Summary

  • Under Stephen Wolff, NSF treated commercialization and privatization as different policy choices. The NSFNET Acceptable Use Policy protected a research-and-education mission while exposing where commercial demand needed other capacity.
  • The policy did not create the commercial Internet by itself. Regional seed funding, private operation, neutral interconnection and the 1995 federal exit converted a contested boundary into a workable transition.

The important line in early commercial Internet history was not drawn on a map. It ran through the purpose of a packet. NSFNET’s backbone existed to support research and education, yet the institutions connected to it were increasingly surrounded by commercial users, suppliers and services. A university might collaborate with a company’s research laboratory. The same company might also want to sell an unrelated product across the network. One use could further the public mission; the other could consume subsidized capacity for private business.

This ambiguity made a short administrative rule unusually consequential. The 1992 Acceptable Use Policy said that use of the NSFNET Backbone should support open research and education among American research and instructional institutions. Research arms of for-profit firms could participate when they engaged in open scholarly communication. Other for-profit activity and personal business were not acceptable. The policy applied to the backbone, not automatically to every connected network; those networks were expected to establish their own rules.

Stephen Wolff led the National Science Foundation’s networking program while this boundary was being tested. He had taken responsibility for a system intended to connect supercomputing centers, campuses, regional networks and other research communities. NSFNET rapidly became a major national backbone. The Internet Hall of Fame credits Wolff with responsibility for its development, but the institutional record resists a lone-inventor story.

NSF supplied mission and seed funding; Merit coordinated the service; IBM and MCI contributed engineering and communications; regional networks assembled campuses and found additional money; private firms owned circuits and increasingly handled routine operation.

The boundary mattered because demand was already outrunning the categories used to govern it. RFC 1192, a report on a 1990 workshop sponsored by NSF and the congressional Office of Technology Assessment, recorded traffic growth of 15 to 20 per cent a month and repeated requests from commercial services. Letting all such traffic onto a subsidized backbone raised two objections at once: it might overload scarce capacity, and it might place private packet-network providers in unfair competition with a service underwritten for research.

At that workshop, Wolff made a distinction that is still often blurred. Commercialization meant allowing commercial users and providers to reach Internet facilities and services. Privatization meant eliminating the federal role in providing or subsidizing network service. A network could become more commercial while public support continued. Conversely, a government could shift subsidy toward institutions or researchers without abandoning the public purpose it wanted connectivity to serve.

That distinction turns the AUP from a morality rule into a question of market design. By reserving subsidized backbone capacity for a defined mission, NSF left other demand looking for lawful carriage elsewhere. But the rule was only one part of the signal. The physical circuits were already privately owned. Regional networks charged membership or connection fees, drew on state support and contracted out operations while retaining policy control. Public money covered only part of the system.

The emerging market was therefore neither a clean private replacement for a state network nor a spontaneous business sector untouched by public investment.

There was no frictionless way to move the boundary. RFC 1192 warned that a free backbone could discourage investment in competing backbones. It also warned that commercial providers could “cream-skim” dense research markets while nonprofit regional networks remained responsible for broad, expensive geographies. Some regionals were financially stronger than others. If subsidy disappeared before commercial capacity and interconnection were ready, institutions in less attractive markets could lose service rather than gain choice.

Enforcement was another weakness. Federal interconnection policy required traffic using NSFNET to conform to its rules, and routing eligibility could in principle reflect policy. Yet a later history of inter-domain routing noted that there was little or no technology to support the AUP’s legal framework. Packets did not arrive carrying reliable declarations of purpose. Compliance depended on institutions, contracts, routing relationships and judgments by program staff. The boundary shaped incentives even when it could not be reduced to a perfect filter.

The controversy around commercial use of shared facilities made accountability part of the design. NSF’s inspector general concluded in 1993 that the agency could allow commercial use under conditions protecting research service and returning profits to network infrastructure. It also criticized poor public notice and inadequate documentation of important decisions. The audit recommended public comment and a clearer legal basis for applying the AUP. A policy line that affects competitors is not legitimate merely because officials can explain it afterward; affected networks must be able to see how the line is interpreted and who benefits.

The eventual transition required machinery beyond acceptable use. NSF’s 1993 solicitation proposed Network Access Points where providers could exchange traffic, a Routing Arbiter to support policy coordination, assistance for regional providers buying service, and a very-high-speed backbone reserved for advanced research. Commercial backbones could compete, but common interconnection and continuing research obligations reduced the risk of fragmenting reachability. When expanding commercial service made the dedicated NSFNET backbone unnecessary, NSF retired it in 1995.

It would be wrong to compress that sequence into “Wolff privatized the Internet.” He did not write every rule alone, build every regional network or cause every market outcome. Commercial networks such as UUNET, PSINet and CERFnet were already developing; ANS introduced its own contested commercial arrangements; legislation, procurement, engineering and community practice all mattered. Wolff’s contribution is more precise: he articulated the choices that had to be separated, and managed a public program whose boundary made those choices difficult to evade.

The result was not a simple retreat of government. Public funding helped create users, skills, institutions and technical reach. The AUP protected the reason for that support while signaling that unrelated commercial demand needed a different economic path. Later interconnection policy kept competing paths from becoming isolated islands. Public withdrawal became credible only after a market could carry ordinary traffic without dissolving the research network’s mission.

That sequence offers a restrained lesson for current infrastructure policy. A public network, compute facility or data platform can crowd in private supply when it states what subsidy is for, where commercial use begins, how providers interconnect and how public obligations survive an exit. A boundary without enforcement becomes discretionary. Enforcement without open interpretation becomes favoritism. Exit without portability and universal-service protection can merely abandon the users that the market values least.

Wolff’s distinction therefore deserves to be remembered as governance, not slogan. Commercial participation can broaden a system before government leaves it. Public subsidy can move without disappearing. And the line between a public backbone and a market is not found in ownership alone; it is made through purpose, access, interconnection and the sequence in which responsibility changes hands.

Sources