In 1999, on a page selling a product called TITAN kiOSK, we described the internet kiosk as "an Internet Payphone." We meant it as a promise: that web access would roll out across the world the way coin telephones once had, and that whoever owned the payment layer beneath them would own something large. It turned out to be a prophecy of a different kind. The kiosk did share the payphone's fate. Just not the one we were selling.
This is a first-hand account of a small Scottsdale company's four-year attempt to build the settlement layer for public internet access, told from inside it, and sourced to the company's own pages as they were captured at the time. It is also, in hindsight, a clean study in a particular kind of failure: being right about the architecture and wrong about the machine.
One company, three names
Tempest Telecommunications, founded in 1996 in Scottsdale, Arizona, began as a dial-up roaming service for travelers. The product was TITAN, the Tempest Internet Telecommunications Access Network, and by 1997 its pitch was already a clearinghouse, not an internet service. A traveler carried a TITAN card; an ISP forwarded authentication requests by proxy RADIUS to TITAN's servers; TITAN authenticated the user, metered the session, billed the card issuer, and settled payments between the ISP whose customer was roaming and the ISP whose access numbers they had dialed. TITAN guaranteed, by contract, that it would never operate as an ISP itself, and it sold ISPs on exclusive territories: one coverage area, one local provider, no channel conflict.
None of that was unique. Dial-up roaming aggregation was a real category in 1997, and its leaders, iPass and GRIC, both founded around the same period, ran the same proxy-RADIUS-and-settlement mechanism at far greater scale. TITAN was a small shop in a race it had not invented and was not winning. What it had was a correct instinct about where the value sat: in the neutral middle, holding identity and money, competing with neither side.
By April 1999 you can watch the company hunt for its next market. The TITAN card had become a five-service bundle (dial-up roaming across more than 150 countries and 3,000 access points, a toll-free calling-card voice service, a cyber-cafe network, rented GSM handsets, and a direct-dial long-distance product, UniDial, with residential minutes starting at 9.9 cents), with a sixth listed as "coming soon": a public-access terminal and kiosk network. The single-purpose 1997 pitch had become a catalog. That is usually what a company looks like when the floor is moving under its primary product, and both the calling-card and dial-up businesses were deregulation arbitrage, the kind of margin that compresses the moment the underlying telecom regime stabilizes.
The "coming soon" kiosk line is the one that mattered. Within months it had its own sub-site, TITAN kiOSK, and then its own name. By late 1999 or early 2000 the brand had changed from TITAN kiOSK to PATN, Public Access Terminal Networks, while still running inside Tempest. The PATN-branded software pages routed their contacts to tempestcom.com addresses, and the settlement copy still read that Tempest Telecom would make monthly payment for kiosk usage and bill the card provider. PATN was a Tempest product before it was a company. In March 2000 it incorporated as a separate business, with its own switch on a multi-homed OC3 in Fremont, direct peering at MAE West, PAIX, the PacBell NAP, and NASA/Ames, and a projection of 300,000 kiosks by 2003. It went dark by roughly 2003. Tempest's own satellite-phone tail ran on until about 2012.
The system itself
It is worth being specific about what was actually built, because it was a real product and not a slide deck. TITAN kiOSK, later the PATN system, split into a frontend and a backend. The frontend, the Control Software, sat on each terminal as an add-on to the kiosk's existing Public Browser Interface rather than as a reinvented browser; it was compatible with Netshift, the dominant lockdown software of the period, and customizable for others. It accepted magnetic-stripe or smart cards, ran on any connection from a 56k modem to satellite, metered each terminal at a rate keyed to its static IP, logged every event, offered a vandal-resistant on-screen PIN pad, and included a test mode that ran the kiosk with the central server stubbed out for debugging. The roadmap named biometric authentication (eye scan, thumbprint) as a future replacement for the PIN. The backend, the AAA Platform, gave resellers authentication, authorization, and accounting against their own user databases, reachable over SSL.
The smartest single decision in the whole venture is buried in the provider documentation, stated twice and almost defensively: these were not stored-value cards. The company cited Visa's own view that stored-value would not see wide use for over a decade, and chose instead to make the card a near-free magnetic-stripe identifier (pennies each, against dollars for a smart card) with all of the balance, intelligence, and authority living server-side, in the issuer's database, reached through distributed AAA. That was correct in a way that outlived everything else here. The on-card-cash experiments of the late 1990s, Mondex chief among them, failed exactly as Visa predicted. The model that won, a cheap credential as identity token with the account in the cloud, is the architecture of the modern internet. The company articulated why in 1999.
Right architecture, wrong machine
Read end to end, the four years are not really a sequence of products. They are one architecture, a neutral clearinghouse that holds authentication and settlement and competes with neither side, applied to three different machines in turn: the dial-up modem, the cyber-cafe and kiosk, and finally the kiosk clearinghouse itself. And the same force deleted each machine: universal personal connectivity. Home broadband, and then decisively the smartphone, removed the reason to roam onto anyone else's access point, whether that point was a modem bank, a cafe workstation, or a public terminal.
The kiosk thesis failed in a specific and instructive way. The pitch, that public internet kiosks would replace public telephones, that there would be 100,000 terminals in the US alone by 2002, was a straight-line extrapolation from the payphone. The mental model was that the internet kiosk would deploy the way the coin telephone had. But the payphone was not replaced by a better shared terminal. It was annihilated by a personal device that made the entire category of shared public access obsolete. The kiosk followed the payphone's curve, all right: its death curve, a few years behind it. "An Internet Payphone" was the most accurate line on the page, and we read it backwards.
The deepest irony is in the part the company got right. The server-side-identity model (cheap credential, account in the cloud) is precisely what made both the dedicated card and the dedicated terminal unnecessary. Once every traveler carries a personal identity token with its own connection, you need neither a separate card to swipe nor a shared machine to swipe it at. The architecture we correctly bet on is the architecture that erased the product we built on top of it.
The genuine foresight, in the end, was not in any single product but in the shape of the bet: that federated identity and usage-based settlement through a neutral clearinghouse was where communications was heading. That pattern did win. It is Wi-Fi roaming; it is iPass's own later life; it is OpenRoaming and eSIM roaming today. The mechanism was right. The substrate, the public kiosk, was wrong, and the timing put a kiosk-dependent company into the market in the exact quarter the dot-com financing for kiosk rollouts began to disappear.
Why this is here
The consumer-facing layers of the 1990s internet (phone cards, internet cafes, dial-up roaming networks, the first wave of consumer satellite handsets) are fading from sourced public reference faster than the infrastructure layers, because the trade press that covered them has largely vanished or gone behind paywalls. This account is published as primary-source material: named entities, real dates, the actual mechanism, and an honest accounting of what was right and what was not. Corrections and first-hand additions from anyone who worked the same trade are welcome.
Part of the Tempest unified-platform architectural arc, 1997-2012. See One Account, Many Networks for the synthesis across the production service types, and the TITAN Global Communications Card for the physical instrument that carried the four-service bundle.
Researched and written by Jason Jacoby, who operated inside this industry at Interglobe and Tempest Telecommunications, and who now runs AgaveIS, a web and software development practice in Scottsdale, Arizona. Tempest is no longer an operating business; this page is part of a primary-source archive of the 1989-2012 international travel-connectivity industry.