Who Built the First Digital Computer?

 

The honest answer requires a table, not a sentence. The textbook answer — ENIAC, then later ABC after a 1973 patent ruling — is missing a third contender that the British government refused to acknowledge until 1976. The priority dispute was settled in court before all the evidence was permitted in the room.

By Stephen "Pseudo Publius"  |  7 May 2026

BLUF — The claim that Eckert and Mauchly's ENIAC (1946) was the first digital computer was the dominant American consensus from roughly 1946 through 1973. It was overturned, in U.S. patent law, by Judge Earl Larson's October 1973 ruling in Honeywell, Inc. v. Sperry Rand Corp., which invalidated the ENIAC patent on the grounds that John Mauchly had visited John Atanasoff in Iowa in June 1941 and "derived" key concepts from the Atanasoff-Berry Computer. The popular account thus shifted to "ABC came first." Both narratives, however, were arguing on incomplete evidence, because the strongest counter-claimant — the British Colossus, operational at Bletchley Park in February 1944 — was still classified. Brian Randell's October 1976 paper at Los Alamos was the first technical disclosure to the international computing community. By the time Colossus entered the conversation, the legal dispute was already closed and the textbooks were already printed. The defensible position today, accepting all the evidence, is that there is no single "first digital computer" — only firsts under specific qualifiers, and Colossus holds the strongest claim under the qualifier programmable, electronic, digital.^1,2,3

The five machines

A useful priority discussion starts by laying the candidates out on the same axis. There are five machines (six if Babbage's never-built Analytical Engine is counted) that have, at various times, been called the first digital computer. They differ along five technical dimensions that matter:

Machine	Date	Electronic?	Digital?	Programmable?	General-purpose?	Stored-program?
Zuse Z3 (Berlin)	May 1941	No (relays)	Yes	Yes	Turing-complete*	No
Atanasoff-Berry Computer (ABC) (Iowa State)	1942	Yes (~300 valves)	Yes (binary)	No	No (linear systems only)	No
Colossus Mk I (Bletchley Park)	Feb 1944	Yes (1,500 valves)	Yes	Yes (plugboard)	No (Tunny attack only)	No
ENIAC (U. Pennsylvania)	Feb 1946	Yes (~17,500 valves)	Yes (decimal)	Yes (rewiring)	Yes	No
Manchester Baby (SSEM)	June 1948	Yes	Yes	Yes	Yes	Yes
UNIVAC I (Eckert-Mauchly / Remington Rand)	March 1951	Yes	Yes	Yes	Yes	Yes

*The Z3's Turing-completeness was demonstrated in 1998 by Raúl Rojas; Zuse's machine technically lacked conditional branching but could simulate it through a workaround.

Reading down the table, the answer to "first digital computer" depends entirely on which adjectives the questioner allows. Z3 is first if you accept electromechanical relays. ABC is first electronic but it cannot be programmed. Colossus is first programmable electronic, but only for one job. ENIAC is first general-purpose programmable electronic. Manchester Baby is first stored-program — first computer in the modern architectural sense. UNIVAC I is first commercial.^1,2,4

The Eckert-Mauchly claim, briefly

John Mauchly, a physics professor at the Moore School of Electrical Engineering at the University of Pennsylvania, and J. Presper Eckert, a young electrical engineer at the same school, built ENIAC under a U.S. Army Aberdeen Proving Ground contract starting in 1943. The machine went operational in late 1945 and was publicly unveiled in February 1946. It used roughly 17,500 vacuum tubes, occupied 1,800 square feet, weighed 30 tons, and was originally designed to compute artillery firing tables. Reprogramming it required physical rewiring of plugboards — a process that could take days.^2,5

Eckert and Mauchly's claim to "first electronic digital computer" rested on three elements: ENIAC was demonstrably first to be publicly demonstrated; it was demonstrably first to be general-purpose and programmable; and the underlying patent (No. 3,120,606), filed in 1947 and granted in 1964, named Eckert and Mauchly as the inventors. They left Penn in 1946 over an intellectual-property dispute, founded the Eckert-Mauchly Computer Corporation, were acquired by Remington Rand in 1950, and built UNIVAC I — the first commercial computer in the United States, delivered to the U.S. Census Bureau in March 1951. UNIVAC I established the commercial computer industry and gave the Eckert-Mauchly partnership a second, durable claim: not "first computer," but "first computer that could be bought."^2,5

Both claims were broadly accepted in the United States through the 1960s. The standard textbook line — ENIAC was first, UNIVAC was first commercial — entered American technical education and stayed there for a generation.

1973: Larson's ruling

The first major challenge to that consensus came not from a historian but from a patent attorney. Sperry Rand, the owner of the ENIAC patent through its acquisition of Eckert-Mauchly, had been collecting licensing fees from competitors throughout the 1960s. Honeywell, refusing to pay, sued. The case, Honeywell, Inc. v. Sperry Rand Corp., was tried in U.S. District Court in Minneapolis from 1971 to 1973. Honeywell's defense centered on a single fact: in June 1941, four years before ENIAC began, John Mauchly had spent five days at the home of John Vincent Atanasoff in Ames, Iowa, examining a working prototype of the Atanasoff-Berry Computer.^3,6

Atanasoff, a physics professor at Iowa State College, and his graduate student Clifford Berry had built the ABC between 1939 and 1942 to solve systems of linear equations — a problem set Atanasoff had encountered repeatedly in his own research. The machine used about 300 vacuum tubes for its arithmetic logic, regenerative capacitor memory mounted on rotating drums, and binary arithmetic. It was the first electronic digital computer to use vacuum tubes for computation. It was not programmable in any modern sense; it solved one class of problem and only that class. Atanasoff and Berry never patented it. Berry committed suicide in 1963, well before the legal dispute that would make his name famous.⁴

Honeywell's argument was that Mauchly had observed Atanasoff's work in 1941, taken notes, and incorporated the key concept — vacuum-tube-based binary digital computation — into ENIAC without attribution. After a 135-day trial, Judge Earl Larson agreed. His October 1973 ruling invalidated the ENIAC patent on the grounds that "Eckert and Mauchly did not themselves first invent the automatic electronic digital computer, but instead derived that subject matter from one Dr. John Vincent Atanasoff."^3,6

The legal effect was financial: Sperry Rand lost the ability to extract licensing fees. The historical effect was textual: many computing histories quietly amended themselves to credit Atanasoff. The popular framing shifted from "ENIAC was first" to "ABC was first, ENIAC was first general-purpose." This is the version most American textbooks now carry.

What the courtroom didn't know

The 1973 ruling was made on incomplete evidence — not because Honeywell's lawyers were negligent but because the strongest counter-claim was still classified by a foreign government. In October 1973, when Larson handed down his decision, the British Official Secrets Act still applied to the Bletchley Park codebreaking program. The first photographs of Colossus had been quietly released in 1970, but no technical description existed in the open literature. Tommy Flowers, the engineer who had designed and built it, was alive, working at the Post Office, and forbidden to say what he had done.^7,8

Brian Randell, a computer historian at the University of Newcastle upon Tyne, had been working since 1971 to extract a public account of Colossus from the British government. He succeeded only partially. In October 1976 — three years after Larson's ruling — Randell delivered a paper at the International Research Conference on the History of Computing at Los Alamos National Laboratory, titled "The Colossus." It was the first substantive technical disclosure of the Bletchley machine to the international computing community. Even then the disclosure was incomplete; Randell had been permitted to describe the machine's existence and broad architecture but not its full technical detail. The complete General Report on Tunny, written by Bletchley staff in 1945 as a classified internal document, would not be released to the public until the year 2000.^7,8

If Randell's disclosure had come three years earlier — or if the British government had declassified Colossus in 1965 instead of 1976 — Honeywell's lawyers would have had a third candidate to introduce into the priority dispute. Colossus was operational in February 1944, two full years before ENIAC. It was vacuum-tube electronic, fully digital, programmable through plugboard configuration, and operated continuously for sixteen months before ENIAC was first switched on. It was special-purpose, but so was ABC. The argument that Colossus, not ABC, was the first electronic digital computer in operation — and the first programmable one — would have been very difficult to refute.

That argument was never made in court because the witnesses could not legally testify. The U.S. patent dispute therefore proceeded as an ABC-versus-ENIAC question, and the British contribution was quite literally inadmissible. By 1976, when the disclosure came, the legal proceedings were closed, the patent had been invalidated on different grounds, and the textbook narrative had already shifted to its post-Larson form. Colossus arrived too late to participate in the priority dispute, and computing history has been catching up ever since.^1,2

The Zuse claim, which Americans still mostly ignore

There is a fourth claimant the American consensus has historically been even worse at acknowledging. Konrad Zuse, a civil engineer in Berlin, completed the Z3 in May 1941 — eight months before the United States entered the war and four years before ENIAC's first calculation. The Z3 was electromechanical, using approximately 2,000 telephone relays rather than vacuum tubes. But it was binary, it was floating-point (22-bit, with a 14-bit mantissa and 7-bit exponent), and it was programmable through a punched film tape. Zuse demonstrated it to officials of the German Aircraft Research Institute, who used it for statistical analyses of wing flutter. The original Z3 was destroyed in the Allied bombing of Berlin on 21 December 1943 — the same month Colossus was being delivered to Bletchley Park. Zuse rebuilt it, eventually, in the 1960s; the rebuild is on display at the Deutsches Museum in Munich.⁹

The Z3 was shown by Raúl Rojas in 1998 to be Turing-complete in principle, despite the lack of a true conditional-branch instruction (a workaround was possible). Under the qualifier "first programmable, Turing-complete digital computer of any technology," the answer is the Z3 — by a wide margin. Under the qualifier "first programmable, electronic digital computer," the Z3 is excluded by the relay technology and the answer is Colossus. The Z3's claim has been historically diminished partly because it was destroyed in the war, partly because Zuse worked in isolation under wartime conditions and could not publish, and partly because postwar computing history was written largely in English by people who had limited contact with the German engineering community.^1,9

The honest answer

Given the table at the top of this article, the honest answer to "who built the first digital computer" is that there is no single answer. Different machines hold different priority claims, and the claims are technically distinguishable:

• First conception: Charles Babbage's Analytical Engine, designed in the 1830s but never built in his lifetime.
• First programmable digital computer (any technology): Zuse Z3, May 1941.
• First electronic digital computing device: Atanasoff-Berry Computer, 1942 (special-purpose, not programmable).
• First programmable electronic digital computer: Colossus Mk I, February 1944 (special-purpose, for Lorenz cipher attack).
• First general-purpose programmable electronic digital computer: ENIAC, February 1946.
• First stored-program (von Neumann architecture) computer: Manchester Baby (SSEM), June 1948.
• First commercial digital computer: UNIVAC I, March 1951.

Stephen's intuition — that the British Colossus was really the first, but received no press because of classification — is essentially correct, with the qualifier that "first" needs the modifier "programmable, electronic, digital." Under that qualifier, Colossus has the strongest claim, and the textbook answer of "ENIAC, then ABC" reflects what American historians and patent attorneys could see, not what was actually true.

Why this still matters

The priority dispute is, on one level, a historians' quarrel. On another level it is a parable about how technical history gets written under classification. The 1973 court ruling that crowned the ABC was not wrong on the evidence available; it was wrong on the evidence permitted. The textbook line that crowned ENIAC was not a conspiracy; it was the best account that could be written by people who legitimately did not know about Colossus. The British government's three-decade silence, motivated by concerns about Cold War cryptanalytic capabilities — and partly by sheer institutional inertia — distorted the historical record in ways that are still being corrected.

The Memphis facility now bearing the name Colossus inherits this legacy whether its operators know it or not. The actual first programmable electronic digital computer was a British military secret for thirty years; its engineer, Tommy Flowers, was paid less than the cost of his own materials and forbidden to claim credit. The frontier-AI infrastructure being built today is, by contrast, the most extensively publicized engineering effort of the 21st century — billion-dollar press releases, FCC filings of unprecedented scale, IPO roadshows. The contrast is worth holding onto. The first computer of consequence was built by a man who could not tell anyone what he had done. The latest one is being built, in part, by people who appear to be telling everyone they meet.

Which approach will produce more durable engineering, history will eventually decide. But on the priority of "first digital computer," the record is now, with Colossus included, as clear as it is going to get: there is no single first, and the British claim is stronger than the textbooks built between 1946 and 1976 ever permitted it to be.

---

Sources

1. Computer History Museum. "The Neverending Quest for 'Firsts.'" Blog post, 2019. https://computerhistory.org/blog/the-neverending-quest-for-firsts/
2. Computer History Museum, "Timeline of Computer History." https://www.computerhistory.org/timeline/computers/
3. Honeywell, Inc. v. Sperry Rand Corp., 180 USPQ 673 (D. Minn. 1973). Memorandum opinion, Judge Earl R. Larson, 19 October 1973. Iowa State University maintains the trial archives at https://jva.cs.iastate.edu/courtcase.php
4. "Atanasoff-Berry computer." Wikipedia, retrieved 7 May 2026. https://en.wikipedia.org/wiki/Atanasoff%E2%80%93Berry_computer
5. McCartney, S. ENIAC: The Triumphs and Tragedies of the World's First Computer. Walker & Company, 1999. Standard popular history of the Penn project; Eckert and Mauchly's commercial trajectory through UNIVAC.
6. Burks, A.R. and Burks, A.W. The First Electronic Computer: The Atanasoff Story. University of Michigan Press, 1988. The standard pro-Atanasoff scholarly account, written by participants in the Honeywell case.
7. Randell, B. "The Colossus." Paper presented at International Research Conference on the History of Computing, Los Alamos National Laboratory, 10 June 1976. Reprinted in Metropolis, N., Howlett, J., and Rota, G.-C. (eds.), A History of Computing in the Twentieth Century, Academic Press, 1980.
8. Good, I.J., Michie, D., and Timms, G. General Report on Tunny, with Emphasis on Statistical Methods. Bletchley Park, 1945. Declassified by the UK Government in 2000. Primary technical record of the Lorenz attack and Colossus's role.
9. Rojas, R. "How to make Zuse's Z3 a universal computer." Annals of the History of Computing, IEEE, vol. 20, no. 3, July-September 1998. The proof that the Z3 was Turing-complete despite lacking explicit conditional branching. Z3 timeline and technical specifications: see also Computer History Museum timeline (Ref. 2).