The first computer was invented in the third century BC: the Antikythera mechanism, a Greek bronze gear device built around 150–100 BC to calculate astronomical positions. If you mean the first programmable electronic computer, that is the ENIAC, completed in 1946. If you mean the first programmable machine of any kind, the Jacquard loom of 1804 holds the claim. Each answer is correct inside a different definition of "computer."

What Counts as a Computer

The word has drifted for two centuries, and the drift explains most of the disagreement. Early "computers" were people, often women, hired to perform arithmetic by hand. The machine sense arrived later and kept expanding.

Four tests separate the candidates. Does it execute a stored sequence of instructions? Can that sequence be changed without rebuilding the hardware? Does it compute automatically, without a human turning cranks? And is it electronic?

Almost no early device passes all four. The Antikythera mechanism automates calculation but is not programmable. The Jacquard loom is reprogrammable but computes nothing. The ENIAC is electronic and programmable but was rewired, not loaded from memory. Understanding why each candidate fails or passes one test is the real history.

The Gear Mechanism That Predates Everything

The Antikythera mechanism was recovered in 1901 from a Roman-era shipwreck off the Greek island of Antikythera. It is a hand-cranked box of at least 30 interlocking bronze gears, and it is the oldest known analog computer.

Its function was astronomical prediction. A front dial tracked the date against the zodiac. A rear spiral dial predicted solar and lunar eclipses using the Saros cycle, a 223-lunar-month pattern. The device also modeled the Moon's variable orbital speed.

That speed variation is the deep mechanism. The Moon appears to move faster at some points in its orbit than others, and Greek astronomers modeled this with a pin-and-slot design. A gear mounted slightly off-center drives a pin in a slot, so the output rotation speeds up and slows down through each turn. This is a mechanical implementation of an inequality, a mathematical function of time.

The device could not be reprogrammed. Its instructions were cut into its gears. That single fact disqualifies it under the stored-program test but not under the older meaning: a machine that computes a predictable output from inputs.

Quantified Context: From 30 Gears to 18,000 Vacuum Tubes

Scale separates the eras more sharply than dates do.

The Antikythera mechanism used roughly 30 gears. Charles Babbage's Analytical Engine, designed from 1837 but never completed in his lifetime, called for about 25,000 parts. The ENIAC, finished in 1946, filled 1,800 square feet and contained roughly 18,000 vacuum tubes, 70,000 resistors, and 10,000 capacitors. It weighed about 30 tons and drew roughly 150 kilowatts.

That power draw matters. The ENIAC's tubes burned out constantly, and engineers replaced them in shifts. Reliability, not logic, was the binding constraint. A machine that fails every few hours cannot run a long computation, which is why stored-program designs won: the same hardware could run any program without physical reconfiguration.

Speed tells the same story. The ENIAC performed roughly 5,000 additions per second. A modern processor core does billions per second, a gap of roughly six orders of magnitude.

Machine Date Core mechanism Programmable? Electronic?
Antikythera mechanism c. 150–100 BC Interlocking bronze gears No (fixed gearing) No
Jacquard loom 1804 Punched cards steering hooks Yes (card sequence) No
Babbage Analytical Engine Designed 1837 Mechanical mill and store Yes (on paper) No
Z3 1941 Electromechanical relays Yes (tape) No (relay)
Atanasoff–Berry computer 1939–1942 Vacuum tubes, capacitors Not general-purpose Yes
ENIAC 1946 Vacuum tubes Yes (rewiring) Yes
Manchester Baby 1948 Stored-program CRT memory Yes (stored program) Yes

Edge Cases: Who Gets the Title

The boundary cases are where professional historians earn their keep. Treat each claim as conditional, not as a fact to defend.

The Z3, completed by Konrad Zuse in Berlin in 1941, was the first working programmable digital computer. It used electromechanical relays rather than tubes, so it was digital and programmable but not electronic. It was destroyed in 1943.

The Atanasoff–Berry computer, built at Iowa State between 1939 and 1942, was the first electronic digital computing device. It was not general-purpose and was never fully operational as a complete system. A 1973 court ruling in Honeywell v. Sperry Rand invalidated the ENIAC patent partly on the basis of Atanasoff's earlier work.

The Manchester Baby, formally the Small-Scale Experimental Machine, ran the first program stored in electronic memory on June 21, 1948. That is the birthdate of the stored-program architecture that every modern computer inherits.

A practical caution follows. Any "first computer" claim is a claim about a definition, and the definitions are not arbitrary: they track which capability mattered for later machines. The stored-program test is the one most engineers accept, because it separates machines you rebuild from machines you reprogram.

Misconceptions Worth Correcting

Several popular claims are wrong or incomplete, and each error teaches something.

The ENIAC was not the first computer. It was the first general-purpose electronic digital computer, and even that title is contested.

The ENIAC was not programmed by plugging cables alone. Six women, including Jean Bartik and Betty Holberton, designed the program sequences, and their work was treated as clerical until historians recovered it. Programming was mathematical labor, not wiring.

Babbage never built his machine. The Analytical Engine exists as drawings, and a portion was assembled by his son Henry. The design was sound in principle, and the failure was funding and manufacturing tolerance, not concept.

The Antikythera mechanism is not a clock. It had no escapement and no continuous timekeeping. It was a calculator you cranked to a chosen date.

Red flags to watch for in popular retellings:

A single "first" with no stated definition of computer. Dates given to the year for ancient devices, when the real uncertainty is decades. Claims that a design counts as an invention even when no working machine existed. Ignoring the human computers whose labor the machines replaced. Treating patent rulings as settled history rather than as legal findings.

A Timeline of Firsts

Year Event Why it matters
c. 150–100 BC Antikythera mechanism built Oldest known analog computer
1804 Jacquard loom demonstrated First widely used punched-card control
1837 Babbage designs Analytical Engine First general-purpose programmable design
1843 Lovelace publishes notes on the Engine First published algorithm for a machine
1890 Hollerith tabulator used in US census Punch cards at industrial scale
1941 Z3 operational First working programmable digital computer
1946 ENIAC completed First general-purpose electronic digital computer
1948 Manchester Baby runs stored program Birth of modern computer architecture
1971 Intel 4004 released First commercial single-chip microprocessor

Practical Actions: How to Date a "First" Yourself

You can resolve almost any first-computer claim with a short checklist, and the checklist scales to other contested firsts.

Step one: write down the claimant's definition of computer. Step two: test the claimant against the four criteria above, recording pass or fail for each. Step three: check whether a working machine existed or only a design. Step four: check whether the machine ran a stored program and whether that program could change without rewiring.

Thresholds help here. A device needs at least one stored, changeable instruction sequence to qualify as programmable. It needs electronic switching elements, meaning tubes, transistors, or chips, to qualify as electronic. It needs to be general-purpose to qualify as a computer in the modern sense.

Reconcile competing thresholds explicitly. The Antikythera mechanism fails the programmability threshold but passes the automation threshold, so it is a computer only under a premodern definition. The ENIAC passes programmability and electronics but fails the stored-program threshold, so it is a first only if you exclude memory architecture. The Manchester Baby passes all four, which is why many historians place the true origin of the modern computer there.

For research, go to primary sources: the 1973 Honeywell ruling, the Manchester machine's archived program listings, and museum catalogs from the National Museum of Computing and the Computer History Museum.

Limits of the Claim and the Evidence

The honest limits are large. Ancient dates carry decades of uncertainty, and the Antikythera mechanism's exact origin is unknown. Much early computing history was classified or discarded, and the ENIAC's tube count is reported as "roughly 18,000" because counts vary by source and configuration.

Definitions do more work than evidence in most first-computer arguments. That is not a defect: it is how historians of technology organize a messy record. The useful takeaway is that the question has several defensible answers, and the definition you choose determines which one is true. When someone states a first without a definition, ask for the definition before the date.