Summary

  • A 28 February 1995 notice set 30 April as the end of NSFNET Backbone Service at named sites and four Network Access Points.
  • The same notice said many recipients already depended on an alternate network for their primary path, making the deadline a service boundary inside a longer migration.

NSFNET did not disappear in one switch-flip. On 28 February 1995, Merit’s Elise Gerich forwarded a formal 60-day termination notice to the North American Network Operators’ Group (NANOG). It gave a precise end date: 30 April, at midnight in each site’s local time zone. It also contained a revealing qualification. “Many” recipients, the notice said, already depended on an alternate network for their primary communication path.

That sentence changes how the headline date should be read. It documents movement before the service contract ended, but it does not say how many networks had moved, which alternatives each used, or whether every route had changed. The wording is qualitative. It supports a transition already under way—not a claim that the entire Internet moved at once.

The notice listed engineering network service sites (ENSS) from Palo Alto and Seattle to Cambridge and College Park. It named service cessation at the Ameritech, MFS, Pacific Bell and Sprint Network Access Points as well. Atlanta’s ENSS 138 was omitted because it had received an earlier termination notice. This was a schedule for identified facilities and services, not a universal off-switch for everything called NSFNET.

NSF had planned a successor structure well before the final weeks. Its 1993 Solicitation 93-52 described Network Access Points, a Routing Arbiter, regional-network providers and a separate very-high-speed backbone service for research. The design separated functions: commercial networks could exchange traffic, a routing-arbiter function could help coordinate routes, regional providers could connect users, and selected research traffic could continue on a distinct service. The solicitation records a plan and procurement framework; by itself, it cannot prove how each network used the architecture in practice.

That distinction matters because NSFNET named more than one thing. It could mean the program, the national backbone service, regional networks connected to it, or research services created around the transition. Merit’s later account dates the backbone service’s end to April 1995 while describing final dismantling as extending to April 1996. The physical dismantling date and the service end date are not interchangeable.

Nor did the public Internet begin on 1 May. Commercial providers and alternative networks were already carrying traffic; the notice says so directly. After 30 April, traffic continued over regional and commercial backbones. NSF also supported a separate very-high-speed network for selected research. The institutional boundary moved while the larger network remained in use.

The historical significance is therefore more precise than “the government shut down the Internet.” NSF ended one backbone service on a scheduled, location-specific timetable after a replacement arrangement had been designed and as some networks had already shifted their primary paths. That transition left operators to coordinate real routing and connectivity across multiple networks. The record tells us the date and some of the handoff’s contours; it does not provide a complete, route-by-route migration ledger.

Heng Lu’s Note 65, “Running-Code Primacy,” offers an editorial question about operational continuity during institutional change. It is not evidence for this chronology. The period record itself supports the narrower conclusion: a formal closing date can organize a transition without making every network’s cutover simultaneous.

Sources