costs, so low-cost new desktops can have a lower total cost than secondhand PCs. Refurbishment nonprofits operate on this model; one reports that it has shipped over 50,000 computers to 3,729 sites across 54 developing countries since 2000. Some researchers argue that shared access points matter more than private ownership for low-income users, and that collective access points where users benefit from internet services on equipment they do not own may be the most important policy for the poor.

Research on access and outcomes

Evidence on whether access changes results is mixed. A study of twelve Sydney high schools found that 1:1 laptop provision had statistically significant, positive associations with science attainment, and the effect was larger in physics, where students and teachers used simulations and spreadsheets more. Evaluations commissioned by Microsoft reported that laptop students showed significantly higher achievement in nearly all measures after one year, although these were vendor-linked studies.

Other evidence is less favourable. In Peru, the OLPC program did not improve academic performance, and students had slightly lower on-time completion rates. A long-term follow-up concluded that laptops did not have a positive impact on educational outcomes. In Catalonia, a study found that students in schools that adopted the program scored about 5.1% lower on average, with a larger negative effect among boys. In Norway, the program showed no overall effect on well-being outcomes, although it reduced the frequency of bullying at school.

These contrasting results point to a central disagreement. Some researchers argue that the nature of computer use matters more than access itself, and that digital inequality extends to skills and use, not only to ownership. The OLPC evaluation in Peru found that students improved at using the provided laptops, but only marginally improved their ability to use PCs and showed no improvement in internet skills. Others note that access has become widespread at many income levels, yet outcomes still differ. The debate is therefore less about whether devices matter than about which combination of hardware, teaching, connectivity and skills is required to change results.

Security, privacy and malware

From floppy disks to networked threats

Self-replicating code was described before personal computers existed. The concept of a computer virus, a program capable of replicating itself, was first mentioned in John von Neumann's 1949 essay. The first well-documented network program of this kind was the Creeper, which was detected on ARPANET in the early 1970s and could gain access independently through a modem and copy itself to the remote system.

Most early PC malware instead spread offline, through shared media. Elk Cloner, written around 1982 by a high school student, infected the Apple DOS 3.3 operating system using a technique now known as a boot sector virus. Its typical trigger was a game disk, and every 50th time the program was run, instead of executing normally, it would change to a blank screen that displayed a poem about the virus. The 1986 Brain virus was the first global IBM-compatible virus epidemic, which infected the boot sector, and was able to spread practically worldwide within a few months. Its authors reportedly intended it to deter copying of their software rather than to cause damage. Because each shared disk could carry an infection onward, these outbreaks depended on human exchange rather than on any network.

Networked worms and later malware followed the spread of the internet and email. A long list of named threats, including the Morris worm, ILOVEYOU, Melissa, Code Red, Slammer, Stuxnet and CryptoLocker, illustrates how the methods changed over time. Ransomware, which encrypts a victim's data and demands payment, is now the most visible criminal category. Counting it is difficult. Comparitech's tracker logged a record 2,627 attacks in the third quarter of 2026, while Ransomnews had confirmed only 756 attacks for the year to date at the time of its update. Black Kite recorded 7,551 victims over its reporting period, and one industry estimate holds that nearly 85% of attacks go unreported. These figures use different sources and methods and should not be compared directly.

Boot-time and firmware protections

Because firmware runs before the operating system, it is a target for attackers who want to persist across reinstallations. Antivirus software, which runs after the operating system has loaded, is ineffective at detecting and remediating malware in the early-boot firmware environment that executes before the operating system.

UEFI Secure Boot addresses this gap. It checks the digital signatures of boot components before running them. Secure boot validates signatures using RSA-2048 public key certificates, and hashes are listed in SHA-256 format. The feature was first specified in UEFI 2.3.1, and it first became available as a feature in UEFI in 2016, about the time of Windows 8. Secure Boot does not cover every scenario: UEFI relies upon Secure Boot or vendor-specific boot protection solutions – no validation or protection of the boot process is granted simply by choosing UEFI over legacy mode.

The Trusted Platform Module (TPM) complements this with measurement rather than blocking. The recorded measurements are stored in a tamper-resistant log, such as in a TPM. Windows uses this to load only trusted OS bootloaders on PCs with UEFI firmware and a TPM. Operating-system permission models, such as separate user accounts and restricted administrator rights, add another layer by limiting what a compromised program can change.

Patching, updates and privacy

Security depends on keeping software current. Government guidance recommends that firmware should be updated regularly and treated as importantly as operating system and application updates. Unpatched systems remain a major risk: Black Kite reported that 43.5% of victims still carried a critical patch vulnerability when Black Kite checked their posture again.

Privacy concerns arise from the other side of the same relationship. Operating systems and applications collect diagnostic data, usage information and account identifiers, often by default, and the boundaries of that collection are frequently unclear to users. These questions connect to data privacy and data governance, and they become more pressing when a PC is tied to cloud accounts.

Usability, compatibility and responsibility

Stronger protections often reduce flexibility. Secure Boot can block unsigned operating systems, so installing a Linux distribution alongside Windows may require configuration changes. Some older hardware and software do not function in UEFI mode. Requirements have also spread into other software. Activision's guidance for Call of Duty asks players to enable TPM 2.0 and Secure Boot, arguing that these features help protect your system against cheats and unauthorized access by verifying your PC's integrity from startup to gameplay. Windows 11 likewise requires TPM 2.0, which excludes some older machines.

Responsibility is shared. Vendors must sign boot components, issue timely firmware and OS patches, and design defaults that are safe without being unusable. Users and administrators must install updates, protect firmware with passwords where possible, and maintain backups against ransomware. Security measures that are disabled, skipped or left unpatched account for many real-world incidents, so the effectiveness of these features depends on how they are maintained as much as on how they are designed. Mainstream protections such as cybersecurity practices, cryptography and operating-system permission models only work when vendors and users both keep their part of the arrangement.

Repairability, right to repair and lifecycle

Early PCs were built to be opened. In most older laptops, memory modules came as removable sticks that sat in slots, so owners could easily swap them. From the 2010s, makers moved components onto the board to make machines thinner. Apple began to solder certain components as it pushed its laptops to be ultra thin and ultra light. Today many modern laptops have low-power DDR memory soldered directly to the motherboard. Soldered chips take less board space, which designers can give to a larger battery or better cooling. The cost falls on the owner: a memory fault can mean a whole motherboard swap instead of a simple RAM replacement. Soldering is not always absolute. A 2021 Hackaday case study describes an engineer who raised a Dell XPS 13 7390 from 8 GB to 16 GB, though the work required specialist rework. Specialist equipment for such work is costly, and upgrading requires specialised microsoldering equipment (£500–5,000+) and carries high risk of motherboard damage. Some workstations still keep removable memory, and newer CAMM2 modules are proposed as a way to combine thin designs with upgradeability, though it's too early to tell whether they will succeed.

The right-to-repair movement spread through the 2010s, moving from the east into Western Europe. In 2017 the European Parliament recommended that member states give consumers a legal right to repair electronics. The EU's directive 2024/1799 is now the central instrument. According to the European Commission, it starts to apply on 31 July 2026 and lets consumers request that manufacturers repair products that are technically repairable under EU law. Repairs must be done for free or at a reasonable price, within a reasonable timeframe. Whether laptops fall within scope is disputed. One analysis reads the annex as excluding them, arguing that the underlying EU documents each refer to very narrow definitions from which laptops are explicitly excluded. By contrast, iFixit expected a separate laptop regulation, with spare parts and documentation for diagnosis and repair should be available, devices should be modular, and software updates guaranteed for the device's expected life. Outside the EU, US states like Oregon and California have passed their own Right to Repair laws, with Oregon going as far as banning parts pairing entirely. Elsewhere, Sweden has gone further than most of the EU, making repairs and spare parts subject to lower value-added tax.

Voluntary programmes and legal duties differ in reach. Apple, for example, announced it would start providing training and spare parts to certified independent repair stores fixing Mac computers, not just iPhones. Such programmes depend on the company's choice and can be withdrawn, whereas statutes create enforceable rights. Manufacturers defend tight design on grounds of reliability and safety. Soldered memory, for instance, is much less likely to be damaged if a device is dropped, than if the chips are locked into slots designed to allow removal. Sealed designs also reduce the chance of user error during upgrade or repair. Repair advocates counter that companies may use intellectual property law to block independent repairers; iFixit has argued that companies may use copyright laws in order to ensure that people don't use an unauthorized repairer.

Lifespan evidence is contested. Researchers were unable to draw conclusive proof that manufacturers are intentionally putting out products with shorter lifespans. Consumer surveys do show rising replacement of defective goods: one summary reports an increase in proportion of consumers replacing defective appliances from 3.5% in 2004 to 8.4% in 2012. These figures cover household appliances rather than PCs specifically. Older estimates suggest PC life spans have shortened, with one California document stating that the average life span of a personal computer has shrunk from four or five years to two years. Such figures predate current data and should be read with caution.

End-of-life handling carries health and environmental risks. Electronic scrap components, such as CPUs, contain potentially harmful materials such as lead, cadmium, beryllium, or brominated flame retardants. Recycling and disposal of e-waste may involve significant risk to the health of workers and their communities. Burning plastics that contain halogenated compounds can also result in the formation of dioxin. Recyclers typically follow a sequence. Working devices are tested, wiped, and repaired before anything gets broken down. For data security, certified recyclers use NIST-approved wiping or physically destroy the drive. In California, hazardous parts, like leaded TV glass, mercury lamps, and lithium batteries, are handled separately under California law. Recovery rates remain low by some measures. A 1999 estimate found that only 11% of discarded computers were recycled, compared with 28% of overall municipal solid waste.

Environmental impact and electronic waste

Energy use in operation

A PC's electricity use depends heavily on how much of its time it spends idle. The EPA's voluntary ENERGY STAR label, introduced in the early 1990s, required a qualifying machine to drop to about 30 W when inactive. In one mid-1990s analysis, a compliant system's peak draw was about 120–130 W, with the monitor using 85–95 W and the CPU 25–35 W. The display was therefore the largest load in many office setups. Later test data reached a similar conclusion from the other direction: an early-2000s ENERGY STAR draft found that idle periods made up 69 to 97% of total annual energy use, even with power management enabled. Sleep and hibernate modes address this directly; when these features are active during inactivity, they have been estimated to cut a computer's total energy use by up to 50%.

Processor design has also lowered idle power. Testing of the most efficient desktop processors from AMD and Intel found roughly 1 to 7 W in idle mode, including voltage-regulator losses. Older architectures consumed 20 to 34 W in the same state. Standards bodies now measure annual consumption by weighting the time spent in each power state, a method codified in ECMA-383 and used by ENERGY STAR.

Embodied energy and material demands

For most current PCs, the largest share of emissions occurs before the machine is switched on. A 2025 screening study comparing a desktop and a laptop found that the desktop had a higher overall carbon footprint (679.1 kg CO2eq) than the laptop (286.1 kg CO2eq). In both cases manufacturing contributed the largest share of emissions, followed by use. Other lifecycle analyses put the embodied share even higher; one arXiv study reports that embodied footprint accounts for 70–75% of total emissions for laptops. Memory and storage are a notable part of this burden. Flash and DRAM manufacturing is markedly energy-intensive, which makes solid-state drives a significant contributor to embodied carbon.

Manufacturing a PC also requires large quantities of metals, plastics and semiconductor materials, and the production of a single laptop has been estimated in industry compilations to involve substantial water and mined earth, although such figures vary widely between studies. The chip fabrication stage is discussed in the semiconductor and microprocessor entries.

Electronic waste flows and recycling

When PCs reach end of life, they enter the broader electronic waste stream. The UN-linked Global E-waste Monitor 2024 reported that a record 62 billion kg of e-waste was generated globally in 2022, equivalent to 7.8 kg per person, and 22.3% of this mass was documented as formally collected and recycled. Since 2010, generation has grown far faster than documented recycling. The growth of e-waste generation is outpacing formal collection and recycling by almost a factor of five. Personal computers are a small part of this total: small IT and telecommunications equipment, a category that includes PCs, printers and routers, totaled 5 billion kg in 2022.

Recovery rates differ sharply by region. African countries had the lowest rate, with less than 1 per cent of e-waste documented as formally collected and recycled. Cross-border movement is also significant. About 5.1 million tonnes, or 8.2% of the global total, was shipped across borders in 2022, and roughly 65% of that moved from high-income to middle- and low-income countries. Much of the material then passes through informal recycling, where valuable metals are extracted by hand or by burning. Delhi's Seelampur district, for example, is a major informal e-waste market. The same monitor values the metals embedded in global e-waste at US$91 billion, which explains both the economic incentive for recovery and the pull toward informal channels.

Mitigation and the debate over product lifetimes

Energy-efficiency standards, power-management defaults and more efficient processors are the main operational mitigations. Their effect is limited when a machine spends most of its life idle or when grid electricity is carbon-intensive, because the largest emissions may already be embodied in the hardware. This leads to the central debate. Keeping a device longer spreads its manufacturing emissions over more years, and one life-cycle study found that laptops had slightly lower total costs than desktops over a four-year period, despite higher initial investment costs. Replacing an old, power-hungry machine with a more efficient one, by contrast, can reduce use-phase emissions, but only if the operational savings outweigh the new device's embodied carbon. A refurbishment vendor citing a Cranfield University study reports that professionally remanufactured laptops produced only around 6% of the carbon emissions of an equivalent new laptop across their full lifecycle.

The answer depends heavily on assumptions. Estimates of the manufacturing share differ between analyses, and the same MDPI study notes that manufacturer-provided data generally showed lower carbon footprint values than the modeled scenarios. Results also depend on the regional electricity mix, the assumed lifetime and the end-of-life treatment. Repairability and the right to repair, which determine how long a device can realistically remain in service, are covered in another section. Comparisons across life-cycle assessments should therefore be read with attention to their system boundaries.

AI PCs and on-device inference

The "AI PC" is a marketing category more than a fixed hardware class. Its defining addition is a neural processing unit (NPU), a processor for the low-precision matrix arithmetic behind neural-network inference. The NPU complements the CPU and the GPU rather than replacing them. The most concrete benchmark comes from Microsoft's Copilot+ label, which requires an NPU above 40 TOPS (trillions of operations per second), at least 16 GB of memory and a 256 GB drive. The 40-TOPS Copilot+ floor is a Microsoft labelling gate, not a hardware capability boundary. TOPS figures are also easy to inflate. Some vendors add CPU, GPU and NPU throughput into one "platform" number, although applications usually target a specific processor.

On current Windows machines, the NPU handles tasks such as live captions with translation, Windows Studio Effects and AI-enhanced search. GPU compute is power-intensive and not suitable for continuous background AI workloads. The NPU's value therefore depends on software. NPUs are most useful when applications are specifically optimized to use them. The laptop is the main target, because it has the battery and thermal budget for always-on inference.

Analysts report a sharp rise in AI-capable shipments. Gartner estimated that AI PCs would make up 31% of worldwide PC shipments in 2025, totaling 77.8 million units. Gartner analysts forecast AI PC shipments will total 143 million units and are projected to represent 55% of total PC market in 2026. Counterpoint uses a stricter definition, projecting that AI Advanced PC penetration is projected to climb 52% YoY to reach ~59% of global shipments in 2026; up sharply from ~39% in 2025. Vendors give several reasons for the shift. Gartner ties it to productivity gains from on-device AI, and expects 100% of commercial (enterprise) PC purchases will be AI PCs by the end of 2026. It also expects software vendors to invest more in AI that runs directly on PCs, with 40% of software vendors will focus their investments on developing AI capabilities that run directly on PCs, compared to a figure of just 2% in 2024. Gartner has also linked the upturn to the Windows 10 end-of-support deadline.

Local inference offers three main benefits. It reduces latency by avoiding wide-area round trips and backhaul jitter. It supports privacy by keeping sensitive data (e.g., voice, medical text, personal documents) local. It also works offline. Privacy has limits, however. Intermediate values such as the KV cache, held in GPU memory, can leak conversation content.

The constraints are mainly memory and bandwidth. The constraining factor for local LLMs is memory. A language model must fit entirely in RAM (or VRAM) during inference. At 4-bit precision, a 7B model requires 3.5 GB for the weights alone. Larger models quickly exceed typical laptop memory, since a 70B model at Q4 needs roughly 40–48 GB. Speed is also limited by bandwidth, because the speed of data transfer between RAM and the processor often limits performance more than raw computational capacity. Local models generally trail cloud systems in capability. On-device LLMs minimize latency and keep data local, yet lightweight models typically underperform relative to state-of-the-art cloud models. Power and thermal limits also apply, although the cited research examines phones, where smartphones operate under limited compute throughput, memory bandwidth, thermal headroom, and finite battery capacity.

Forecasts should be read with care. Their definitions differ: Gartner's 54.7% share, Counterpoint's roughly 59% for "AI Advanced PCs," and a Canalys figure of 43.5% that counts machines with 40-plus-TOPS NPUs measure different things. Gartner's 2026 projection also predates the current memory squeeze. IDC counted 68.2 million PC shipments in Q2 2026, down 4.9% year over year, and attributed the decline to memory costs. Marketing labels and technical capability are therefore only loosely connected. A logo tells a buyer little about which models will run locally, how fast they will do so, or how much memory they will have.

Controversies and open questions

Decline or redefinition?

Predictions that handheld devices would displace the PC have recurred for more than a decade. Gartner once forecast that more than half of all users would use a tablet or smartphone first, rather than a laptop or desktop, for online activities. Defenders of the PC answered with shipment data. IDC argued that a slowdown in tablet shipments showed PCs still played a key role for consumers and businesses. The tablet market itself split: keyboard-less tablets fell into a steep decline, while devices with detachable keyboards, which closely resemble laptops, did well. Supporters of the PC also stress that document-heavy work, which remains a major workplace requirement, is still better suited to full-size machines than to phones. Critics reply that the category is being redefined rather than abandoned, since smartphones and cloud services absorb tasks that once required a desktop.

Memory costs and the 2026 contraction

The most recent swings have different causes. Shipments rose as organisations replaced machines before Microsoft ended free support for Windows 10 on October 14, 2025. Gartner, however, called the first-quarter 2026 gain "artificially inflated", arguing that channel partners had stockpiled inventory ahead of price increases. The larger force is memory. Gartner estimates a 130% surge in combined DRAM and solid-state drive prices by the end of 2026, which it expects to raise PC prices by 17% compared with 2025. IDC counted 68.2 million PC shipments in Q2 2026, a 4.9% year-on-year fall that it attributed to memory costs. Forecasts for the full year vary: IDC's latest figure is an 11.3% shipment decline, while Omdia expects a 10.7% decline and a further 4.9% drop in 2027 for US shipments.

AI PCs: forecasts and definitions

Forecasts for AI-capable machines diverge. Gartner projects that they will reach 54.7% of total PC shipments in 2026, or 143.1 million units, while Counterpoint expects AI Advanced PC penetration to climb to about 59% of global shipments in 2026, up sharply from roughly 39% in 2025. The Gartner figure predates the memory squeeze, and rising prices have already delayed its estimate of 50% market penetration of AI PCs until 2028. The label is also partly definitional. Gartner's count covers machines with AI embedded in the hardware components and software stack, whereas vendor marketing applies the term more loosely. Technically, on-device AI workloads benefit from lower latency, from keeping data on the device, and from working offline. The limits are set by GPU and NPU power budgets and by memory capacity, which is now the most expensive component to expand.

Open and closed platforms

The PC began as an open system: the IBM-compatible market depended on third parties building interchangeable parts. The modern architecture is more contested. Firmware protections such as secure boot defend against malware, but they can also restrict which operating systems or boot images run on a given machine. Some vendors and platform owners control app stores and software signing, and regulators in several jurisdictions have examined this gatekeeping. Open-source systems, notably Linux, and the wider open-source movement remain the main counterweight, though they depend on hardware makers releasing drivers and documentation. The balance between security and user control has no settled answer.

Platform competition

Processor competition is widening. Qualcomm has moved downstream to the $700 and below market with a new Snapdragon chip, while the x86 makers Intel and AMD see growth opportunities ahead. Apple's own silicon has shown that a non-x86 PC can compete on efficiency. Whether a new platform displaces the x86-Windows standard, however, depends less on raw speed than on software compatibility. Applications, drivers and enterprise management tools create switching costs and network effects that favour incumbents, and that inertia is the main reason the question remains open.

Long-term support and unresolved questions

Support lifecycles have become a practical concern. Microsoft's commercial extended security updates for Windows 10 are priced at $61 per device in year one, and the second year doubles to $122 per device. That structure makes keeping older hardware progressively more expensive. Several questions remain open: how long manufacturers will supply firmware and drivers; who owns and can audit the data that on-device AI assistants process; how privacy is protected given telemetry from operating systems and applications; and whether the PC stays a local, general-purpose machine or becomes a thin terminal for cloud-hosted AI. The answers will depend on regulation, memory and chip cost curves, and whether users value local control enough to pay for it.

References

1. Personal Computerscourses.lumenlearning.com
2. PERSONAL COMPUTER Definition & Meaning - Merriam-Webstermerriam-webster.com
3. What is a hybrid laptop?lenovo.com
4. Laptop & Chromebook Buying Guide via @ConsumerReportsconsumerreports.org
5. techcrunch.comtechcrunch.com
6. Rise and Fall of Minicomputers - Engineering and Technology History Wikiethw.org
7. 1961computerhistory.org
8. At MIT Fernando Corbato develops the "Compatible Time Sharing System," Precursor of Word Processing and Email : History of Informationhistoryofinformation.com
9. PDP-8 Minicomputeramericanhistory.si.edu
10. PDP-8en.wikipedia.org
11. Xerox Altoen.wikipedia.org
12. Xerox PARC had created a computer with a mouse and windows and bitmap screen before Apple or Microsoft, in fact Steve Jobs saw it and copied it for the Macintosh. So how did they completely fail to lead personal computers?worldhatchlearning.wordpress.com
13. Compatible Time-Sharing Systemen.wikipedia.org
14. Compatible Time-Sharing System — Grokipediagrokipedia.com
15. History of the graphical user interfaceen.wikipedia.org
16. Xerox Altoarchania.org
17. Xerox Alto — Grokipediagrokipedia.com
18. Alto computerbritannica.com
19. Xerox Alto - The Interface Experience: Bard Graduate Centerinterface-experience.org
20. History of Microprocessors Timelinepreceden.com
21. Microprocessor chronologyen.wikipedia.org
22. Intel 8080en.wikipedia.org
23. MOS Technology 6502en.wikipedia.org
24. Microprocessors Flourish - CHM Revolutioncomputerhistory.org
25. 6502, Z80, 8086, 68000: the four CPUs that dominated the 1980siljitsch.com
26. The History of Personal Computingeverything-everywhere.com
27. The Homebrew Computer Club - CHM Revolutioncomputerhistory.org
28. Apple Ien.wikipedia.org
29. Sol-20en.wikipedia.org
30. Home computeren.wikipedia.org
31. 45 Years Ago, Apple Kickstarted the Personal Computer Industrymedium.com
32. Altair 8800 - CHM Revolution - Computer History Museumcomputerhistory.org
33. Altair at 50: Remembering the first Personal Computer - Goto 10goto10retro.com
34. Byte Shopen.wikipedia.org
35. IBM Introduces the IBM 5150- The IBM PChistoryofinformation.com
36. IBM 5150 (Original PC) - School of Computer Sciencecarleton.ca
37. Wintelen.wikipedia.org
38. The IBM PCibm.com
39. IBM PC compatibleen.wikipedia.org
40. BIOSen.wikipedia.org
41. Phoenix Technologiesen.wikipedia.org
42. Send in the Clones - CHM Revolutioncomputerhistory.org
43. Phoenix Technologies Produces the First Commercially ...historyofinformation.com
44. Early memory testimage-ppubs.uspto.gov
45. Attack of the Clones - This Day in Tech Historythisdayintechhistory.com
46. Of Mice and Modes: The Story of the Graphical User Interface - IEEE Spectrumspectrum.ieee.org
47. How the Graphical User Interface Was Invented - IEEE Spectrumspectrum.ieee.org
48. Xerox Alto Computer with Mouse, GUI, and Ethernet Introduced in 1973cedmagic.com
49. Xerox PARC is Foundedhistoryofinformation.com
50. How Xerox Inspired Apple and Microsoftdopartners.be
51. GUI: the computer game changerarlingtonmnnews.com
52. Apple Lisaen.wikipedia.org
53. The Lisa: Apple's Most Influential Failure - CHMcomputerhistory.org
54. Macintosh 128Ken.wikipedia.org
55. Why the Apple Lisa Failed Despite Inventing the Future — Apple Scoopapplescoop.org
56. Apple Lisa - Mac Historymac-history.net
57. The Lisa - MacStoriesmacstories.net
58. Timeline of Microsoften.wikipedia.org
59. The Apple vs. Microsoft GUI Lawsuit - Low End MacLow End Maclowendmac.com
60. Apple v. Microsoft: The End at Last?freibrunlaw.com
61. Laws Violated: Apple Vs Microsoftlawshun.com
62. Apple v. Microsoftitlaw.fandom.com
63. OPEN LOOKen.wikipedia.org
64. Stored program paradigmuvm.edu
65. Von Neumann architectureen.wikipedia.org
66. ATXen.wikipedia.org
67. Motherboard form factoren.wikipedia.org
68. kindatechnical()kindatechnical.com
69. Difference Between RAM, Cache, and Registers: The Memory Hierarchy Showdownvalleyai.net
70. Unit 3.3: The Computer Memory Hierarchyarjankc.com.np
71. Douglas Engelbartbritannica.com
72. The computer mouse and interactive computing - SRIsri.com
73. The Keyset Was Designed to Make Computing More Powerful. Apple’s User-Friendly Mouse Made It Obsolete.slate.com
74. History of the floppy disken.wikipedia.org
75. The History of Computer Storage Devicesrecordnations.com
76. What is an Expansion Slot and How Does It Work?lenovo.com
77. Laptop Upgrade Planning Guide 2026: What to Check Before Buyingnewegg.com
78. A Brief History of Keyboards & the Birth of the Mouse - Computer Historytomshardware.com
79. The Mouse - CHM Revolution - Computer History Museumcomputerhistory.org
80. The Mother of All Demosinvention.si.edu
81. Why Are Laptops Moving to Soldered RAM? - Slashdothardware.slashdot.org
82. Laptop Soldered RAM — What It Means and Can You Upgrade?computercompatibility.com
83. Computer Data Storage Historycomputerhope.com
84. The Evolution of Storage Deviceslinkedin.com
85. Evolution of Data Storageplatinumdatarecovery.com
86. Desktop OS Market Share By Country Statistics 2026commandlinux.com
87. Mac vs Windows Market Share 2026: Usage, Revenue, and Segment Data - Axis Intelligenceaxis-intelligence.com
88. Usage share of operating systemsen.wikipedia.org
89. Decoding UEFI Firmwareuefi.org
90. Unified Extensible Firmware Interfacesimple.wikipedia.org
91. UEFI firmware requirementslearn.microsoft.com
92. Windows Market Share Statistics 2026: Global OS Share, Desktop Dominance, Windows 11 Adoption, and Regional Trends -requirementspc.com
93. Desktop Operating System Market Share Worldwidegs.statcounter.com
94. Goodbye BIOS, hello UEFIcomputerworld.com
95. What Is UEFI? UEFI vs BIOS for IT Teams - Flamingoflamingo.run
96. BIOS vs UEFI: Which Firmware Is Better for Modern PCs?superops.com
97. Worldwide PC market update Q2 2026: Omdia: overall decline 4%, Apple up 16%. IDC: overall decline of 4.9%, Apple up 10%asymco.com
98. IDC - WW Personal Computing Device Market: Q1 2026 Performance & 2026 Outlookidc.com
99. Chromebook Market Share By Country Statistics 2026commandlinux.com
100. Chromebook Market Share Statistics 2026: Shipments, Education Adoption, Growth Trendsfosspost.org
101. Chromebook Detachable VS Clamshell Sales Statistics 2026aboutchromebooks.com
102. Chromebook Market Size to Hit USD 23.79 Billion by 2034growthmarketreports.com
103. Chromebook Market Size, Trends, Share Analysis & Research Report, 2031mordorintelligence.com
104. Chromebook Market Size, Share, and Future Scope 2026-2033skyquestt.com
105. Global Chromebook Market Size, Trends, Share 2025 - 2034custommarketinsights.com
106. Intel 8086en.wikipedia.org
107. x86: History, what is it, and everything else you need to knowxda-developers.com
108. How is AMD allowed to make x86 microprocessors? Doesn't Intel own x86? Do AMD and other companies pay Intel royalty for the licence? - Quoraquora.com
109. CPU Architecture Explained: ARM, x86, Intel vs AMD — Full Guidemaxtangpc.com
110. Why does Intel control x86?forums.tomshardware.com
111. PC Market Size, Growth Trends & Industry Analysis, 2031mordorintelligence.com
112. Arm Dominates Laptops in 2026 — the Next Generation of Chips Has Arrived - Nextage Blognextage.com.br
113. Arm-based Processor Market Size, Share & Forecast Report 2025 to 2034dataintelo.com
114. NVIDIA Joins Windows on Arm, Driving Arm-Based AI Notebook Share to 34.2% by 2029 :: I-Connect007iconnect007.com
115. Mac processor comparison: Apple Silicon vs Intelmacworld.com
116. Apple Intel Transition: From x86 to Apple Silicon Integration - AppleMagazineapplemagazine.com
117. Today in Apple history: Apple chooses Intel over PowerPCcultofmac.com
118. The Evolution of Apple MacBooks: From Intel to Apple Siliconlaptopoutlet.co.uk
119. Apple A12Xen.wikipedia.org
120. The thrill of Apple silicon is already gonemacworld.com
121. Apple Silicon: The Complete Guide - MacRumorsmacrumors.com
122. Windows On Arm In 2026: The Laptop Architecture Shifteditorialge.com
123. A Brief History of CPUs: 31 Awesome Years of x86pcgamer.com
124. ARM CPUs predicted to hit 30% PC market share by 2026windowscentral.com
125. Modems: 60 Years of Hooking Uppcworld.com
126. Modemen.wikipedia.org
127. From dial-up to 5G: a complete guide to logging on to the internetqz.com
128. When did you first use the Internet at home (dial-up, DSL, fiber)? - Quoraquora.com
129. Dual WiMAX radio modemimage-ppubs.uspto.gov
130. The TCP/IP Guide - DNS Overview, History and Standardstcpipguide.com
131. Domain Name Systemen.wikipedia.org
132. History of DNSsimson.net
133. Owning Your Home Network: Router Security Revisitedarxiv.org
134. Your broadband router is not as secure as you think it istheconversation.com
135. Outdated WiFi routers may pose a huge security risk to millions of peoplezmescience.com
136. Home Network Threats: The Biggest Risks Hiding Inside Your Wi‑Fi - Fingfing.com
137. Outdated routers expose millions of broadband customers to security risksmoneysupermarket.com
138. Domain Name System (DNS) Historylivinginternet.com
139. An Overview of TCP/IP Protocols and the Internetgarykessler.net
140. Smart routerimage-ppubs.uspto.gov
141. The decline of the PC industrystatista.com
142. Personal Computer Market Share: 1975-2004 - Pegasus3Dpegasus3d.com
143. Global PC Shipments Decline By 100 Million Unitselectronicsforyou.biz
144. Worldwide PC Shipments by Top Vendors by Year - Dazeinfodazeinfo.com
145. Global PC Shipments Surpassed 275 Million Units in 2020databoks.katadata.co.id
146. www.businesswire.combusinesswire.com
147. Gartner Says Worldwide PC Shipments Increased 9.3% in Fourth Quarter of 2025 and 9.1% for the Full Yeargartner.com
148. More than 4% PC shipment growth predicted for 2025, but not for what you expect, says IDCcomputerworld.com
149. Windows 10 EOL 2026: The Billion-PC Refresh, By The Numbersservnetuk.com
150. Windows 10 end-of-support boosts global PC shipment growthciodive.com
151. Half of PCs still run Windows 10 despite looming end of support, PC makers saycomputerworld.com
152. Global PC shipments increase 8.9% - Taipei Timestaipeitimes.com
153. IDC - PC Market 2026: Memory Shortage Drives 11% Shipment Dropidc.com
154. TOP 20 PERSONAL COMPUTER MARKETING STATISTICS 2025amraandelma.com
155. US PC sales grow by 3% in late 2025 as companies and consumers scrambled to replace Windows 10 — 2026 forecast sees 13% drop as storage and memory prices expected to climb another 60%tomshardware.com
156. VisiCalcbritannica.com
157. (1979) VisiCalc: The First Spreadsheet Programlinkedin.com
158. VisiCalc (1979) (Software Arts) : Free Borrow & Streaming : Internet Archivearchive.org
159. The Rise and Fall of VisiCalc: How the First Spreadsheet Software Drove the PC Revolution - History Toolshistorytools.org
160. Personal computer gameusmodernculture.fandom.com
161. The History of the Gaming PCstyletech.net
162. graphsearch.epfl.chgraphsearch.epfl.ch
163. 1 A Brief History of Computer Games Mark Overmars, January 30, 2012stichtingspel.org
164. Harvard homework assignment led to creation of first computerized spreadsheetfiftyplusadvocate.com
165. How spreadsheets changed decision-makingtechnopran.com
166. The Complete History of Video Games 1952 - 2026blog.udonis.co
167. Falling Through the Net: Part I -- Household Accessntia.gov
168. The Digital Divide: Bridging the Divide Naturallybrookings.edu
169. The digital divide continues in the US as lower-income American fall behindweforum.org
170. Digital Divide Persists even as Lower-Income Americans Make Gains in Tech Adoption - Urbanism Nexturbanismnext.org
171. Computer Ownership and the Digital Dividebenton.org
172. Computing Solutions for Education in the Third Worldcs.purdue.edu
173. A socioeconomic related 'digital divide' exists in how, not if, young people use computersncbi.nlm.nih.gov
174. One Laptop per Child: Lessons from a long-term follow upvoxdev.org
175. Developers unveil low-cost computer for third-world nations - EE Timeseetimes.com
176. AMD discontinues low-cost PCnbcnews.com
177. Affordable Computing for Schools in Developing Countries Page 1 of 34vitalwave.com
178. Get Computers - World Computer Exchangeworldcomputerexchange.org
179. Information Societies and Digital Dividesarxiv.org
180. An Evaluation of the Impact of 1:1 Laptops on Student Attainment in Senior High School Sciences: International Journal of Science Education: Vol 37 , No 2 - Get Accesstandfonline.com
181. (PDF) Learning With Technology: The Impact of Laptop Use on Student Achievementresearchgate.net
182. IDBiadb.org
183. Computers and students’ achievement: An analysis of the One Laptop per Child program in Catalonia - ScienceDirectsciencedirect.com
184. Well-being and technology: The effect of individual laptops in high school - ScienceDirectsciencedirect.com
185. Household computer and Internet access: The digital divide in a pediatric clinic population - PubMedpubmed.ncbi.nlm.nih.gov
186. Computer Virus Historycomputerhope.com
187. Studying and Classification of the Most Significant Malicious Softwarearxiv.org
188. Elk Cloneren.wikipedia.org
189. UEFI Secure Boot Customizationmedia.defense.gov
190. BOOT SECURITY MODES AND RECOMMENDATIONSmedia.defense.gov
191. How to fix Windows 11 when Secure Boot and TPM don't worktechtarget.com
192. Secure Boot Explained — Every system boot is a negotiation of trust.medium.com
193. Secure the Windows boot processlearn.microsoft.com
194. 2026 Ransomware Report: 7,551 Victims, Up 24.9% - Black Kiteblackkite.com
195. Trusted Platform Module (TPM) 2.0 and Secure Boot for Call of Dutysupport.activision.com
196. Malware Detection and Prevention using Artificial Intelligence Techniquesarxiv.org
197. The History of Malwareibm.com
198. The history of computer viruses: Journey back to where it all began!dig.watch
199. The Early Days of Computer Viruses: A Simple Tale of Digital Troublemedium.com
200. UEFI and Secure Boot Explained: What They Do and Why They Matter - The Infobitstheinfobits.com
201. Ransomware attacks jump 29% to hit new quarterly high in Q3chainstoreage.com
202. Ransomware statistics 2026: confirmed attacks by monthransomnews.com
203. Ransomware Trends 2026: AI Attacks & Defense Strategiesadaptivesecurity.com
204. You Need To Avoid This Huge RAM Problem When Buying Your Next Laptop - BGRbgr.com
205. Soldered Components vs Repairability: Pros, Cons and Right to Repair Impactbiwintech.com
206. Upgradable RAM is overrated: Why I finally made peace with soldered memorytech.yahoo.com
207. Soldered RAM Explained — Why You Can't Upgrade Some Laptopscomputercompatibility.com
208. The real reason so many laptops have moved to soldered RAM - Digital Trendsdigitaltrends.com
209. Right to repairen.wikipedia.org
210. Right to repair: New consumer rights for easy and attractive repairs - European Commissioncommission.europa.eu
211. Does the EU Directive 2024/1799 (‘Right to Repair’) also apply to laptops?help.xmg.gg
212. What to Expect from Right to Repair in 2024 - iFixitifixit.com
213. Did EU ‘right to repair’ law force Apple to finally make a repairable Macbook?euobserver.com
214. New EU ‘right to repair’ laws require technology to last for a decadenews.yahoo.com
215. How to Sustainably Dispose of Your Technological Wastesustainability.yale.edu
216. PCLiquidations: A Helpful Guide to Electronic Waste Recyclingpcliquidations.com
217. Poison PCS and Toxic Californians Agalnst Wasteiwaste.epa.gov
218. Electronic wasteen.wikipedia.org
219. What Happens to Old Electronics After Recycling 2026 Guidepctechmag.com
220. Can You Upgrade Soldered Laptop RAM? What’s Really Possibleitechguides.com
221. Can I Upgrade a Laptop with Soldered RAM? Models & Alternatives Explainedcomputercompatibility.com
222. Energy Savings from Energy Star Personal Computer Systemsaceee.org
223. Recommendations for Tier I ENERGY STAR® Computer Specification Austin, TXenergystar.gov
224. ENERGY STAR* Version 5.0 System Implementation Whitepaperenergystar.gov
225. Rationale for Data Analysis of Idle Power Levels for the ...energystar.gov
226. Environmental and Economic Assessment of Desktop vs. Laptop Computers: A Life Cycle Approachmdpi.com
227. Sustainable Hardware Specializationarxiv.org
228. The Dirty Secret of SSDs: Embodied Carbonarxiv.org
229. The Global E-waste Monitor 2024itu.int
230. (PDF) GEM 2024 18-03 web page per page webresearchgate.net
231. Global e-Waste Monitor 2024: Electronic Waste Rising Five Times Faster than Documented E-waste Recyclingunitar.org
232. Environmental Impact of a Refurbished PCrootrefurb.co.uk
233. IoT System Case Study: Personal Office Energy Monitor (POEM)arxiv.org
234. The Carbon Impact of Manufactured VS Refurbished Computers - Get Online @ Homegetonlineathome.org
235. E-Waste On The Rise Worldwide - ChemistryViewschemistryviews.org
236. NPU Explained: What a Neural Processing Unit Is, How It Differs From a CPU and GPUguptadeepak.com
237. AI PC NPU Dashboard: Check Your Copilot+ TOPS Rating (2026) - SolidAITechsolidaitech.com
238. What Is An NPU? How On-Device AI Chips Work, Explained - TechDogstechdogs.com
239. Gartner Says AI PCs Will Represent 31% of Worldwide PC Market by the End of 2025gartner.com
240. AI Advanced PCs to Surpass Half of Global Shipments in 2026 as NPU-powered Laptops Go Mainstreamcounterpointresearch.com
241. AI PCs Have Arrived: Shipments Rising, Competition Heating Upio-fund.com
242. AI PCs to make up 31% of global shipments by end of 2025itbrief.asia
243. Network Edge Inference for Large Language Models: Principles, Techniques, and Opportunitiesdl.acm.org
244. On-Device LLM Inferenceemergentmind.com
245. How to Run LLMs Locally: A Complete Guide (2026) - Xybrid Blogxybrid.ai
246. On-Device LLMs: Complete Guide to Local AIpicovoice.ai
247. Sustainability Is Not Linear: Quantifying Performance, Energy, and Privacy Trade-offs in On-Device Intelligencearxiv.org
248. AI Hardware Market Share Statistics 2026aboutchromebooks.com
249. AI PC penetration rate broke 50% for the first time in 2026:143 million units of replacement tide, what exactly is the replacement?accpc.com.cn
250. AI PC Statistics By Usage, Shipment, Market Size and Facts (2025)electroiq.com
251. What Is TOPS in AI? NPU Performance and AI PC Specs Explainedstore.kamrui.com
252. Users will soon opt first for smartphones and tablets, not laptops and PCscomputerworld.com
253. Tablet Shipments Are On the Decline for the 10th Quarter in a Rowdigitaltrends.com
254. PC shipments grew in Q1, but questions remain - Resource Recyclingresource-recycling.com
255. Gartner Says Surging Memory Costs Will Reduce Global PC and Smartphone Shipments in 2026gartner.com
256. Windows 10 ESU Costs Double in 2026 as Memory Shortages Reshape PC Refresh Planswindowsforum.com
257. Omdia: US PC shipments grew 1.0% in 2Q26, while full-year market forecast to decline 10.7%omdia.tech.informa.com
258. AI PCs to surge, claiming over half the market by 2026computerworld.com