Why “first” is a tricky word
Asking which machine was the first computer sounds simple, but the answer depends on how you define the term. Must the machine be electronic, or can it use relays and gears? Must it be programmable? Must it be general purpose? Must it keep its program in memory? Different early machines satisfy different parts of this list, and historians and national traditions sometimes disagree. This article describes several of the best-known early computers and what set each apart, rather than naming one winner. Earlier mechanical attempts are covered in our broader history of computing.
Machines of the early 1940s
In Germany, Konrad Zuse built a series of programmable calculating machines in the late 1930s and early 1940s. His Z3, completed in 1941, used electromechanical relays and read instructions from punched film. It is often described as an early programmable, automatic computer, though it was not electronic.
In the United States, John Atanasoff and Clifford Berry at Iowa State developed a special-purpose electronic machine around 1939 to 1942 for solving systems of linear equations. It used electronic switching and regenerative memory but was not programmable in the general sense. Its role in later patent disputes made its place in history contested.
At Harvard, the Mark I, completed in 1944 with support from IBM, was a large electromechanical calculator controlled by a paper tape. It performed long calculations for wartime and scientific work.
Colossus and codebreaking
In Britain, codebreakers needed to analyze enciphered German messages quickly. The Colossus machines, first operating in 1943 to 1944, used a large number of vacuum tubes to carry out logical operations on message data. They were electronic and could be configured for different tasks, but they were designed for a specific kind of cryptanalysis rather than general computation. Their existence stayed secret for decades after the war, which is one reason the early history was told incompletely for a long time.
ENIAC
The Electronic Numerical Integrator and Computer, or ENIAC, was built at the University of Pennsylvania and became operational in 1945, with a public unveiling in 1946. It was designed to calculate artillery firing tables for the U.S. Army, and was soon used for other scientific problems. It used many thousands of vacuum tubes, took up a large room and weighed on the order of tens of tons. Its speed was a leap over mechanical calculators, performing thousands of simple operations per second.
ENIAC was programmable, but not in the way we now think of the term. Changing the task meant rewiring cables and setting switches, which could take days. Several women who worked as programmers on the machine developed much of the practical know-how for operating it, and their contribution was underrecognized for years.
The stored-program idea
The weakness of ENIAC’s set-up helped drive the idea of storing instructions in the same memory as data, so that a program could be loaded and changed quickly. This concept is associated with a 1945 report by John von Neumann and with the work of many others on the same problem, so crediting a single inventor is misleading.
In 1948, the Manchester Small-Scale Experimental Machine, nicknamed the Baby, ran a program from its own electronic memory. It is generally regarded as the first stored-program computer to work, though it was a small demonstration. In 1949, the EDSAC at Cambridge became an early practical stored-program machine used by researchers. In the United States, the UNIVAC I, delivered in the early 1950s, was among the first computers sold for business and government use.
Size, speed and limits
Machines of this era had very little memory by today’s standards, often storing only a few thousand numbers or instructions. Vacuum tubes burned out and generated enormous heat, so reliability was a constant concern, and teams had to schedule maintenance around calculations. Input and output relied on punched cards, paper tape and printed output. Programming meant writing numeric codes by hand, which was slow and error-prone.
Cost and expertise limited who could use them. Only governments, universities and a few large organizations could afford one, and a typical machine served a team of specialists rather than individual users. The later move to semiconductor chips is what finally broke this pattern.
What they were used for
Typical early uses included artillery tables, codebreaking, atomic research, weather and engineering calculations, and large census or payroll data processing. These jobs involved either repetitive arithmetic or sorting large records, which were exactly what the machines did well.
The early programs were not just calculations. People soon explored whether machines could play simple games or prove theorems, which led into questions later pursued in the history of AI.
Why this period matters
The first computers established several principles that still hold: electronic switching is fast, general-purpose machines can be reprogrammed for new tasks, and programs can be treated as data. They also showed how quickly a technology can improve. Within about a decade, machines moved from one-off experiments to products with customers, setting the stage for the transistor-based computers that followed.