Dead End: WAP and the Mobile Web

Abstract

The Wireless Application Protocol was the mobile industry’s answer to a question everyone agreed was urgent: how do you put the internet on a phone? The answer that Nokia, Ericsson, Motorola and Unwired Planet arrived at in 1997 was a parallel web for phones, with its own markup language, its own browser, its own protocol stack, and a gateway in every carrier’s machine room. The first WAP service went live in the Netherlands in October 1999. By the end of 2000 a Nielsen Norman field study in London found that 70% of the users given WAP phones said they would not be using WAP in a year, and the British trade press had settled on the phrase “WAP is crap”. Japan’s i-mode, built on a subset of HTML and a revenue split that paid content providers, had 30 million subscribers by then. When Apple shipped the iPhone in 2007 with a browser that rendered ordinary websites, the argument was over.

The Problem WAP Was Solving

In 1997 mobile phones were voice devices. GSM networks carried data over circuit-switched connections at 9.6 kbit/s, about a sixth of what a 56k modem managed on a phone line, and the user paid for the time the connection was open rather than for the bytes moved. Screens were monochrome and tiny: the Nokia 7110, the first phone with a WAP browser, had 96 × 65 pixels. Memory was measured in tens of kilobytes.

An ordinary web page of the period, with images, tables and scripts, could not be fetched, stored or rendered on such a device. The mobile industry’s judgment was that the web therefore had to be rebuilt for the phone.

The WAP Forum and WML

Ericsson, Motorola, Nokia and Unwired Planet announced the initiative on 26 June 1997, published the architecture for public comment that September, and formally incorporated the WAP Forum in December 1997. WAP 1.0 followed in April 1998, after which membership was opened to the rest of the industry. Unwired Planet, which had built carrier data services since the mid-1990s, renamed itself Phone.com in 1999 and, after merging with Software.com in November 2000, Openwave.

The architecture was layered, and every layer was a replacement for something the web already had.

WML (Wireless Markup Language) replaced HTML. It was an XML application: strictly structured, no optional end tags, no layout tables, no scripting in the page. Content came in “decks” of “cards”; a deck was roughly a page, a card was a screen inside it, and the phone fetched a whole deck at once to save round trips. Images were possible but only in WBMP, a one-bit black-and-white bitmap format with no compression, chosen because the handsets of 1998 had no power to spare for decoding anything richer.

WMLScript replaced JavaScript with no DOM to act on. It could calculate, validate form input, and show alerts.

The WAP Gateway sat in the carrier’s infrastructure between phone and internet. The handset spoke WAP’s compressed binary protocol to the gateway; the gateway fetched content over ordinary HTTP, converted it, and pushed it down. That let WAP content live on standard web servers, and it put the carrier in the path of every request.

WAP 1.1 arrived in 1999 and WAP 1.2 in December 1999, with a conformance release in June 2000. WAP 2.0 went out for public review on 1 August 2001 and gave up most of what made WAP distinct: XHTML Mobile Profile instead of WML, TCP/IP and HTTP/1.1 instead of the WAP stack, TLS instead of WTLS. The WAP Forum itself was folded into the Open Mobile Alliance in 2002.

The Experience in Practice

The first commercial WAP service was launched by the Dutch operator Telfort in October 1999, in the same month as the Nokia 7110.

Speed. GPRS, which reached European networks in 2000 and 2001, delivered 20 to 40 kbit/s. Before that, every session began by dialling a data call. A request then travelled to the carrier’s gateway, out to the origin server, back, and through conversion before anything appeared on the screen.

Cost. Data was billed by connection time and later per kilobyte. Charges accumulated while nothing visible happened, which is a poor way to teach a habit.

Content. Carrier portals, the pages the browser button opened, were stocked with ringtones, wallpapers, horoscopes and weather. Ordinary websites built WAP versions reluctantly and maintained them worse.

The WAP Gap. WAP 1.x could not do end-to-end encryption. The phone spoke WTLS to the gateway; the gateway spoke TLS to the server; in between it decrypted and re-encrypted, so the traffic existed in plaintext inside carrier infrastructure. For banking and payments this was not a risk assessment but an architectural fact, and it is the problem WAP 2.0’s move to TLS was meant to end.

What the Field Study Found

Jakob Nielsen’s group gave WAP phones (Ericsson R320s and Nokia 7110e) to 20 users in London in the autumn of 2000, tested them at the start and end of a week, and had them keep diaries. Reading a TV listing took 2.6 minutes at the start of the week and 1.6 minutes by the end; checking local weather, 2.7 minutes falling to 1.9. Web experts asked beforehand had estimated under 30 seconds for such tasks. One participant worked out that buying a newspaper and throwing away everything except the listings would have been cheaper than looking up that evening’s BBC programmes on her phone. At the end of the week, 70% of the group said they would not be using WAP in a year. The Register had already run the headline “WAP is crap” that July, and ran “WAP is still crap” the following May.

i-mode in Japan

i-mode launched at NTT DoCoMo on 22 February 1999, built by a team under Keiichi Enoki, with Mari Matsunaga designing the service and its content strategy and Takeshi Natsuno running business development. Its markup was cHTML, a subset of HTML, so ordinary web tools and ordinary web knowledge transferred to it.

The larger difference was commercial. DoCoMo billed users for content and kept a 9% commission, passing 91% to the provider, and vetted providers before listing them on the official i-mode menu. That gave content businesses a working payment channel from day one, which is what WAP’s open-but-unbilled model never supplied. News, banking, games, restaurant reservations and ringtones appeared quickly because someone was being paid for them.

Growth followed: 100,000 subscribers within three months, one million by August 1999, ten million by August 2000, and around 30 million by May 2001, roughly 60% of DoCoMo’s subscriber base.

The comparison flatters i-mode on architecture, but the packaging mattered as much: packet pricing rather than metered call time, handsets DoCoMo specified and subsidised, and a menu that told users where to start.

The walled garden set the limit too. DoCoMo exported i-mode to 17 countries and it took hold in none of them, and when smartphones arrived the closed menu had no answer to the open web. DoCoMo announced the end on 29 October 2019 and shut i-mode down on 31 March 2026, together with its 3G network, 27 years after launch.

Real Browsers on Real Phones

WAP assumed the mobile web had to be a different web: simpler, smaller, shaped by 1997 hardware. The assumption held only as long as the hardware did.

Opera Mobile shipped in July 2000 for Psion’s EPOC devices, and Opera Mini followed as a pilot in Norway on 10 August 2005, going worldwide on 24 January 2006. Opera Mini put the work on Opera’s servers: they fetched the real page, compressed it, and sent a rendered version to a phone that could never have parsed it alone. It ran on the same handsets that ran WAP and produced recognisable websites on them. Opera Mini passed 50 million monthly users in February 2010 and 100 million in March 2011.

Mobile Safari arrived with the iPhone on 9 January 2007 and did no simplification at all. It rendered desktop pages on a 480 × 320 screen and let the user pinch and zoom. Jobs loading the full New York Times site on stage, then zooming into a single column of text, was the demonstration that ended the argument: a 412 MHz ARM processor with 128 MB of memory could run the real web.

Dead End

WAP’s designers were right that a 1997 handset could not render HTML, and they built a permanent replacement for a temporary condition. Within about five years the constraint was gone, and what remained was a stack of infrastructure, content practices and user expectations that had been built around it.

The economics mattered more than the markup. WAP put carriers at the centre: they ran the gateways, curated the portals, and billed the traffic, which reproduced the control they had over voice. i-mode was also carrier-centred, but it paid the people making the content, and that single difference produced a working ecosystem in Japan while Europe’s filled with ringtone vendors. The iPhone removed the carrier from the middle of the connection entirely.

The WAP stack did find one long-lived tenant. MMS, the carriers’ picture-messaging service, used WAP as its delivery mechanism from 2002 until the last German networks switched it off in 2026, a toll road running inside another toll road.

The companies that dominated the PC web (Google, Amazon, eBay, the New York Times) had no presence in the WAP ecosystem, and the companies that dominated WAP (ringtone vendors, SMS quiz services, portal operators) had none in the smartphone web. The mobile internet was not an evolution of WAP. It was the ordinary internet, arriving once the phones could hold it.