The History of Computing: From Abacus to Artificial Intelligence

The history of computing is the history of humanity's effort to automate thought — to build machines that could carry out calculations faster, more accurately, and eventually more broadly than any human mind. This history stretches from the earliest counting devices of ancient Mesopotamia to the quantum processors and large language models of the twenty-first century, passing through mechanical calculators, vacuum tube mainframes, transistor-based minicomputers, the personal computer revolution, the internet, and the emergence of artificial intelligence as a transformative technological force.

The story of computing is not simply an American story, though the United States played a dominant role in the industry's twentieth-century development. It is a global story shaped by British mathematicians, German engineers, Japanese semiconductor manufacturers, Indian software developers, Chinese hardware assemblers, and researchers and entrepreneurs from dozens of countries who collectively built the information infrastructure of the modern world.

The Ancient and Medieval Roots of Computation

The impulse to mechanize counting is ancient. The abacus — a frame with sliding beads used for arithmetic — was used in ancient Mesopotamia as early as 2700 to 2300 BCE and in China (the suanpan) from at least the second century BCE. The abacus spread across the ancient world and remained in widespread commercial use in China, Japan (the soroban), Russia (the schoty), and the Middle East well into the modern era; Japanese soroban experts competed successfully with early electronic calculators in speed demonstrations through the 1950s and 1960s.

The Antikythera mechanism — recovered from a Greek shipwreck in the Aegean Sea in 1901 and dated to approximately 100 BCE — is the oldest known complex mechanical device and the first known analog computer. The bronze mechanism used a sophisticated system of interlocking gears to predict astronomical positions, eclipse cycles, and Olympic Games dates. Its existence demonstrates that ancient Greek engineers achieved a level of mechanical sophistication not matched again in the Western world for over a thousand years.

Medieval Islamic scholars made foundational contributions to the mathematics underlying modern computing. Muhammad ibn Musa al-Khwarizmi, a ninth-century mathematician working in Baghdad at the House of Wisdom under Caliph al-Ma'mun, developed the systematic procedures for solving linear and quadratic equations that he called al-jabr (from which the word algebra derives). His name was Latinized as Algorismus, giving the English language the word algorithm — the concept of a step-by-step procedure for solving a problem that lies at the heart of all computer programming.

The Mechanical Calculator Era

The seventeenth century saw the first mechanical calculating machines capable of performing arithmetic operations automatically. Wilhelm Schickard, a German professor at the University of Tübingen, designed what is believed to be the first mechanical calculator in 1623 — a device he described in letters to the astronomer Johannes Kepler that used rotating wheels to perform addition and carry operations. Schickard's machine was apparently built but lost in the disruptions of the Thirty Years' War.

Blaise Pascal, the French mathematician and philosopher, built the Pascaline in 1642 — a mechanical adding machine using interlocking toothed wheels that he designed to help his father with tax calculations. Pascal constructed approximately fifty Pascalines, some of which survive in museums, making his device the first calculating machine to be manufactured and sold commercially.

Gottfried Wilhelm Leibniz, the German mathematician and philosopher who independently developed calculus alongside Isaac Newton, built the Stepped Reckoner in 1673 — a mechanical calculator that could perform multiplication and division as well as addition and subtraction. Leibniz also conceived the binary number system (representing all numbers using only 0 and 1), publishing his ideas in 1703. The binary system, which Leibniz saw as having philosophical and theological significance (representing creation from nothing), became the mathematical foundation of all digital computers two and a half centuries later.

Charles Babbage and the Conception of the Programmable Computer

The conceptual leap from mechanical calculator — a device that performs a fixed set of operations — to programmable computer — a device that can be instructed to perform any sequence of operations — was made by Charles Babbage, an English mathematician and inventor who lived from 1791 to 1871.

Babbage conceived of two extraordinary machines. The Difference Engine, designed in 1822, was a specialized calculator using the method of finite differences to compute polynomial functions automatically — intended for generating mathematical tables for navigation and astronomy. Babbage began construction of the Difference Engine with government funding but never completed it due to engineering disputes, cost overruns, and his growing interest in a more ambitious project.

The Analytical Engine, which Babbage designed from 1837 onward and worked on for the rest of his life, was the first design for a general-purpose programmable mechanical computer. It incorporated an arithmetic logic unit (which Babbage called the Mill), a memory store (which he called the Store), and the ability to be controlled by punched cards — inspired by the Jacquard loom's use of punched cards to control textile patterns. The Analytical Engine would have been programmed using sequences of punched cards, with conditional branching (the ability to take different actions depending on results of calculations) and loops. Babbage never built the Analytical Engine, but his detailed designs were visionary.

Ada Lovelace — the daughter of poet Lord Byron, educated in mathematics at her mother's insistence, and a close intellectual collaborator of Babbage — wrote an extensive commentary on the Analytical Engine in 1843 that included what is recognized as the first computer program: a detailed algorithm for computing Bernoulli numbers using the Analytical Engine. Lovelace's notes articulated, with remarkable clarity, the concept that the Analytical Engine could manipulate symbols beyond numbers — that it was not merely a calculator but a general-purpose symbol-manipulating machine. This insight anticipated the theoretical foundations of modern computer science by nearly a century.

The Electromechanical Era: Punched Cards and Tabulating Machines

The development of practical computing technology in the late nineteenth century was driven in significant part by the need to process the enormous volumes of data generated by modern industrial society — census records, business accounts, railroad timetables, and government statistics that overwhelmed the capacity of human clerks.

Herman Hollerith, an American statistician who had worked on the 1880 US Census, developed a system of punched cards and electromechanical tabulating machines to process the 1890 Census. Hollerith's system encoded information about each census respondent in patterns of holes punched in cards that could be read by electrical sensing pins, with the results tabulated by counters and sorted into bins automatically. The 1890 Census was processed in approximately two years using Hollerith's machines — compared to the eight years required to process the 1880 Census by hand.

Hollerith founded the Tabulating Machine Company in 1896, which through a series of mergers became the Computing-Tabulating-Recording Company in 1911 and subsequently the International Business Machines Corporation (IBM) in 1924, under the leadership of Thomas J. Watson Sr. IBM would dominate the computing industry for much of the twentieth century.

The punched card became a fundamental medium of data processing and programming throughout the early computer era, remaining in use from the 1890s well into the 1970s and in isolated applications even longer. The eighty-column IBM card format, standardized in 1928, became the universal standard for data exchange in an era when programs and data were literally physical cards that programmers carried in boxes and guarded from damage.

World War II and the Birth of the Electronic Computer

The Second World War (1939-1945) was the crucible in which electronic digital computing was born, driven by urgent military needs for codebreaking, ballistic calculations, and scientific computation. The war produced several landmark computing machines in Britain, Germany, and the United States, each representing different approaches to the fundamental problem of automatic computation.

In Britain, the need to decode German military communications encrypted by the Enigma machine and the more complex Lorenz cipher drove the development of code-breaking computing at Bletchley Park, the British signals intelligence center in Buckinghamshire. Alan Turing, a Cambridge mathematician who had published the theoretical foundations of computation in his 1936 paper "On Computable Numbers" (introducing the concept of the Turing Machine — an abstract model of a computing device that could simulate any algorithm), played a central role in breaking the Enigma cipher and in designing the Bombe, an electromechanical device that systematically tested Enigma cipher settings.

The Colossus — developed at Bletchley Park from 1943 onward by engineer Tommy Flowers and his team at the Post Office Research Station at Dollis Hill — was the world's first programmable electronic digital computer. Colossus used vacuum tubes (thermionic valves) rather than mechanical relays for its logic operations, making it orders of magnitude faster than electromechanical alternatives. Ten Colossus machines were in operation by the end of the war, processing intercepted Lorenz-encrypted German High Command communications. The existence of Colossus was classified by the British government until 1975, significantly affecting later historical accounts of who built the first electronic computer.

In the United States, two major computing projects emerged from wartime needs. The ENIAC (Electronic Numerical Integrator and Computer), built at the University of Pennsylvania by engineers J. Presper Eckert and John Mauchly and first operational in November 1945, was designed to calculate artillery firing tables for the US Army. ENIAC used approximately eighteen thousand vacuum tubes, occupied a room approximately twenty-seven meters long and three meters tall, consumed approximately one hundred and fifty kilowatts of power, and could perform approximately five thousand additions per second — roughly a thousand times faster than electromechanical computing devices. ENIAC was programmed by physical rewiring of its circuits — a laborious process that could take days. Six women — Kay McNulty, Betty Jennings, Betty Snyder, Marlyn Meltzer, Fran Bilas, and Ruth Teitelbaum — were among its primary programmers and became known as the ENIAC programmers.

In Germany, Konrad Zuse, a civil engineering student working independently in his parents' living room and later with limited support from the German aircraft industry, built a series of binary mechanical and relay-based computers during the 1930s and 1940s. Zuse's Z3, completed in May 1941, was the world's first functional programmable fully automatic digital computer — using relay switches rather than vacuum tubes but incorporating all the key elements of a modern computer including floating-point arithmetic and program control via punched film tape. Zuse also developed the first programming language specifically designed for computers — Plankalkül (Plan Calculus), designed in 1945 though not implemented for decades. Germany's defeat ended Zuse's wartime computing work, but he subsequently founded the first commercial computer company in Germany.

The Stored-Program Concept and Early Computer Architecture

The conceptual breakthrough that transformed the early electronic computers from specialized calculating devices into general-purpose machines was the stored-program concept: the idea of storing both program instructions and data in the same memory, allowing the program to be modified without physically rewiring the computer.

The theoretical basis for stored-program computers was articulated in several documents from 1944 to 1946. John von Neumann, the Hungarian-American mathematician, produced a draft report in June 1945 — the "First Draft of a Report on the EDVAC" — that described a computer architecture with a central processing unit, memory for both instructions and data, input and output devices, and the ability to modify programs during execution. This architecture became known as the "von Neumann architecture" and remains the dominant design for general-purpose computers to this day, despite von Neumann's draft having been partly drawn from ideas developed collaboratively with Eckert and Mauchly.

The dispute over intellectual priority for the stored-program concept involved multiple parties: von Neumann's report circulated widely and established his name as synonymous with the architecture; J. Presper Eckert and John Mauchly claimed they had discussed the stored-program concept with von Neumann before his report; and Alan Turing had described a universal stored-program computer in theoretical terms in his 1936 paper. The British computing pioneers Freddie Williams and Tom Kilburn at the University of Manchester built the first working stored-program computer — the Manchester Small-Scale Experimental Machine (SSEM), known as the "Baby" — which ran its first program on June 21, 1948.

The late 1940s and early 1950s saw the construction of stored-program computers at universities and research institutions on both sides of the Atlantic. Cambridge University's EDSAC (Electronic Delay Storage Automatic Calculator), designed by Maurice Wilkes and operational from May 1949, was the first practical stored-program computer to run useful programs. Princeton's IAS machine, designed by von Neumann's team, was completed in 1952 and became a template copied by computing projects at the Australian National University (SILLIAC), the University of Illinois (ILLIAC), and elsewhere.

Transistors and the Second Generation: the 1950s and 1960s

The vacuum tube computers of the first generation were expensive, unreliable, power-hungry, and physically enormous. The invention of the transistor at Bell Laboratories in December 1947 by William Shockley, John Bardeen, and Walter Brattain — for which they shared the Nobel Prize in Physics in 1956 — offered a revolutionary alternative: a solid-state switching device far smaller, more reliable, cooler, and more efficient than the vacuum tube.

The transistor is a semiconductor device (typically made from silicon or germanium) that can amplify or switch electrical signals. A transistor operates as a switch by using a small input current or voltage at one terminal to control a larger current flowing between two other terminals — the same function performed by the vacuum tube triode, but without the hot cathode, vacuum envelope, glass housing, warm-up time, or fragility of the tube.

Commercial transistor computers began appearing in the late 1950s. IBM's 7090, introduced in 1959, was a fully transistorized successor to the vacuum tube 704, used for scientific and engineering computation. The transistor-based IBM 1401, introduced in 1959, became one of the most widely sold computers in history — providing business computing power to thousands of companies that could not afford the larger mainframes. By the early 1960s, vacuum tube computers were obsolete, and the second generation of transistor-based computers dominated the industry.

The Soviet Union developed a substantial indigenous computer industry during this period, largely parallel to and independent of Western developments due to Cold War technology restrictions. The MESM (Small Electronic Computing Machine), built by a team led by Sergei Lebedev at the Institute of Electric Engineering in Kiev, Ukraine and first operational in November 1950, was among the first operational stored-program computers in continental Europe. The subsequent BESM series (Large Electronic Computing Machine) and the Ural, Minsk, and STRELA computer families established Soviet computing as a significant research and industrial force through the 1950s and 1960s, though Cold War isolation ultimately limited the Soviet industry's ability to keep pace with Western commercial development.

Japan entered the computing industry in the late 1950s, with major Japanese electronics companies including Fujitsu, NEC, Hitachi, and Toshiba developing domestic computer products initially based on licenses from Western companies and subsequently on independent Japanese designs. The Japanese government's Ministry of International Trade and Industry (MITI) played an active role in coordinating the development of a domestic computer industry as part of Japan's broader industrial strategy.

The Integrated Circuit and Moore's Law

The transistor replaced the vacuum tube, but transistor computers were still built by manually soldering individual components — transistors, resistors, capacitors, and connecting wires — to circuit boards. Each connection was a potential point of failure, and the density and speed of circuits were limited by the physical dimensions of discrete components.

The integrated circuit — placing multiple transistors and their interconnections on a single piece of semiconductor material — was invented independently and nearly simultaneously in 1958-1959 by two engineers: Jack Kilby at Texas Instruments, who demonstrated the first working integrated circuit in September 1958 using germanium; and Robert Noyce at Fairchild Semiconductor, who developed the planar process for fabricating silicon integrated circuits in 1959, a more manufacturable approach that became the foundation of the modern semiconductor industry. Both men are credited as co-inventors of the integrated circuit; Kilby received the Nobel Prize in Physics in 2000 (Noyce had died in 1990 and was thus ineligible).

The planar process, developed at Fairchild by Jean Hoerni, used photolithography — the same technique used to print photographs — to define circuit patterns on a flat silicon wafer, then deposited or etched materials to create transistors and their connections. This process allowed many transistors to be fabricated simultaneously on a single wafer, and allowed circuit patterns to be shrunk by simply reducing the size of the photographic masks used to define them. As manufacturing processes improved, more transistors could be placed on a chip, increasing circuit density and reducing cost per transistor.

Gordon Moore, a co-founder of Fairchild Semiconductor and later Intel, published an observation in 1965 that has become known as Moore's Law: the number of transistors on an integrated circuit was doubling approximately every two years (originally he said every year), while the cost per transistor fell proportionally. Moore's observation, which he presented as an extrapolation of empirical trends rather than a physical law, proved remarkably accurate for over five decades. The number of transistors on a microprocessor grew from a few dozen in the early 1960s to thousands in the early 1970s, millions in the 1990s, billions in the 2010s, and over one hundred billion in advanced chips of the early 2020s.

Moore's Law shaped the entire computing industry: semiconductor manufacturers invested enormous capital in ever-more-advanced fabrication plants, confident that miniaturization would continue; software developers could assume that hardware would be dramatically faster in two or three years; and the declining cost of computation enabled entirely new applications that would have been economically impossible with earlier hardware.

Fairchild Semiconductor was extraordinarily influential as a nursery for the semiconductor industry. Founded in 1957 by eight engineers who left Shockley Semiconductor Laboratory (founded by transistor co-inventor William Shockley) — the "Traitorous Eight" as Shockley called them, including Noyce and Moore — Fairchild trained many of the engineers and entrepreneurs who went on to found or lead other semiconductor companies. Silicon Valley, the region of Santa Clara County in California that became the global center of the semiconductor and technology industry, takes its name from the silicon-based semiconductor industry that Fairchild and its spinoffs established there.

Robert Noyce and Gordon Moore left Fairchild in 1968 to found Intel Corporation (from "Integrated Electronics"), recruited engineer Andy Grove, and directed the company toward a then-novel product: semiconductor memory chips. Intel's first product was the 3101 static random-access memory (SRAM) chip in 1969. The subsequent development of the microprocessor — a complete central processing unit on a single chip — would transform the computing industry and make Intel one of the most valuable companies in the world.

The Microprocessor Revolution

The microprocessor — integrating all the functions of a computer's central processing unit onto a single semiconductor chip — was the development that made personal computers possible and fundamentally changed the relationship between humans and computation.

Intel developed the first commercial microprocessor, the Intel 4004, in 1971. The 4004 was designed by Federico Faggin, Ted Hoff, and Stanley Mazor for a Japanese calculator company, Busicom. The 4004 contained 2,300 transistors on a single chip, processed four-bit data (sufficient for calculator operations), and ran at 740 kilohertz. Its successor, the Intel 8008 (1972) and then the Intel 8080 (1974), provided eight-bit processing and became the basis for early personal computers and embedded systems.

Texas Instruments and other companies also developed microprocessors in the early 1970s, but Intel's 8080 and the competing Motorola 6800 established the design patterns that would dominate personal computing. The Intel 8080 was the processor used in the Altair 8800, the computer kit introduced in January 1975 by MITS (Micro Instrumentation and Telemetry Systems) of Albuquerque, New Mexico, that is generally considered the first commercially successful personal computer. The Altair appeared on the cover of Popular Electronics magazine in January 1975 and generated overwhelming demand from hobbyists, triggering the personal computer revolution.

Two young programmers in Seattle — Paul Allen and Bill Gates — saw the Popular Electronics cover and contacted MITS to offer a BASIC programming language interpreter for the Altair. They formed Micro-Soft (later Microsoft) in 1975 to develop and sell software for microcomputers. The ability to program the Altair in BASIC — a relatively accessible high-level language — rather than in machine code was essential to its appeal and to the growth of the hobbyist computing movement.

The Homebrew Computer Club, a hobbyist meeting in the San Francisco Bay Area founded in 1975, became a gathering point for the engineers, programmers, and entrepreneurs who would build the personal computer industry. Among its members were Steve Wozniak and Steve Jobs, who founded Apple Computer in 1976 and introduced the Apple I (a single-board computer sold as a kit) in 1976 and the Apple II (a complete personal computer with color graphics and an expandable bus) in 1977.

The Apple II was one of three machines that arrived in 1977 that established the personal computer as a consumer product: alongside the Apple II were the Tandy TRS-80 (from Radio Shack, making computers available in electronics retail stores across America) and the Commodore PET. These machines moved personal computing from hobbyist kits requiring soldering to complete, usable systems with keyboards, displays, and the beginnings of practical software.

IBM, the dominant force in the existing computer industry, had watched the personal computer market develop and in 1980 decided to enter it. IBM's Personal Computer, introduced in August 1981, was revolutionary in one crucial respect: IBM's project team, led by Don Estridge, made the unusual decision to build the PC from standard, commercially available components (using the Intel 8088 processor and Microsoft's MS-DOS operating system) and to publish the technical specifications of the machine — creating an open architecture that any company could copy. IBM's decision, intended to accelerate product development, had the unintended consequence of creating the IBM PC-compatible ecosystem: dozens of companies (led by Compaq, which introduced the first IBM PC clone in 1982) built computers using the same architecture, creating a massive competitive market while the key intellectual property — the processor (Intel's) and the operating system (Microsoft's) — remained in the hands of IBM's suppliers.

The IBM PC and its clones established the Wintel architecture (Windows operating system, Intel processor) that dominated personal computing for the following three decades. Microsoft, which owned the MS-DOS and later Windows operating system, grew from a small Seattle software company to one of the world's most valuable corporations. Intel, which had negotiated the right to continue selling its processors to PC clone makers (IBM had assumed the clone market would be limited), grew its microprocessor business into the foundation of one of the world's most profitable and strategically important industries.

The Chip Manufacturing Industry: Design, Fabrication, and the Global Supply Chain

The semiconductor manufacturing industry that emerged from the integrated circuit revolution is among the most capital-intensive, technically demanding, and geographically concentrated industries ever created. The production of modern semiconductor chips requires mastery of materials science, photolithography, chemistry, vacuum technology, precision engineering, and computational design at scales and tolerances that have required decades of continuous improvement to achieve.

The fundamental process of semiconductor chip manufacturing — planar fabrication using photolithography — has remained conceptually similar since Noyce and Hoerni's work at Fairchild in the late 1950s, but the engineering refinements have been extraordinary. The feature size (the minimum dimension of a transistor or connecting wire) in the first integrated circuits was approximately twenty-five microns (micrometers). By 2024, leading-edge chips from Taiwan Semiconductor Manufacturing Company (TSMC) and Samsung were manufactured at process nodes below five nanometers (five billionths of a meter) — a roughly five-thousandfold reduction in feature size over sixty years.

Achieving these sub-nanometer dimensions requires extreme ultraviolet (EUV) lithography — a technology that uses light with a wavelength of thirteen nanometers (compared to visible light's four hundred to seven hundred nanometer wavelengths) to expose circuit patterns on silicon wafers. The only commercially available EUV lithography machines in the world are made by ASML, a Dutch company headquartered in Eindhoven. Each EUV machine costs approximately two hundred million dollars, weighs approximately one hundred and eighty tonnes, and contains over one hundred thousand components sourced from suppliers across the world. ASML's EUV technology, developed over more than two decades with partial funding from Intel and other chipmakers, represents one of the most concentrated technological monopolies in modern industry.

The semiconductor industry has bifurcated between "fabless" companies that design chips but do not manufacture them, and foundries (or "fabs") that manufacture chips designed by others. This model was pioneered by TSMC (Taiwan Semiconductor Manufacturing Company), founded by Morris Chang in Taiwan in 1987 with backing from the Taiwanese government and investment from Philips. TSMC's pure-play foundry model — manufacturing chips exclusively for others rather than competing with its own chip designs — attracted customers who were unwilling to share their designs with vertically integrated competitors like Intel or IBM.

By the early 2020s, TSMC manufactured chips for Apple (the A-series and M-series processors in iPhones, iPads, and Macs), Nvidia (graphics processors and AI accelerators), AMD (processors and GPUs), Qualcomm (smartphone processors), and hundreds of other companies. TSMC's dominance of leading-edge semiconductor manufacturing — the company accounted for the majority of the world's most advanced chip production — created extraordinary strategic concentration: the world's most sophisticated chips were manufactured in Taiwan, an island whose political status was disputed by China. The US CHIPS and Science Act (2022), providing approximately fifty-two billion dollars in subsidies for domestic semiconductor manufacturing, and equivalent programs in the EU, Japan, South Korea, and India, reflected the strategic risks of this concentration.

Intel, which had long been the world's leading semiconductor manufacturer by revenue and technology, struggled to maintain its manufacturing leadership in the 2010s. After repeated delays in transitioning to new manufacturing process nodes, Intel fell behind TSMC and Samsung in leading-edge manufacturing, forcing it to announce in 2021 that it would adopt the foundry model and have some of its products manufactured by TSMC. Intel's subsequent launch of Intel Foundry Services (IFS) and its investment in new fabrication plants in Ohio and Arizona represented an attempt to rebuild US semiconductor manufacturing capacity.

Samsung Electronics of South Korea is the only company other than TSMC that manufactures leading-edge chips for external customers, operating large-scale foundry operations alongside its own memory chip business and consumer electronics division. South Korea has developed one of the world's strongest semiconductor industries: Samsung and SK Hynix together dominate global DRAM and NAND flash memory chip production. China has invested heavily in catching up with US, Taiwanese, and South Korean semiconductor technology, but US export controls imposed from 2019 onward — restricting the sale of advanced semiconductors and semiconductor manufacturing equipment to Chinese companies — significantly constrained Chinese semiconductor development.

Operating Systems, Software, and the Graphical User Interface

The hardware of computing — processors, memory, and storage — is only useful when animated by software: the programs that direct the hardware to perform useful work. The development of operating systems, programming languages, and user interfaces constitutes a parallel history to the hardware story, shaping how humans interact with machines and what tasks computers can perform.

Early computers had no operating systems: each program had to manage all machine resources directly, and a human operator physically loaded programs and data. The development of batch processing systems in the 1950s — software that could automatically load and execute a sequence of jobs without operator intervention between each — was the first step toward the operating system concept. The IBM OS/360, introduced with the IBM System/360 family of mainframes in 1964, was one of the first comprehensive operating systems, managing multiple programs, peripheral devices, and memory allocation. Its development — requiring over a thousand programmers and thousands of person-years of effort — was the subject of Fred Brooks's influential 1975 book "The Mythical Man-Month," which documented the difficulties of large-scale software projects and introduced "Brooks's Law" (adding programmers to a late project makes it later).

Unix, developed at Bell Laboratories beginning in 1969 by Ken Thompson and Dennis Ritchie, became one of the most influential operating systems ever written. Unix introduced a hierarchical file system, the concept of software tools that could be combined using pipes, a consistent programming interface, and portability across different hardware platforms. The C programming language, developed by Dennis Ritchie at Bell Labs in 1972 primarily to implement Unix, became the most important systems programming language of the twentieth century — used to write operating systems, compilers, databases, and embedded software for decades. Unix spread widely through universities (where AT&T licensed it cheaply for academic use) and commercial systems. Its descendants include the BSD Unix variants, Linux, macOS (derived from BSD Unix via NeXT), and indirectly Android (which uses the Linux kernel).

The graphical user interface (GUI) — using a mouse to point and click on visual representations of files, programs, and controls rather than typing text commands — was developed at Xerox's Palo Alto Research Center (PARC) in the early 1970s. Xerox PARC, founded in 1970 and staffed with exceptional researchers recruited from universities and research labs, was one of the most productive research centers in computing history: its researchers invented or pioneered the GUI, the mouse, the desktop metaphor, the laser printer, object-oriented programming (Smalltalk), Ethernet local area networking, and the concept of what-you-see-is-what-you-get (WYSIWYG) document editing.

The Alto, a personal workstation developed at Xerox PARC in 1973, incorporated all of these innovations — a bit-mapped graphical display, a mouse, overlapping windows, icons, menus, and a network connection. It was not a commercial product, but its influence was extraordinary: Steve Jobs visited Xerox PARC in December 1979 (Xerox received Apple shares in exchange for the visit), saw the Alto's interface, and directed Apple to incorporate the GUI into the Lisa (1983) and Macintosh (1984) computers. Microsoft's Windows, first released in 1985 as a graphical interface layer on top of MS-DOS, drew on similar ideas and eventually became dominant after Windows 3.0 (1990) and especially Windows 95 (1995).

The Linux operating system, created by Finnish computer science student Linus Torvalds in 1991, developed as a free, open-source alternative to Unix. Torvalds announced his project on the Usenet newsgroup comp.os.minix in August 1991 and invited contributions from the programming community. The resulting collaborative development model — hundreds and later thousands of contributors improving a shared codebase — produced a remarkably capable operating system that came to dominate server computing, embedded systems, the Android mobile platform, and supercomputers. By 2024, all five hundred of the world's most powerful supercomputers ran Linux.

The Internet and the World Wide Web

The internet — the global network of networks that underlies modern digital communication — has its origins in ARPANET, a computer network funded by the US Defense Advanced Research Projects Agency (DARPA) that first became operational in October 1969 when messages were sent between computers at UCLA and the Stanford Research Institute.

ARPANET was designed around packet switching — a method of transmitting data developed independently by Paul Baran at RAND Corporation (1964) and Donald Davies at the UK National Physical Laboratory (1965). In packet switching, data is divided into small packets that travel independently through the network and are reassembled at the destination, rather than establishing a dedicated circuit between sender and receiver as telephone systems did. Packet switching proved far more robust and efficient for data communication than circuit switching.

The Transmission Control Protocol and Internet Protocol (TCP/IP), developed by Vint Cerf and Bob Kahn and published in 1974, provided the common language that allowed different types of computer networks to communicate — the essential protocol that made the internet an internet (a network of networks). The US Department of Defense adopted TCP/IP as the standard for ARPANET in January 1983, the date often cited as the birthday of the modern internet.

Through the 1970s and 1980s, the internet was primarily an academic and governmental network used for email, file transfer, and remote login. The democratization of internet access came through the commercial internet service providers that emerged in the early 1990s — CompuServe, AOL (America Online), and Prodigy offering dial-up internet access to home users — and through the development of the World Wide Web.

The World Wide Web was invented by Tim Berners-Lee, a British computer scientist working at CERN (the European Organization for Nuclear Research) in Geneva, Switzerland. Berners-Lee proposed a hypertext-based information system in 1989, developed the key technologies (HTML for formatting pages, HTTP for transferring them, and URLs for addressing them), and put the first website online on August 6, 1991. Berners-Lee made the Web's core technologies freely available without patent or royalty, a decision he has credited as essential to the Web's explosive adoption.

The Web browser — software for navigating the World Wide Web — was key to making the Web accessible to non-technical users. NCSA Mosaic, released in 1993 by a team including Marc Andreessen at the National Center for Supercomputing Applications at the University of Illinois, was the first browser to display images inline with text and to bring the Web to millions of ordinary users. Andreessen subsequently co-founded Netscape Communications, whose Navigator browser dominated the early commercial Web.

The growth of the commercial internet in the 1990s was extraordinary. Internet traffic roughly doubled every year through the 1990s. The dot-com boom of 1995-2000 attracted hundreds of billions of dollars of investment to internet companies, many of which failed in the subsequent crash of 2000-2002, but the survivors — including Amazon, Google, and eBay — emerged as dominant forces in the new digital economy. The internet transformed commerce, media, communication, and entertainment, and the question of how it should be governed became one of the central policy questions of the early twenty-first century.

Mobile Computing and the Smartphone Era

The convergence of wireless communication, miniaturized processors, and capacitive touchscreens in the early twenty-first century produced the smartphone — a pocket-sized computer connected to the internet that became the primary computing device for the majority of humanity.

The personal digital assistant (PDA) — handheld computers for organizing contacts, calendars, and notes — appeared in the early 1990s. Apple's Newton MessagePad (1993) was an early example, featuring handwriting recognition that worked poorly and became a cultural reference point for over-ambitious technology. Palm's PalmPilot (1996) succeeded by offering a simpler, more reliable device with a stylus-based interface that worked within the limitations of the hardware.

The mobile phone and the handheld computer converged in devices including the Nokia Communicator (1996), the Blackberry (1999, developed by the Canadian company Research in Motion), and early Windows Mobile phones. These devices required styluses or miniature keyboards and were primarily used by business customers.

Apple's iPhone, unveiled by Steve Jobs on January 9, 2007, at the Macworld Conference in San Francisco, was a transformative product that combined a phone, music player, and internet device in a single package with a large multi-touch screen that could be operated entirely by fingers. The iPhone introduced smartphone concepts that became industry standards: a large capacitive touchscreen, an app ecosystem (the App Store launched in 2008), mobile web browsing with pinch-to-zoom, visual voicemail, and deep integration with desktop software. The iPhone's success was followed by Google's Android operating system (first released in 2008 on the HTC Dream), which adopted a more open model allowing multiple phone manufacturers to use the platform.

The smartphone market became one of the largest and most competitive consumer electronics markets ever: Apple, Samsung, Huawei, Xiaomi, OPPO, and dozens of other manufacturers sold billions of devices annually. By the early 2020s, there were approximately six billion smartphone subscriptions globally — more than one for every adult on earth. The smartphone brought internet access, digital commerce, navigation, photography, and computing capability to populations in developing countries that had never had access to personal computers.

The History of Artificial Intelligence

Artificial intelligence — the attempt to create machines that can perform tasks that would, if performed by humans, require intelligence — has a history that both predates and is intertwined with the history of electronic computing. The ideas, methods, and ambitions of AI have evolved through cycles of optimism and disappointment, each wave of enthusiasm producing genuine advances but also unmet promises, leading to periods of reduced funding and interest known as "AI winters."

The conceptual roots of artificial intelligence extend to antiquity: myths of artificial beings with intelligence or consciousness appear in ancient Greek literature (Talos, the bronze giant of Crete; the bronze automata of Hephaestus), and medieval Jewish legend (the Golem of Prague). But the modern scientific approach to artificial intelligence begins with the question posed by Alan Turing in his 1950 paper "Computing Machinery and Intelligence": "Can machines think?" Turing proposed the Imitation Game (subsequently known as the Turing Test) as an operational definition: a machine could be considered intelligent if a human interrogator could not distinguish its responses from a human's through text-based conversation. The paper anticipated and refuted many standard objections to machine intelligence and established the philosophical framework for AI research for decades.

The formal founding of artificial intelligence as an academic field is conventionally dated to the Dartmouth Conference of the summer of 1956, organized by John McCarthy, Marvin Minsky, Nathaniel Rochester, and Claude Shannon. McCarthy, who coined the term "artificial intelligence" for the conference proposal, was a mathematician who had developed Lisp (a programming language for symbol manipulation that became the standard language for early AI research) and would go on to found the Stanford Artificial Intelligence Laboratory. The Dartmouth Conference brought together the founders of what became AI research and established the field's central problem: creating machines that could perform any intellectual task a human could perform.

The early years of AI research (1956 to approximately 1974) produced genuine accomplishments and extravagant optimism. The Logic Theorist, developed by Allen Newell and Herbert Simon at Carnegie Mellon University (then Carnegie Institute of Technology) in 1955-1956, proved theorems in symbolic logic and is sometimes considered the first true AI program. The General Problem Solver (1957), also by Newell and Simon, was an attempt to create a general-purpose problem-solving algorithm. Arthur Samuel's checkers program (1959), which learned to play checkers by playing games against itself, introduced the concept of machine learning — a program that improves its performance through experience rather than explicit programming.

The Marvin Minsky and Seymour Papert co-authored book "Perceptrons" (1969) critically analyzed the limitations of artificial neural networks — an approach to AI modeled on the structure of the brain — concluding that single-layer perceptrons could not solve certain classes of problems. This analysis, partly misunderstood as condemning neural network approaches more broadly, contributed to reduced funding for neural network research and redirected AI toward symbolic, rule-based approaches.

LISP-based expert systems — programs that encoded human expert knowledge as sets of rules — became the dominant AI approach of the 1980s. MYCIN (developed at Stanford in the early 1970s), which diagnosed bacterial infections and recommended antibiotic treatments, demonstrated that an AI system could match expert physician performance in a narrow domain. The commercialization of expert systems technology by companies including Symbolics, Lisp Machines Inc., and the "Fifth Generation Computer Systems" project launched by Japan's MITI in 1982 (aiming to build knowledge-based computing systems that would surpass Western AI) triggered significant investment and the creation of a commercial AI industry.

The collapse of the commercial expert systems market in the late 1980s — as the cost and difficulty of maintaining large rule bases became apparent, and specialized Lisp hardware proved unable to compete with rapidly improving conventional workstations — constituted the second AI winter (approximately 1987-1993), following an earlier period of reduced interest in the mid-1970s.

Machine Learning and the Neural Network Renaissance

The resurgence of neural network research and the development of modern machine learning represents the most consequential development in computing since the microprocessor.

Artificial neural networks — computing systems loosely modeled on the structure of biological neural networks, with layers of interconnected nodes (neurons) whose connection strengths (weights) are adjusted during training — had been studied since the 1940s. Frank Rosenblatt's perceptron (1957) was an early trainable neural network that could learn to classify simple patterns. However, as noted above, the limitations of shallow (single-layer) networks and the computational demands of training deeper networks limited progress through the 1970s and 1980s.

The backpropagation algorithm — a method for efficiently computing how the weights in a multi-layer neural network should be adjusted to reduce prediction errors — was known in the 1960s and early 1970s but was not widely applied to neural networks until a landmark 1986 paper by David Rumelhart, Geoffrey Hinton, and Ronald Williams, "Learning Representations by Back-propagating Errors," published in Nature. Backpropagation made it practical to train multi-layer (deep) neural networks, and the subsequent decade saw significant research activity in neural networks applied to pattern recognition, speech recognition, and natural language processing.

Geoffrey Hinton, a British-Canadian cognitive scientist and computer scientist at the University of Toronto (and later Google), became the central figure in the revival of neural network approaches to AI. Hinton, along with his students and collaborators, developed several key techniques including Boltzmann machines (1983), deep belief networks (2006), and contributed to the development of dropout regularization and other training methods that made deep neural networks practical.

Yann LeCun, a French computer scientist who worked at Bell Labs and later joined NYU and Facebook (Meta), developed convolutional neural networks (CNNs) optimized for image recognition, demonstrating in the early 1990s that CNNs could reliably recognize handwritten digits (used in bank check processing). LeCun's work on CNNs presaged the deep learning revolution in computer vision.

Yoshua Bengio at the University of Montreal was another central figure in developing the theoretical and practical foundations of deep learning. Hinton, LeCun, and Bengio shared the 2018 Turing Award — the highest honor in computer science — for their contributions to deep learning.

The breakthrough moment of the modern deep learning era is often identified as the 2012 ImageNet Large Scale Visual Recognition Challenge, when a deep convolutional neural network called AlexNet — developed by Alex Krizhevsky, Ilya Sutskever, and Geoffrey Hinton at the University of Toronto — achieved an error rate of fifteen point three percent on the image classification benchmark, compared to twenty-six percent for the second-place entry and dramatically better than all previous approaches. AlexNet was trained on two NVIDIA GTX 580 GPUs — a pivotal demonstration that graphics processing units (GPUs), originally designed for video game rendering, could dramatically accelerate deep learning training. Nvidia's CUDA parallel computing platform, released in 2007, had made GPU computing accessible to researchers; the AlexNet result made it essential.

The explosion of deep learning research and applications from 2012 onward was driven by three converging factors: the availability of large labeled datasets (ImageNet itself had over fourteen million labeled images), the availability of powerful GPU computing at reasonable cost, and the algorithmic advances in deep network training. Between 2012 and 2016, deep learning systems surpassed human-level performance on multiple standard benchmarks including image classification, speech recognition, and game playing.

Deep Learning Applications: from Games to Language Models

The most dramatic early demonstration of deep learning's capabilities in complex strategy came from DeepMind, a London-based AI research company founded in 2010 by Demis Hassabis, Shane Legg, and Mustafa Suleyman and acquired by Google in 2014 for approximately four hundred million pounds. DeepMind's AlphaGo program, which combined deep learning with Monte Carlo tree search, defeated Lee Sedol, one of the world's top Go players, four games to one in March 2016 — a milestone that arrived at least a decade earlier than most AI researchers had predicted, given the enormous complexity of the game of Go (which has more possible board positions than atoms in the observable universe).

AlphaGo's successor, AlphaZero (2017), taught itself to play Go, chess, and shogi at superhuman levels by playing against itself from random initial play, without any human game knowledge beyond the rules. AlphaZero's chess play was described by world champion Garry Kasparov as exhibiting a style of play unlike any previously seen. DeepMind subsequently applied similar reinforcement learning techniques to protein structure prediction: AlphaFold2 (2020) solved one of the most important problems in molecular biology — predicting the three-dimensional structure of a protein from its amino acid sequence — with accuracy comparable to experimental methods, winning the Critical Assessment of Protein Structure Prediction (CASP) competition decisively and earning the 2024 Nobel Prize in Chemistry for Demis Hassabis and John Jumper.

The development of transformer neural network architecture, introduced in the 2017 paper "Attention Is All You Need" by Vaswani, Shazeer, Parmar, Uszkoreit, Jones, Gomez, Kaiser, and Polosukhin (mostly Google Brain researchers), provided the foundation for large language models. The transformer's self-attention mechanism allowed neural networks to efficiently capture long-range dependencies in sequential data (text, speech, protein sequences), enabling the training of far larger and more capable language models than the recurrent neural networks previously used for natural language processing.

OpenAI, a non-profit AI safety organization founded in San Francisco in December 2015 by Elon Musk, Sam Altman, Greg Brockman, Ilya Sutskever, and others (with initial funding commitments of one billion dollars), developed the GPT (Generative Pre-trained Transformer) series of large language models. GPT-1 (2018), GPT-2 (2019), and GPT-3 (2020) showed that scaling transformer models — increasing the number of parameters and the quantity of training data — produced qualitative improvements in language understanding and generation. GPT-3, with one hundred and seventy-five billion parameters, demonstrated remarkable few-shot learning ability: given a few examples of a task in its input, it could perform the task on new examples without any additional training.

ChatGPT, launched by OpenAI in November 2022 as a conversational interface to a fine-tuned version of the GPT-3.5 model, reached one million users in five days and one hundred million users in two months — the fastest consumer product growth in history. ChatGPT demonstrated that large language models could carry on coherent, helpful, and often impressive conversations across a vast range of topics, and triggered a wave of investment in generative AI companies and a competitive response from Google (Bard/Gemini), Microsoft (Copilot, integrated into Bing and Office), Meta (LLaMA), Anthropic (Claude), and dozens of other companies and research organizations.

Quantum Computing: Principles and Progress

Quantum computing represents the most radical departure from the classical computing paradigm since the invention of the transistor — a fundamentally different approach to information processing that exploits quantum mechanical phenomena to perform certain computations exponentially faster than any classical computer could achieve.

Classical computers process information as bits — binary digits that are either 0 or 1. Quantum computers use quantum bits, or qubits, which can exist in a superposition of 0 and 1 simultaneously until measured. This property, along with quantum entanglement (correlations between qubits that have no classical analog) and quantum interference (the ability to manipulate probability amplitudes to enhance correct answers and cancel incorrect ones), allows quantum computers to explore vast computational spaces in parallel in ways that classical computers cannot efficiently simulate.

The theoretical foundations of quantum computing were laid in the early 1980s by physicist Richard Feynman, who observed at a 1981 conference that classical computers cannot efficiently simulate quantum mechanical systems, and proposed that quantum computers might be able to. David Deutsch at Oxford University published the first theoretical model of a quantum computer in 1985, describing a "universal quantum computer" analogous to Turing's universal classical computer.

The development of practical quantum algorithms in the 1990s demonstrated that quantum computers could solve specific important problems exponentially faster than classical alternatives. Peter Shor at Bell Labs developed an algorithm in 1994 that could factor large integers exponentially faster on a quantum computer than any known classical algorithm — a result of enormous cryptographic significance because the security of widely used public-key cryptography (RSA encryption) rests on the difficulty of factoring large numbers. Lov Grover at Bell Labs developed a quantum algorithm in 1996 that provides a quadratic speedup for searching unsorted databases. These theoretical results demonstrated that quantum computing was not merely academic — it had practical implications for security and optimization.

Building a practical quantum computer requires maintaining qubits in coherent quantum states long enough to perform computations — an extraordinarily difficult engineering challenge, because qubits are easily disturbed by any interaction with their environment (a process called decoherence). Various physical systems have been pursued as qubit platforms: superconducting circuits (cooled to near absolute zero), trapped ions, photonic qubits, nitrogen-vacancy centers in diamond, topological qubits, and others, each with different trade-offs between coherence time, gate fidelity, scalability, and interconnect capability.

Google's quantum computing team, led by John Martinis and subsequently Hartmut Neven, developed superconducting qubit systems and in October 2019 published a claim of "quantum supremacy" — demonstrating that their fifty-three-qubit Sycamore processor performed a specific sampling calculation in two hundred seconds that would take the world's most powerful classical supercomputer approximately ten thousand years. IBM disputed the time estimate for classical simulation, but the result was widely recognized as a milestone demonstrating that quantum processors had reached capabilities beyond classical simulation for at least some narrow tasks.

IBM has pursued quantum computing through its IBM Quantum platform, providing cloud access to superconducting quantum computers with progressively increasing qubit counts. IBM's quantum volume metric — measuring both qubit count and circuit quality — was used to track progress in the early 2020s as qubit counts grew into the hundreds. IBM announced a one-thousand-and-nine-qubit processor (Condor) in 2023, alongside a modular architecture for connecting multiple quantum processors.

IonQ, a company spun out of the University of Maryland and Duke University, developed trapped-ion quantum computers — using individual ions held in electromagnetic traps as qubits — which offer longer coherence times than superconducting qubits but face different scaling challenges. Honeywell (subsequently merged with Cambridge Quantum Computing to form Quantinuum) also developed trapped-ion systems. D-Wave Systems, a Canadian company, commercialized quantum annealing processors — a different and more limited form of quantum computation suited to optimization problems — from 2011 onward.

Quantum hardware development outside the United States has been substantial. China has invested heavily in quantum computing through state programs: the University of Science and Technology of China (USTC) developed the Jiuzhang photonic quantum computer, which in 2020 claimed a quantum advantage in Gaussian boson sampling two hundred times faster than classical simulation. China's Zuchongzhi superconducting processor achieved sixty-six qubits and competed with Google's claims of quantum advantage in 2021. The EU funded quantum technology through its Quantum Flagship initiative, a one-billion-euro ten-year program supporting quantum computing, communication, simulation, and sensing research across member states.

Progress in quantum error correction — protecting quantum computations from decoherence by encoding logical qubits redundantly across multiple physical qubits — is considered essential for achieving fault-tolerant quantum computers capable of running Shor's algorithm against real-world encryption key lengths. Google's demonstration of below-threshold error correction in 2023 (showing that adding more physical qubits improved rather than degraded logical qubit performance) was a milestone toward fault-tolerant quantum computing, though the number of physical qubits required to implement a fault-tolerant logical qubit capable of breaking real encryption remains in the hundreds to thousands.

The timeline for commercially relevant quantum computing remains uncertain: most researchers expect that near-term quantum devices will find applications in quantum chemistry simulation (designing drugs and materials), financial optimization, and machine learning before tackling the cryptographic applications that originally motivated quantum computing research. Quantum-resistant cryptography — developing encryption methods secure against quantum attacks — has become an active area of standardization: the US National Institute of Standards and Technology (NIST) finalized its first post-quantum cryptographic standards in 2024.

Supercomputing and High-Performance Computing

While personal computers and smartphones represent computing as a mass consumer product, the frontier of raw computing power is found in supercomputers — the largest and most powerful computing systems built, used for scientific research, engineering simulation, weather forecasting, nuclear weapons simulation, and increasingly AI model training.

The history of supercomputing begins with Seymour Cray, an electrical engineer who designed some of the fastest computers of each era from the 1960s through the 1990s. Cray designed the CDC 1604 (1960) and CDC 6600 (1964) at Control Data Corporation — the latter was the world's fastest computer for five years and is often considered the first true supercomputer. Cray left CDC to found Cray Research in 1972, and the Cray-1 (1976) — a distinctive cylindrical machine with a bench seat along its base containing the cooling system — became the iconic supercomputer of its era, using vector processing (performing the same operation on multiple data simultaneously) to achieve speeds far beyond conventional computers.

The TOP500 list, first published in 1993, has tracked and ranked the world's five hundred most powerful supercomputers biannually, providing a benchmark of national and institutional computing ambitions. The processing speed of the top supercomputer has grown from approximately fifty-nine gigaflops (the Intel Paragon in 1993) to over one exaflop (one billion billion floating-point operations per second) in the early 2020s.

The United States dominated supercomputing through most of the TOP500's history, with machines from vendors including Cray, IBM, SGI, and HP leading performance rankings through the 1990s and 2000s. Japan mounted serious challenges: the Earth Simulator (2002), built by NEC for the Japan Agency for Marine-Earth Science and Technology, was the world's fastest supercomputer for nearly three years, prompting concerns in the US computing community about competitiveness.

China's ascent in supercomputing has been dramatic. The Tianhe-1A at the National Supercomputing Center in Tianjin became the world's fastest computer in November 2010. The Sunway TaihuLight (2016), built by the National Research Center of Parallel Computer Engineering and Technology using entirely Chinese-designed Sunway processors without any foreign chips, topped the TOP500 list for two years — a demonstration of China's growing independent capabilities in high-performance computing. By the early 2020s, China had more computers in the TOP500 than the United States.

The Frontier supercomputer at Oak Ridge National Laboratory in Tennessee, operational from 2022, was the first machine to achieve exascale computing — over one billion billion floating-point operations per second. Frontier used AMD processors and AMD Instinct GPU accelerators. The Aurora system at Argonne National Laboratory, using Intel processors and Intel Ponte Vecchio GPUs, reached exascale capability in 2023. El Capitan at Lawrence Livermore National Laboratory, using AMD Instinct MI300 accelerators, set new performance records in 2024.

The use of graphics processing units (GPUs) as accelerators in supercomputers became dominant in the 2010s. GPUs, originally designed for the parallel floating-point operations needed to render video games, were found to provide extraordinary throughput for the massively parallel computations of scientific simulation and, crucially, deep learning neural network training. Nvidia's dominance of the GPU market — with its CUDA software platform and its A100 and H100 data center GPUs — positioned the company at the center of both supercomputing and AI infrastructure. Nvidia's market capitalization surpassed one trillion dollars in 2023 and three trillion dollars in 2024, making it one of the most valuable companies in the world, driven primarily by demand for its AI training and inference accelerators.

Cloud Computing and the Data Center Era

Cloud computing — the provision of computing resources (processing, storage, networking, software) as a service over the internet, charged by usage — has transformed the economics of computing and concentrated enormous computing infrastructure in the hands of a small number of large companies.

Amazon Web Services (AWS), launched publicly in 2006, pioneered the commercial cloud computing market, offering virtual computing capacity (Elastic Compute Cloud, EC2), storage (Simple Storage Service, S3), and databases as pay-per-use services. Amazon built AWS partly to make use of computing infrastructure it had built for its own retail operations, recognizing that other companies would benefit from access to similar capabilities without building their own data centers. AWS grew rapidly to become the dominant cloud platform, generating more than half of Amazon's corporate operating profit by the early 2020s despite representing a fraction of its revenue.

Google Cloud Platform and Microsoft Azure followed AWS as the other two major cloud providers in a market that became known as the "hyperscalers" — companies building data centers at a scale previously unknown, with individual facilities consuming hundreds of megawatts of electricity and serving millions of customers simultaneously. By the mid-2020s, the three hyperscalers together controlled approximately sixty-five percent of the global cloud market, with AWS the leader, followed by Microsoft Azure and Google Cloud.

The data center infrastructure required for cloud computing has become one of the world's largest consumers of electricity and water. A large modern hyperscale data center consumes one hundred to three hundred megawatts of electricity continuously — comparable to a small city — and uses enormous quantities of water for cooling. The location of data centers is influenced by electricity costs (favoring regions with cheap renewable power, such as Iceland, Norway, and the US Pacific Northwest), cooling climate, land availability, and network infrastructure.

Key Figures in Computing History

The history of computing has been shaped by a relatively small number of individuals whose theoretical contributions, engineering innovations, or entrepreneurial decisions had disproportionate impact on the development of the field.

Alan Turing (1912-1954) is widely considered the father of theoretical computer science and artificial intelligence. His 1936 paper on computable numbers introduced the Turing Machine — an abstract computational model that defined what could be computed in principle — establishing the mathematical foundations of computer science. His work at Bletchley Park was crucial to British codebreaking in World War II. His 1950 paper "Computing Machinery and Intelligence" framed the question of machine intelligence in terms that shaped AI research for decades. Turing was prosecuted for homosexual conduct in 1952, subjected to chemical castration, and died in 1954 at age forty-one — officially ruled a suicide by cyanide poisoning, though the circumstances have been questioned. He received a royal pardon in 2013.

John von Neumann (1903-1957) — a Hungarian-American mathematician of extraordinary breadth — contributed to the foundations of quantum mechanics, game theory, economics, nuclear weapons design, and computer architecture. His articulation of the stored-program computer architecture in the 1945 EDVAC report influenced virtually every general-purpose computer subsequently built. Von Neumann also contributed to Monte Carlo methods (random simulation techniques used in physics and finance), cellular automata (a model of self-reproducing systems), and game theory.

Grace Hopper (1906-1992) — an American mathematician and US Navy rear admiral — developed the first compiler: a program that translates human-readable programming language into machine code. Hopper's A-0 compiler (1952) and subsequent B-0 compiler (FLOW-MATIC, 1955) established the concept that computers could be programmed in languages closer to human language than machine code. Hopper was also involved in the development of COBOL (Common Business-Oriented Language), the programming language that became the dominant language for business computing and remains in use in financial systems today. Her discovery of a literal bug — a moth found in a relay of the Harvard Mark II computer in 1947, pasted into the logbook with the note "First actual case of bug being found" — gave the computing world the metaphor of a "bug" for a program error.

Claude Shannon (1916-2001) — an American mathematician and electrical engineer at Bell Labs — founded information theory with his 1948 paper "A Mathematical Theory of Communication," which introduced the bit (binary digit) as the fundamental unit of information, defined channel capacity, and established the mathematical framework for understanding the transmission of information through noisy channels. Shannon's information theory underlies all of modern digital communication and data compression.

Tim Berners-Lee (born 1955) — the British physicist and computer scientist who invented the World Wide Web — is unique in the computing pantheon for having given away, without patent or royalty, a technology of extraordinary commercial value. Working at CERN, Berners-Lee proposed and implemented the Web's core technologies — HTML, HTTP, and URLs — and made them freely available, a decision he has said was essential to the Web's universal adoption. He subsequently founded the World Wide Web Consortium (W3C) to develop Web standards and has become an advocate for an open, decentralized Web and for the protection of users' privacy.

Linus Torvalds (born 1969) — the Finnish-American software engineer who created the Linux kernel — initiated and continues to coordinate the development of the most widely deployed operating system kernel in the world, found in Android smartphones, web servers, supercomputers, and embedded devices from automobiles to spacecraft. Linux is developed collaboratively by thousands of contributors under the GNU General Public License, and Torvalds's role as the final arbiter of what code is accepted into the kernel has given him an unusual position as the ultimate authority over a global open-source project of extraordinary importance.

Steve Jobs (1955-2011) — the co-founder of Apple and its chief executive from 1997 to 2011 — had a unique talent for combining technology, design, and marketing into products that created new categories: the Apple II, the Macintosh, the iPod, the iPhone, and the iPad each transformed their respective markets. Jobs's insistence on integrated hardware-software design, his attention to aesthetics and user experience, and his willingness to remove features (including the physical keyboard from the iPhone) that he judged unnecessary set Apple apart from competitors and established design principles that influenced the entire industry.

Geoffrey Hinton (born 1947) — the British-Canadian cognitive scientist whose decades of research on neural networks laid the foundations of deep learning — shared the 2018 Turing Award with Yann LeCun and Yoshua Bengio for contributions to the field. Hinton worked at the University of Toronto for decades before spending time at Google Brain. In 2023, Hinton resigned from Google and publicly expressed concerns about the risks posed by the AI systems whose development he had contributed to — an unusual and influential act of public warning from one of the founders of modern AI.

The Ai Safety Movement and Governance

The rapid advancement of artificial intelligence capabilities, particularly after the deep learning revolution of the 2010s and the emergence of large language models in the early 2020s, has generated serious academic, policy, and public debate about the risks posed by increasingly capable AI systems.

The AI safety field — focused on ensuring that AI systems behave as intended, remain under human control, and do not cause catastrophic harm — developed in parallel with AI capabilities research. Eliezer Yudkowsky at the Machine Intelligence Research Institute (MIRI), Nick Bostrom at Oxford (whose 2014 book "Superintelligence" brought existential risk arguments to a wide audience), and Stuart Russell at UC Berkeley (co-author of the most widely used AI textbook and author of "Human Compatible," arguing for AI systems that defer to human preferences) were prominent voices articulating concern about advanced AI systems.

OpenAI was founded explicitly around the mission of ensuring that artificial general intelligence (AGI) benefits humanity, though its transition from a nonprofit to a "capped profit" structure, its large investments from Microsoft, and its rapid commercial deployment of powerful language models generated controversy about how effectively it pursued its stated safety mission. Anthropic, founded in 2021 by former OpenAI executives including Dario Amodei and Daniela Amodei, positioned itself as an AI safety-focused research company, developing Constitutional AI methods and publishing research on how to make language models more honest and harmless.

Government responses to AI governance accelerated from 2023 onward. The US Executive Order on AI Safety (October 2023) required companies developing large AI systems to share safety test results with the government. The UK hosted the first AI Safety Summit at Bletchley Park in November 2023, producing a declaration signed by twenty-eight countries (including the US, EU, China, and others) acknowledging potential catastrophic risks from frontier AI. The EU's Artificial Intelligence Act, agreed in December 2023 and entering force in 2024, established the world's first comprehensive legal framework for AI systems, imposing requirements based on assessed risk levels.

China has pursued its own AI development strategy aggressively, with companies including Baidu, Alibaba, Tencent, Huawei, and a new generation of AI startups including Zhipu AI, Moonshot AI, and DeepSeek developing large language models and AI applications. DeepSeek's R1 reasoning model, released in January 2025, demonstrated performance comparable to leading US models at dramatically lower training cost, disrupting assumptions about the resources required for frontier AI development and temporarily affecting Nvidia's share price.

Computing's Global Reach: India, China, and the Developing World

Computing's impact on economic development has been profound and uneven. The digital divide — the gap between those with access to digital technology and those without — has been a persistent concern since the personal computer era, and the smartphone has done more to bridge it than any previous technology.

India developed a significant computing and information technology industry from the 1980s onward, initially focused on software services exported to Western companies. The software export model — using India's large population of English-speaking, mathematically trained engineers to develop software for Western clients at lower labor costs — proved remarkably scalable. Infosys (founded 1981 by N.R. Narayana Murthy and others), Wipro, Tata Consultancy Services (TCS), and HCL Technologies became global IT services companies with hundreds of thousands of employees. Bangalore (now Bengaluru), Hyderabad, and Pune became major technology centers. India's IT services exports exceeded two hundred billion dollars annually by the early 2020s.

The Indian IT industry also produced a generation of technology executives who led major global companies: Satya Nadella (Microsoft), Sundar Pichai (Google/Alphabet), Shantanu Narayen (Adobe), Arvind Krishna (IBM), and Indra Nooyi (PepsiCo) among them. India's technology diaspora in Silicon Valley became one of the most influential immigrant communities in the US technology industry.

China's technology sector followed a different trajectory: rather than primarily exporting services, China developed its own large-scale digital economy, protected from foreign competition by the Great Firewall that blocks Western platforms (Google, Facebook, Twitter, YouTube, WhatsApp) and supported by government industrial policy. Baidu (search), Alibaba (e-commerce and cloud), Tencent (social media, games, payments), and ByteDance (TikTok/Douyin) became among the world's largest internet companies. China's mobile payment ecosystem — dominated by Alipay and WeChat Pay — achieved adoption levels and transaction volumes far exceeding those of Western equivalents, leapfrogging credit card infrastructure and becoming the dominant payment method in China.

Africa has seen mobile computing and internet access grow rapidly since the mid-2000s, with smartphone adoption spreading even in countries with limited fixed-line infrastructure. Mobile money systems, pioneered by M-Pesa in Kenya (launched 2007 by Safaricom and Vodafone), demonstrated that mobile networks could provide financial services to populations without bank accounts. M-Pesa grew to process a significant fraction of Kenya's GDP in transactions and was adopted across multiple African countries. The spread of affordable Android smartphones has brought internet access, digital payment, and mobile services to hundreds of millions of people across Sub-Saharan Africa, Southeast Asia, and Latin America who never had personal computers.

The Future of Computing

The computing industry faces several fundamental transitions in the coming decades that will shape what computers can do, where they operate, and who controls the most powerful systems.

The slowdown of Moore's Law — as transistor sizes approach physical limits (a silicon atom is approximately two-tenths of a nanometer, and transistors are already only a few dozen atoms across) — is forcing the industry to find performance improvements from new directions: specialized processors (AI accelerators, networking chips, signal processors), new memory architectures (high-bandwidth memory stacked close to processors), chiplet designs (assembling chips from smaller tiles manufactured separately), and new materials (gallium nitride, silicon carbide, two-dimensional materials like graphene and molybdenum disulfide).

Neuromorphic computing — designing chips that more closely mimic the structure and energy efficiency of biological neural circuits — is pursued by Intel (Loihi chips), IBM (TrueNorth), and academic researchers as a potential path to dramatically more energy-efficient AI computation. Today's AI training requires millions of watts; the human brain performs comparable cognitive tasks on approximately twenty watts.

The intersection of computing and biology is opening new frontiers: DNA data storage (encoding digital information in the base sequences of synthetic DNA molecules), biological computing using engineered cells, and the use of AI in drug discovery, genomics, and personalized medicine all represent the convergence of computing and the life sciences.

The geopolitical dimension of computing has become increasingly central to international relations. Semiconductors, AI systems, and the data they process are recognized by governments as strategic assets, and the US-China technology competition has established competing visions for how the digital infrastructure of the twenty-first century will be governed, owned, and secured.