Short answer
ENIAC (Electronic Numerical Integrator and Computer) was a room-sized, programmable electronic computer built at the University of Pennsylvania for the U.S. Army and shown to the public on February 14, 1946. It was big news because it calculated far faster than any earlier machine, proving that general-purpose electronic computing was possible.
Follow-up question How does ENIAC compare with the computers that run AI today?
In February 1946, newspapers announced a machine that seemed to belong to the future. ENIAC, the Electronic Numerical Integrator and Computer, filled a large room at the Moore School of Electrical Engineering, part of the University of Pennsylvania. It had about 18,000 vacuum tubes, weighed some 30 tons, and could finish in seconds calculations that took people hours. The press called it a "Giant Brain."
ENIAC was not the first calculating machine, nor the first to use electronics. What made it special was the combination: it was fully electronic, very fast, and general-purpose. By rewiring it, its operators could set it to work on very different problems, from artillery tables to weather and atomic-energy calculations.
Why it was built
During the Second World War, the U.S. Army needed firing tables: long lists telling gunners how to aim artillery for different distances, winds and shells. Every table required thousands of trajectory calculations. At the Moore School, teams of women, called "computers" because computing was their job, worked these out with desk calculators and a mechanical device called a differential analyzer. They could not keep up.
John Mauchly, who had come to the Moore School through a wartime electronics course and stayed on to teach, proposed using vacuum tubes, the glowing glass switches found in radios, to do the arithmetic electronically. J. Presper Eckert, a young Penn engineer, figured out how to make it work. The Army's Ballistic Research Laboratory funded the project. The Army signed the construction contract on June 5, 1943, and work began in secret under the code name "Project PX." The mathematician Herman Goldstine oversaw it for the Army.
The unveiling
A press briefing took place on February 1, 1946. On the evening of February 14, the machine was shown to the public, with demonstrations of its speed, and on February 15 it was formally dedicated at the university. The New York Times described it as an amazing machine bringing electronic speeds to mathematical tasks for the first time.
The trajectory demonstration that impressed reporters was the work of women who barely appeared in the coverage. Betty Snyder and Jean Jennings developed the trajectory program, and Marlyn Wescoff and Ruth Lichterman carried out the public demonstration. None of the six women programmers was invited to the dedication dinner. Their names, along with those of Kathleen McNulty and Frances Bilas, were largely forgotten for decades; in 1997 the six joined the Hall of Fame of Women in Technology International.
How ENIAC worked
ENIAC was a collection of units, about 40 panels arranged in a U shape around the room. Twenty of them were accumulators, each able to store a ten-digit decimal number and add or subtract. Others multiplied, divided or took square roots. A master programmer unit controlled the sequence, and punched cards carried numbers in and out.
There was no keyboard and no stored program. To "program" ENIAC, the operators plugged cables between panels and set thousands of switches, routing numbers from one unit to the next in the right order. Setting up a new problem could take weeks. Here is roughly what that looked like for a simple job:
| Step | What the programmers did |
|---|---|
| 1. Break the problem down | Turn the math (say, a trajectory) into a long sequence of additions, multiplications and tests |
| 2. Assign hardware | Decide which accumulator holds each number and which unit does each operation |
| 3. Wire it | Connect panels with cables so results flow to the right place at the right moment |
| 4. Set switches | Enter constants and the number of times each step repeats |
| 5. Test and debug | Run small cases, trace errors through the panels, fix wiring |
Speed was the payoff. Each accumulator could do 5,000 additions a second. A trajectory that took a skilled person about 12 hours with a desk calculator took ENIAC about 30 seconds. Eckert also solved the biggest worry, tube failure, by running the tubes well below their rated power so they burned out less often.
Try it: Pick a small calculation, such as the total cost of a shopping list with tax. Write it as a numbered list of single operations (add, multiply, compare) and decide where each intermediate result would be "stored." You have just done step 1 and 2 of ENIAC programming.
Why it mattered
ENIAC proved that a large electronic computer could be built and kept running, which convinced others to build better ones. In 1947 it moved to the Army's Aberdeen Proving Ground in Maryland, where it ran until October 1955, working on ballistics, weather prediction, cosmic rays, wind-tunnel design and more.
Its biggest weakness pointed the way forward. Because programs lived in cables, changing tasks was slow. Even before ENIAC was finished, its designers were planning a successor, EDVAC, that would store instructions in memory alongside data, the idea at the heart of almost every computer since. ENIAC itself was modified in 1948 to run a simple form of stored program.
How does ENIAC compare with the computers that run AI today?
The gap is hard to picture. ENIAC used about 150 kilowatts and filled a room; its 20 accumulators held twenty ten-digit numbers. A 2019 smartphone fits in a pocket, holds billions of numbers in memory and performs billions of operations every second. The chips used to train neural networks today, such as graphics processors, run many calculations in parallel and are faster again.
Yet the basic recipe has not changed. Modern AI is, underneath, an enormous number of simple arithmetic steps, mostly multiplications and additions, carried out at electronic speed. A language model such as GPT-2 repeats those operations billions of times to produce a paragraph. ENIAC showed in 1946 that machines could do such arithmetic tirelessly and fast; today's AI is what happens when that ability is scaled up by many orders of magnitude.
The ENIAC story also carries a human lesson. The machine was famous, but the people who programmed it were nearly invisible. In 2019, when we talk about an AI system, it is worth asking the same question: whose work, from engineers to the people who label training data, actually made it run?