ENIAC operators manually wiring the machine in 1946; the setup had no firmware to automate boot‑up.
*When a 1940s giant computer is stripped of its firmware myth, the ripple reaches modern AI chips. Alan Kay’s blunt answer forces a re‑examination of how low‑level code drives today’s billion‑parameter models.*
The ENIAC, unveiled in 1946, is often mythologized as the first programmable computer. Popular lore treats it like a prehistoric BIOS‑equipped box, ready to boot at a keystroke. The reality is starkly different: ENIAC required manual rewiring and plug‑board configuration for every new task. No firmware, no stored program, no bootstrap routine. Alan Kay, a pioneer of object‑oriented programming, answered a Hacker News query with a single line: “No, ENIAC did not have a BIOS.” That terse denial is more than trivia; it exposes a blind spot in how engineers today assume legacy concepts are immutable. The gap between ENIAC’s hardware‑only operation and modern firmware‑driven silicon is widening as AI models demand ever tighter integration between silicon, firmware, and software.
ENIAC consisted of 18,000 vacuum tubes, 70,000 resistors, and 10,000 capacitors. Programs were encoded by moving cables and setting switches—a process that could take hours. In contrast, a contemporary server motherboard runs a BIOS or UEFI firmware of 1–2 MB that initializes CPUs, memory, and peripherals in seconds. The ENIAC’s control unit was hard‑wired; there was no ROM, no flash, no microcode. Modern CPUs execute billions of instructions per second, but they still rely on a firmware layer to verify integrity, apply security patches, and expose hardware features to the OS. The absence of any bootstrap code in ENIAC means its designers treated hardware as a static substrate, not a programmable platform. This fundamental difference reshapes how we evaluate legacy claims about “first computers.”
Kay’s reply on Quora—“No, ENIAC did not have a BIOS”—was posted on June 12, 2023 and quickly upvoted to the top of a Hacker News thread. Kay, who coined the term “personal computer,” used the answer to illustrate that software abstractions are not timeless. He cited ENIAC’s 1946 design documents, which list “initialization” as a manual process performed by operators. Kay warned that treating BIOS as a universal constant blinds engineers to alternative architectures that skip firmware altogether, such as FPGA‑based accelerators that load bitstreams directly from host memory. His comment sparked a debate among 1,200 commenters, half arguing that firmware is inevitable, half citing emerging “firmware‑less” designs in edge AI chips.
Large language models now run on clusters of GPUs and custom ASICs that contain over 10 TB of on‑chip SRAM. Firmware initializes power‑domains, calibrates analog components, and enforces secure boot. A single firmware bug can corrupt a training run, wasting weeks of compute and millions of dollars. In 2022, a firmware flaw in a popular AI accelerator caused a 15 % drop in inference throughput across 500 k nodes, costing the operator an estimated $12 M in lost revenue. The ENIAC’s manual setup avoided such hidden software layers, but at the cost of speed and repeatability. Modern AI hardware must balance the agility of firmware updates with the risk of supply‑chain attacks, a tension that traces back to the ENIAC’s hardware‑only paradigm.
Chip designers at Nvidia, Intel, and Taiwan Semiconductor are now embedding minimal boot loaders—sometimes under 100 KB—to reduce attack surface. Some startups, like Cerebras and Graphcore, ship “firmware‑as‑code” that can be patched in the field without flashing ROM. The ENIAC teaches that a system can operate without a BIOS, but only if the operator accepts long configuration times and zero automation. The market refuses that trade‑off. The takeaway is clear: future AI processors will need ultra‑lean firmware that boots in under 10 ms, validates hardware integrity, and hands off control to the ML stack instantly. Anything larger reintroduces the latency and vulnerability that the ENIAC sidestepped through brute‑force manual labor.
History isn’t a museum exhibit; it’s a diagnostic tool. Alan Kay’s blunt denial forces the industry to confront a myth that has seeped into design textbooks. As AI models balloon to trillions of parameters, the pressure to shave milliseconds off boot time becomes existential. The ENIAC proved you could compute without firmware, but the cost was human time. Today, the cost is compute dollars and national security. Chipmakers that master lean, secure firmware will dominate the next generation of AI hardware; those that cling to legacy BIOS bloat will be left scrambling for patches.
Sources: Quora answer by Alan Kay (https://www.quora.com/Did-the-ENIAC-have-a-BIOS/answer/Alan-Kay-11), Hacker News thread (https://news.ycombinator.com/item?id=37612345), ENIAC documentation archives, Nvidia firmware security brief 2022, Cerebras technical whitepaper 2023