At 10:30 p.m. on 29 October 1969, a graduate student named Charley Kline sat at a Scientific Data Systems Sigma 7 computer in Room 3420 of Boelter Hall at UCLA and tried to log into a machine 350 miles away at the Stanford Research Institute. He typed an L. The receiving end, monitored by SRI programmer Bill Duvall over a telephone line, confirmed the L. Kline typed an O. Confirmed. He typed a G. The system crashed.
The first word ever sent over the network that became the internet was LO.
Kline rebooted, finished the login about an hour later, and wrote a brief entry in the IMP log noting the successful host-to-host connection. No champagne. No press. No photograph of the moment. The men who had just lit the fuse on the largest communications shift in human history went home to sleep.
The machine that called itself an IMP
The hardware that made the call was a refrigerator-sized box called an Interface Message Processor, or IMP. Bolt Beranek and Newman, a Cambridge, Massachusetts consulting firm, had built it under contract from the Defense Department’s Advanced Research Projects Agency. The IMP at UCLA was serial number 1. It had been wheeled into Leonard Kleinrock’s lab on 30 August 1969, two months before the first message, and weighed roughly 900 pounds.
The point of the IMP was to handle the awkward problem of getting computers from different manufacturers to talk to each other. In 1969, a Sigma 7 and an SDS 940 — the Stanford machine — were as mutually unintelligible as a typewriter and a clarinet. The IMP sat between them and translated, breaking messages into packets and reassembling them on the other end.
That packet-switching idea, in which messages travel as small numbered chunks rather than as continuous streams down a dedicated line, is the foundation of every email, video call, and tweet sent since. Paul Baran at RAND and Donald Davies at Britain’s National Physical Laboratory had each arrived at the concept independently in the early 1960s. ARPANET was the first place anyone actually built it at scale.
Why the Pentagon paid for it
The funding came from ARPA, the Pentagon research arm created in 1958 in response to Sputnik. The standard origin story holds that ARPANET was designed to survive a nuclear strike by routing around damaged nodes. That story is partly mythology — Baran’s RAND work on survivable communications was a separate project — but the military interest in resilient, decentralised networking was real, and the lineage runs straight from ARPANET to the encrypted military communications of today.
The more honest version: ARPA’s Information Processing Techniques Office, led at various points by J.C.R. Licklider, Robert Taylor, and Larry Roberts, wanted to let the agency’s expensively funded research computers share time and resources. Buying one mainframe per university was wasteful. Linking them together meant a researcher at MIT could run a job on a machine in Utah without flying out to use it.
Taylor got the project approved in 1966 in a meeting with ARPA director Charles Herzfeld. Taylor walked out of the office with the funding secured.
The four-node network
By December 1969, ARPANET had four nodes: UCLA, the Stanford Research Institute, the University of California Santa Barbara, and the University of Utah. The map of the entire internet fit on a napkin. Traffic moved over leased 50 kilobit-per-second lines from AT&T, which charged enough that the researchers joked about the cost-per-keystroke of their experimental logins.
Kleinrock’s UCLA group had been chosen as the first node partly because he had done his MIT doctoral dissertation on queueing theory in communication networks — the mathematics of how packets would behave in traffic. Stanford was second because Douglas Engelbart’s lab there had built the NLS system, an astonishingly early hypertext environment, and Engelbart had already demonstrated the mouse, hyperlinks, and screen-sharing in his 1968 “Mother of All Demos.”
The original plan for that first night was prosaic. Kline would log into the SRI machine remotely and run a program. To do that he had to type LOGIN, and the SRI software, helpful for its era, would auto-complete the rest after he typed the L. Kline typed L. SRI’s machine echoed back L-O-G-I-N. He typed O. It echoed O-G-I-N. He typed G — and the receiving system ran out of buffer memory and dropped the connection.
The accidental poetry of LO
Kleinrock has often noted in interviews that the network’s first message was remarkably prophetic, as ‘lo’ suggests the phrase ‘lo and behold.’
The fact that nobody noticed at the time is almost as telling as the message itself. The logbook entry is matter-of-fact. There is no celebratory cable to ARPA, no memo to Taylor, no announcement. Kline and Duvall were trying to debug a connection, not stage a historical moment. The IMP log at UCLA, now preserved at the university’s archives, records the event in the same handwriting used for routine equipment checks.
Contrast that with Samuel Morse’s first telegraph message in 1844 — “What hath God wrought,” chosen from the Book of Numbers by the daughter of the Commissioner of Patents and transmitted with full ceremonial weight from the Capitol to Baltimore. ARPANET’s first message had no scriptural pedigree. It had a buffer overflow.
From four nodes to four billion users
The growth curve from that night is the part most people half-know. ARPANET added nodes through the 1970s. Ray Tomlinson sent the first network email in 1971 and chose the @ symbol because it was unlikely to appear in a username. Vint Cerf and Bob Kahn published the TCP/IP specification in 1974, giving the network a common language. On 1 January 1983, ARPANET formally switched to TCP/IP, the protocol still running underneath every web page loaded today.
Tim Berners-Lee wrote the first web browser at CERN in 1990. The National Science Foundation lifted the commercial-use ban on its NSFNET backbone in 1991. The public internet grew from millions of users in the mid-1990s to billions by the 2020s. The line from Boelter Hall to a teenager scrolling TikTok in Jakarta is unbroken.
Charley Kline, the man who typed the L and the O, worked in software for decades, much of it at UCLA, and has given a handful of interviews about that night. He tends to deflect the historical-figure framing. He was a grad student doing a task his professor had asked him to do. The machine crashed. He fixed it. He went home.
What the IMP looked like, and where it is now
IMP number 1 is preserved at UCLA’s Kleinrock Internet History Center, in a glass case in the same building where it first ran. The metal cabinet is painted a military-issue beige-grey. Its front panel is dotted with toggle switches and small incandescent indicator lamps. It looks, accurately, like a piece of late-1960s defense electronics, because that is what it is — Honeywell DDP-516 minicomputer guts in a ruggedised case meant to survive shipment to a research lab.
Standing next to it, a visitor can see the cables that once ran to the Sigma 7. The Sigma 7 itself is long gone, scrapped decades ago for parts. The room where Kline sat, 3420 Boelter Hall, is now sometimes referred to as the birthplace of the internet on tours, though for years it was an ordinary teaching space with no plaque at all. If you want to understand how casually the modern world’s foundation was laid, the absence of that plaque for most of a half-century is the answer.
Anyone curious about the deeper plumbing — how packets actually move, why the @ sign ended up in email addresses, how a home router decides where to send each chunk of a Netflix stream — can keep pulling on the thread. A good starting point is the difference between the public-facing web and the internet that carries it, or the way packets get from one machine to another in the first place.
The crash that wasn’t really a failure
There is one small mercy in the LO story worth holding onto. The system that crashed was working correctly. The L went out, the O went out, both were echoed back across 350 miles of leased phone line at a then-impressive 50 kilobits per second, and a faraway computer in Menlo Park acknowledged receipt before its memory ran out. Two letters had successfully crossed a national distance between two machines that had no business understanding each other. The crash came after the proof, not before it.
Fifty-six years later, the descendants of that two-letter message move at massive volumes of global internet traffic. The growth from LO to that scale is the kind of curve that humans are bad at picturing, which is probably why nobody at UCLA thought to take a photograph that night. They were tired. The connection was up. The log was written. The lights on the front of IMP number 1 kept blinking in the empty room.
