TCP/IP and the Open Standard That Ate the Proprietary Networks

A protocol with no owner

By the mid-1970s, the ARPANET was running well as a single network, but it could not talk to the other networks appearing alongside it: packet radio links, satellite links, and the local networks springing up at universities and defense contractors. Bob Kahn moved to DARPA to solve exactly that problem, and working with Vint Cerf at Stanford, he formulated a scheme for joining separate networks into one, an approach the two called internetworking. Neither man built it in isolation. The design drew directly on work already underway elsewhere, most notably the CYCLADES project in France, led by Louis Pouzin, and Bob Metcalfe's research at Xerox PARC.

CYCLADES contributed something specific and lasting. It was the first network to put the job of reliable delivery on the computers at each end of a connection rather than on the network carrying the data, an approach now known as the end-to-end principle. That design choice let the network itself stay simple, and it shaped a core feature of the protocol Cerf and Kahn were building: the Internet Protocol would move packets without promising anything about their delivery, and the computers at either end would handle the rest. Metcalfe's Ethernet work fed the same current from a different angle, giving the pair a working example of a local network built from the start to plug into something larger rather than to stand alone under one vendor's control.

What set the resulting design apart from almost everything commercial vendors were building at the time was what did not happen next. Cerf and Kahn did not found a company around it, patent its core mechanism, or require a license to implement it. The specifications were published, argued over by a working community, and released for anyone to build against, whether the builder was a university, a government agency, or a rival hardware maker. It became, in effect, a piece of public infrastructure authored by two researchers under a government research program, not a product owned by a firm.

The design matured through the rest of the 1970s, and on January 1, 1983, a date remembered ever after as flag day, version four of TCP/IP was switched on for production use across the ARPANET, after the U.S. Department of Defense mandated it as the standard for all military computer networking. Every host that had not converted by that date lost its connection. What had been a research protocol became, overnight, compulsory infrastructure, and the network it ran on would go on to become the internet.

Two networks that tried to own the wire

TCP/IP was not competing against nothing. Through the same decade, the two largest computer makers in the world were each building their own answer to the same problem, and each answer was proprietary by design. IBM's Systems Network Architecture, introduced in 1974, connected IBM machines to other IBM machines on IBM's terms. Digital Equipment Corporation's DECnet, from 1975, did the same for Digital's own hardware line. Both were competent engineering. Both were also, by construction, a way to keep a customer's network locked to a single vendor's equipment, because a customer who wanted to expand or interconnect had little choice but to keep buying from the company that already owned their network.

That lock-in was the point. In a world of proprietary networking, the network itself was where the money sat. A vendor that owned the protocol owned the upgrade cycle, the support contract, and the switching cost a customer faced every time it considered a competitor. Value accrued to whoever controlled the wire and the boxes built to speak to it.

An open, unpatented internetworking standard broke that logic from underneath. Once TCP/IP existed and cost nothing to implement, a customer no longer had to pick a network vendor and stay loyal to it; any hardware that spoke the open protocol could join the same network as any other. Robert Metcalfe, who had helped shape the design current TCP/IP drew on, put this to direct commercial use: he co-invented Ethernet and, in 1979, co-founded 3Com to sell local networking hardware that plugged straight into the open standard rather than into a proprietary architecture owned by IBM or Digital. Competing on hardware that worked with everyone's network turned out to be a viable business in a way competing on owning the network no longer was.

The effect widened as the open commons grew. If the core of the network was free to join and technically neutral between vendors, the place left to compete, and to capture value, was everything sitting at its edges: the hardware that connected to it, the software that ran on top of it, and eventually the services built for the users the network could now reach without any single company's permission. That shift, from value sitting in the wire to value sitting at the edge, is the pattern that would go on to define the commercial internet for decades.

Governance by rough consensus

An open standard still needs a way to change and grow without a single company deciding its direction, and TCP/IP had one from early on. From the ARPANET's first days, technical proposals circulated as Requests for Comments, informal documents that anyone in the working community could write, argue over, and revise. The name undersold what the process became: the RFC series is the mechanism by which internet standards have been proposed and adopted by rough consensus ever since, and it still is.

Jon Postel edited that series through most of its early history and, from the network's inception until his death in 1998, administered the Internet Assigned Numbers Authority, the body responsible for allocating the addresses and numbers the protocol needed to keep functioning. Postel held enormous practical authority over how the network's plumbing was assigned, and he held it as an individual editor inside a research institution, not as an executive of a company with shareholders to answer to.

The internet's technical standards are maintained today by the Internet Engineering Task Force through that same open, consensus-driven process, an arrangement with no single corporate owner and no fee that decides whose proposal wins. The Internet protocol suite organizes communication into four layers, link, internet, transport, and application, and each layer's specification is a public document, not a trade secret.

That governance model was not the only one on offer. A more formal alternative existed, built by the kind of institutions that usually write technical standards for the world, and it lost.

Running code beats treaty law

The formal alternative was the Open Systems Interconnection model, developed under international standards bodies through the late 1970s and 1980s. OSI had what TCP/IP conspicuously lacked: a multilateral pedigree, sign-off from national standards organizations, and the backing that comes from being negotiated the way treaties are negotiated, with delegations representing many countries at once.

TCP/IP had none of that provenance. It had been authored inside a single country's defense department, without an international body's blessing and without any treaty behind it. What it had instead was a network already running on it, expanding through the 1980s while OSI was still being specified. TCP/IP predates and ultimately displaced the more formal, internationally negotiated OSI reference model as the working standard for global networking, despite OSI's institutional backing, because an open standard with running code beat a well-credentialed standard still being negotiated on paper.

The geopolitical consequence is easy to understate because it never took the form of a formal claim. No country signed a treaty ceding the internet's technical architecture to the United States. But a protocol suite written inside the U.S. Department of Defense became, by force of adoption rather than agreement, the de facto technical constitution any network had to speak to join a planetary system. Every government, company, and researcher outside the country of origin who wanted their network connected had to build to specifications authored, published, and for its first decades administered from inside the United States, Postel's stewardship of IANA among them.

That is a form of authorship power distinct from military or economic power, though it sits alongside both: whoever writes the rules of the layer everyone must use to participate sets terms nobody else voted on. The United States acquired that position not by treaty or conquest, but by being first to build the open version of the thing every network would eventually need, and by giving it away before any rival standard, OSI included, was ready to compete for the job.

Liberated and concentrated

The commons the standard created did keep expanding past its military origin. Access broadened in 1981 when the National Science Foundation funded CSNET, extending the network to computer science departments with no Defense Department connection, and again from 1986 through the NSFNET project, which carried the open TCP/IP commons deeper into the broader U.S. academic system and, from there, toward the research networks of other countries.

Each of those expansions is a case for the liberating side of an open standard. A university, in the United States or eventually abroad, did not need a license from IBM or Digital, or a treaty negotiated by its government, to join; it needed hardware that spoke TCP/IP, which by the 1980s a growing market of vendors, Metcalfe's 3Com among them, was already selling. Set against a proprietary network that required a vendor relationship just to join, that is a genuine widening of who could participate on equal technical terms.

The concentrating side sat underneath that widening participation. The institutions that ran the commons, chiefly IANA under Postel, and the bulk of the network's early hosts and funding, sat inside the United States for the whole period this history covers. Openness distributed the right to connect; it did not distribute the stewardship of the system, or, later, the commercial capture of the value flowing across it, nearly as evenly. The edge of the network, established above as where the money would go once the core stopped being ownable, filled disproportionately with American hardware makers, software firms, and eventually platforms, extending a pattern set at the protocol layer up through every layer built on top of it.

Historians of the network's origin add a third complication, a debate over credit rather than power. Accounts of TCP/IP's history that center on Cerf and Kahn alone risk treating their authorship as the whole story; the record credits CYCLADES and Louis Pouzin's end-to-end principle as a direct and specific input, not a footnote, which means the idea that let the network stay simple was French before it was American, even though the network that shipped it, administered it, and set its terms for the following decades was not.

What still runs underneath

TCP/IP won so completely that it became invisible. It is the layer carrying a video call, a bank transfer, and a query typed into an AI answer engine, and almost nobody using any of those services could name the protocol doing the work. That invisibility is itself a kind of proof: an open standard with no owner and no royalty succeeded so thoroughly that using the internet and using TCP/IP became the same fact, unremarked on by the people relying on it.

The pattern it set, an open technical layer nobody owns deciding who may connect on equal terms, has not repeated cleanly at every layer built on top of it. The web that runs over TCP/IP was itself built on an open, unpatented standard, a decision examined elsewhere in this series. But the newest layer determining who gets found on that same open network, the systems that read a page and decide whose business to name in an answer, is not open or neutral in the way the transport layer beneath it is.

TCP/IP will move the query and the answer without caring who wins; it has no opinion on which business gets named. The commons at the bottom of the stack stayed open. What was eventually built on top of it, including the systems now deciding who gets found, did not inherit that same openness by default, and that gap is where a business's visibility is contested today.