Email and the Killer App That Broke the Price of a Message

A feature added to a program nobody had asked for

ARPANET connected its first four nodes in 1969, built to let research institutions share computing power that was still scarce and expensive. Nothing in that founding purpose required a way to write a letter to a colleague. Mail programs already existed on the individual mainframes hooked to the network, but they worked the way office mail slots work: a user could leave a message for anyone else logged into the same machine, and no further than that.

Ray Tomlinson, an engineer at the contractor Bolt Beranek and Newman, changed that in 1971. He took an existing local mail program and combined it with an experimental file transfer utility so that a message could be routed to a user account on a different host computer somewhere else on the network. The specific problem he solved was addressing: once a message could travel to another machine, the software needed a way to say which machine and which user. Tomlinson chose the @ symbol, sitting unused on the keyboard, to separate the two, a convention that has not changed since (Wikipedia, "Ray Tomlinson", 2026).

It was, by most accounts, a modest piece of engineering, not a mandate from ARPA leadership or a planned product. The distinction matters for the economic story that follows. The internet's first mass-adopted application was not commissioned. It was a convenience that spread because it was useful, which is a different and in some ways more durable kind of adoption than one driven by policy.

ARPANET itself grew out of Cold War era research funding, the same institutional context that produced early packet-switching theory and the network's resource-sharing mandate. Nothing about that funding history explains why a defense-research network ended up carrying, within a couple of years, mostly personal and institutional correspondence rather than the resource-sharing traffic it was built for. The gap between a network's designed purpose and its adopted purpose is the recurring shape of this history, and email is one of its earliest and cleanest examples: a general-purpose channel, once opened, tends to fill with whatever its users find most valuable, regardless of what its funders had in mind.

The traffic record: the network filled with letters, not orders

What happened next is the more striking half of the story, though it rests on a figure historians treat with some caution. A study commissioned by ARPA, dated by most secondary accounts to around 1973, and variously attributed to ARPA director Stephen Lukasik or a review by the MITRE Corporation, reportedly found that email made up roughly three quarters of all traffic moving across the ARPANET (Paleofuture, 2016, aggregating primary ARPA-era accounts). No single, fully sourced primary document has been located and republished with the exact figure and its precise date, which is why it is treated here as a contested but widely repeated number rather than a settled one.

What is not in dispute is the direction of the finding and the institutional reaction to it. Larry Roberts, the ARPANET program manager, reportedly required that DARPA-related business correspondence be conducted over email, a policy that pushed adoption across every institution already connected to the network and that, by some accounts, drew pointed questions from within the Defense Department about why its research network had become a messaging system (Paleofuture, 2016).

The lesson generalizes past this one network. A communications system commissioned for one purpose, here, coordinating scarce computing resources among defense researchers, was captured almost immediately by its users for the purpose people actually wanted, which was writing to each other faster and cheaper than any existing channel allowed. That capture, not the original mission, is what made the ARPANET economically significant beyond its own research community.

The scale of that capture is also what makes email a better candidate than any single defense application for the title of the network's first real economic driver. Resource sharing, the ARPANET's founding justification, involved a handful of institutions logging into a handful of remote machines. Email, once Tomlinson's addressing scheme existed, involved every user of every connected institution writing to every other user, a use case that scales with the number of people on the network rather than with the number of expensive mainframes on it. That difference in scaling is, in miniature, the same economic logic that would later make consumer internet access, and not enterprise computing, the larger business.

What email actually did to the price of a message

Before Tomlinson's work, a message between two people separated by distance had a real cost. A letter cost postage and days of transit through a national postal system. A telex message required a subscription to a dedicated telex network, itself a walled garden with its own directory and per-message charges. A long-distance telephone call was billed by the minute, at rates that made a five-minute international call a deliberate expense. Each of these was a mature, capitalized industry: postal services, telex networks, and long-distance carriers all extracted a price for moving a fixed unit of information across a distance.

Email, once it existed on a shared packet network, collapsed that price toward the marginal cost of the bandwidth the message consumed, which for a short text message was close to nothing. The system underlying it, described in later technical accounts as a store-and-forward model, meant sender and recipient never had to be connected to the network at the same moment, unlike a telephone call. A message could sit on a host waiting to be read, at no incremental cost for the wait (Wikipedia, "Email", 2026). That single structural property, asynchronous delivery at near-zero marginal cost, is what let globally distributed collaboration become an everyday habit rather than an occasional expense reserved for urgent business.

It is worth being precise about what changed and what did not. The fixed costs of building and running a network like the ARPANET, and later the wider internet, were real and were paid by governments and later by commercial carriers. What collapsed was the marginal cost per message once that infrastructure existed, the price a sender faced for the next letter after the first. That is the number a postal clerk, a telex operator, and a long-distance billing system had each been charging for, and it is the number email drove toward zero.

This is the mechanism economists describe when a new technology moves an activity from a priced, metered footing onto a shared, largely fixed-cost one. A postal service, a telex network, and a long-distance carrier each depended on charging per unit of correspondence to recover the cost of the physical infrastructure moving that correspondence: trucks and sorting offices, dedicated teleprinter lines, and circuit-switched telephone exchanges. Once a packet network already existed for other reasons, and once a piece of software could address a message across it, the incremental unit of correspondence no longer needed a unit price to justify itself. The institutions built around charging that price did not disappear immediately, but the economic ground under them had already shifted.

The geopolitical thread: a channel that did not need a state's post office

A postal service is, among other things, an instrument of state control over correspondence. Mail crosses a border through customs and through a national post office's handling, a chokepoint that can be taxed, inspected, delayed, or in some periods and places, opened. Telex and international telephony were similarly mediated by state-linked or state-regulated carriers, each charging its own tariff for a cross-border connection.

Email, running over a packet network that treated a message to another country the same as a message across a hallway, did not require that mediation. Two people or two businesses on different sides of a border could correspond, and later transact, without either message ever passing through a postal system that either government controlled. This is a modest claim, not a sweeping one: the underlying network still depended on national telecommunications infrastructure and, eventually, on national regulation of that infrastructure. But the specific chokepoint of the letter carrier and the customs stamp, the mechanism by which a state had historically been able to see, delay, or tax personal and commercial correspondence crossing its border, no longer applied to a message sent by email. That is a real erosion of state mediation over cross-border communication, achieved not by policy or treaty but by a cost structure that made the old channel simply slower and more expensive by comparison.

The pattern recurs across the history this publication has traced elsewhere: control of the dominant medium of an age has repeatedly meant control of the economy and the politics that medium carries. A postal monopoly was, for centuries, one lever of that control. Email is an early and clear case of a technology built for one narrow institutional purpose that ended up, within its own users' hands, loosening that particular lever, well before anyone was designing the internet with geopolitics in mind.

The practical beneficiaries of that loosening were, at first, the same research institutions that already had ARPANET access, and only later a wider public as the network opened commercially and internationally through the 1980s and into the 1990s. A researcher in one country could correspond with a collaborator in another without either message being handled by a postal service, a distinction that mattered little while the network stayed small and institutional, and mattered a great deal once the same channel carried business correspondence, contracts, and eventually payment instructions across borders that had never granted either party's state a look at the message in transit.

The undertow: a marginal cost of zero cuts both ways

A message that costs the sender almost nothing to send is not only a gift to the person who wants to write to a colleague across an ocean. It is also, obviously in hindsight, an opportunity for anyone who wants to reach many people at once without paying for the reach. On May 3, 1978, a Digital Equipment Corporation marketer named Gary Thuerk sent an unsolicited promotional message advertising DEC's DECSYSTEM-20 product line to roughly 400 of the ARPANET's approximately 2,600 addressable accounts, a message that reached essentially every user on the U.S. West Coast segment of the network at the time (The Mary Sue, citing Computer History Museum and Guinness World Records accounts, 2024). It is the earliest documented case of what would later be named spam.

The complaints were immediate, and by several accounts sharp. What matters for the economic thread is what the campaign reportedly produced anyway: approximately 12 million dollars in DEC computer sales attributed to that single unsolicited message (The Mary Sue, 2024, citing Computer History Museum records). That figure is treated here as contested, since it comes from a secondary account of a decades-old campaign rather than an audited sales report, but the underlying logic is not in doubt and is easy to state plainly: if reaching one more recipient costs almost nothing, then even a very low response rate can be profitable, a piece of arithmetic that has since built a real, if unwelcome, industry.

That is the two-sided finding this thread demands. The same structural property, near-zero marginal cost per message, that let a researcher in one country write cheaply to a colleague in another, or let a small business reach a customer without paying postage, also handed anyone willing to send unsolicited bulk mail a profitable business model. Every medium this publication has examined has both liberated and been captured; email liberated correspondence from postal and telephonic tariffs, and within seven years of its invention, it had already been captured by a use its inventor never intended, an undertow that runs, by later industry estimates, into the billions of dollars a year in filtering, lost productivity, and outright fraud.

The scale of that undertow is what eventually forced a policy response. Legislatures spent much of the following two decades trying to write rules for a channel whose entire economic logic ran against per-message pricing, and whose senders could route around any single jurisdiction's enforcement by relaying through a server elsewhere. The details of that regulatory history sit outside this article's scope, but the shape of the problem was set the moment the marginal cost of a message hit zero: a channel cheap enough to liberate correspondence from postal tariffs is also cheap enough that no volume of unwanted mail through it is ever expensive for the sender to try.

The line to the present

Email's two founding facts, that Ray Tomlinson is also credited with contributing the TCP three-way handshake, one of the pieces of protocol plumbing that underlies the wider internet including the web that came after it, and that the same near-free channel was profitable for legitimate senders and spammers alike within a decade, both point toward the same present-day question this publication returns to (Wikipedia, "Ray Tomlinson", 2026). When the cost of reaching someone falls toward zero, the constraint that decides who actually gets through shifts from price to attention, and eventually to whichever system is doing the sorting.

Email itself answered that shift with spam filters, a decades-long arms race between senders and the systems that decide what reaches an inbox. The current chapter of that same problem is not an inbox but an AI answer engine deciding, on a buyer's behalf, which business gets named at all. The mechanism differs. The underlying economics do not: whoever controls the sorting layer of a cheap, high-volume channel controls a meaningful share of the value that channel creates, a pattern this series has traced from the postal monopoly through the telegraph and now into the systems that read a page before a person does.