Skip to content

Dead End: Apple Newton

Abstract

Apple shipped the Newton MessagePad on 2 August 1993 for $699. John Sculley had coined the term “personal digital assistant” for it at the Consumer Electronics Show in January 1992, fourteen months before there was a product to sell. Three weeks after launch, Garry Trudeau spent a week of Doonesbury on the handwriting recognition, and the panel in which “Catching on?” came back as “Egg freckles” outlived the device. Steve Jobs cancelled the line on 27 February 1998. Roughly 200,000 Newtons sold in four and a half years. The low-power processor Apple bought into to make the Newton possible was ARM, and every iPhone since 2007 has run on its descendants.

Origins

Steve Sakoman joined Apple in 1984 from Hewlett-Packard, where he had built the HP-110, the first battery-powered MS-DOS portable. At Apple he became director of hardware engineering and worked on the system design of the Macintosh Plus, SE, II, and the Apple IIGS. In 1987, with backing from Jean-Louis Gassée, he started a project named Newton to build a handheld computer: pocket-sized, pen-driven, wireless, with software that understood what the user wrote rather than merely storing it.

The processor was the first problem. Newton was designed around AT&T’s Hobbit, a CPU built for C code that turned out to be too slow and too power-hungry for a battery device. Apple went looking for an alternative and found one at Acorn in Cambridge. In November 1990, Apple put roughly $3 million into a joint venture with Acorn and VLSI Technology called Advanced RISC Machines Ltd, taking a 43% stake; Acorn supplied twelve engineers, VLSI the tools. Apple’s handheld is the reason ARM exists as an independent company.

Sakoman left Apple in 1990 with Gassée to found Be Inc. Larry Tesler, who ran Apple’s Advanced Technology Group, took over the project. The decisive product call came from a marketing manager, Michael Tchao, who convinced Sculley to build the small, roughly $1,000 “Junior” rather than the tablet-sized, expensive “Senior” the team had been designing.

John Sculley then made the promise. At the Consumer Electronics Show in Las Vegas on 7 January 1992 he used the phrase “personal digital assistant” in public for the first time and said Apple would ship one, for under $1,000, in 1993. A prototype was demonstrated publicly on 29 May 1992. Sculley later described the pressure behind the announcement plainly: “I was getting intense pressure from the Apple board… And that looked like they could ship a product.” The team then had fourteen months to make the demo true.

The MessagePad

The first MessagePad (OMP/H1000) went on sale on 2 August 1993 at Macworld Expo in Boston for $699. It measured 7.25 × 4.5 × 0.75 inches, ran a 20 MHz ARM 610, and carried 640 KB of RAM (490 KB system, 150 KB user) plus 4 MB of ROM on a 336 × 240 monochrome touchscreen. Four AAA batteries powered it. There was no keyboard: input was a stylus on glass.

Newton OS was the more original half of the machine. It had no file system and no save command. Everything the user created (notes, contacts, appointments, sketches) lived in a persistent object store, and applications wrote into shared databases called soups rather than into private files. Data moved between applications because it was never separated in the first place. Write “Call John at 3pm Tuesday” and the system would pick out the time, the day, and the name, and offer to make an appointment. That kind of semantic interpretation did not reappear in mainstream handhelds for over a decade.

Apple reported strong early demand: several thousand units sold out on the Macworld show floor, and about 50,000 shipped by the end of November 1993. Sales never accelerated from there.

The Handwriting Recognition

Recognition was the feature Sculley had sold, and version 1.0 was not ready.

The recognizer was CalliGrapher, licensed from ParaGraph International, a Moscow company founded by Stepan Pachikov. It attacked cursive and mixed writing whole-word at a time, matched against a dictionary, and adapted to a user’s hand over time. In demos, with trained writers and known vocabulary, it worked. In a shop, with an untrained buyer writing a name that was not in the dictionary, it produced confident nonsense.

Late in August 1993, three weeks after launch, Garry Trudeau ran a week of Doonesbury strips in which Michael Doonesbury’s new Newton reduces everything he writes to gibberish. In the strip of 27 August, “Catching on?” comes back as “Egg freckles.” The phrase became the standing joke for expensive technology that does not work, and it stuck to Apple’s product for the rest of its life.

The team fixed the problem, two years too late to matter. Newton OS 2.0, released 14 March 1996, added Rosetta, a print recognizer written at Apple by Larry Yaeger and colleagues: a neural network classifying characters, wrapped in a context-driven search over segmentation and word hypotheses, with dictionaries and letter-frequency models breaking ties. It shipped alongside CalliGrapher, which kept the cursive job. Reviewers who had mocked the 1993 machine called the result the first genuinely usable handwriting recognition in a shipping product. Yaeger’s team published the design in AI Magazine in 1998, by which time the product was dead. The same recognizer came back to Apple in 2002 as Inkwell in Mac OS X 10.2.

Egg freckles, licensed back

Trudeau had never used a Newton when he wrote the strips. Steve Capps, chief architect of Newton OS 2.0, licensed a Trudeau panel from him for the machine that fixed the problem: write “egg freckles” on a Newton running OS 2.0 and tap Assist, and the cartoon appears. The Easter egg was removed again in OS 2.1.

What Newton Got Right

A third-party software market. Newton software shipped on PCMCIA cards and later over serial links, and an independent developer scene grew around medical references, databases, scientific calculators, and games. Apple ran a handheld application market twelve years before the App Store.

Connectivity as a premise. With a modem card the Newton sent faxes and email; infrared beamed contacts between two devices across a table; a serial cable and the Newton Connection Kit moved data to a desktop. In 1993 the assumption that a handheld existed to talk to other machines was not general. It was also the weakest part of the promise: cellular data barely existed, so most Newtons spent their lives unconnected. Palm made desktop synchronization work properly in 1996 with the HotSync cradle, which was Palm’s own design, not Apple’s.

Stylus interaction conventions. Tap to select, drag to scroll, scrub a word out to delete it. These were invented for the device rather than inherited from the desktop, and most touch platforms since have kept some version of them.

Power-first engineering. Designing for AAA cells forced aggressive sleep states, wake-on-input, and deferred computation, the discipline that ARM-based mobile design later ran on.

Newton Inc. and the Cancellation

By 1997 the Newton had lost the market it named. Gil Amelio’s answer was separation: on 22 May 1997 Apple announced that the Newton group would become a wholly owned subsidiary, and Newton Inc. was formalized on 1 July with about 130 employees, an office in Santa Clara, Sandy Benett as chief operating officer, and a plan to raise outside money and go public within two years.

It never moved in. Jobs returned as interim CEO in July 1997 and in September told the Newton Inc. executives not to bother with the new offices. The group was folded back into Apple, and on 27 February 1998 Apple ended the platform: MessagePad 2100, eMate 300, Newton OS, all of it.

Jobs’s stated objection was the stylus. A pen has to be stored, retrieved, and held, which makes it an accessory the device depends on; a finger is always present. He was right, though the finger-driven machine he had in mind took another nine years to build. There was also the organizational argument: Newton had been Sculley’s project, ran as its own division, and had just been spun into a company of its own. Jobs was closing autonomous fiefdoms as fast as he found them.

What Carried Over

ARM. The investment Apple made to get a chip for the Newton produced the architecture behind the iPod, the iPhone, and eventually Apple Silicon. Of everything the Newton caused, this is the piece with the largest downstream effect.

Rosetta. Yaeger’s recognizer became Inkwell in Mac OS X, and handwriting recognition returned as a system service on the iPad in 2020 under a different implementation.

People. Steve Capps, one of the original Macintosh designers and chief architect of Newton OS 2.0, was made an Apple Fellow in 1994 and left for Microsoft in June 1996, where he designed the Search, History, and Favorites panes of Internet Explorer. Jon Rubinstein arrived from NeXT in 1997 to run hardware engineering, led the iPod, and became chief executive of Palm in 2009, where he shipped webOS into the market the Newton had failed to hold.

Multi-touch. Apple bought FingerWorks in 2005 for the gesture technology in the iPhone screen, solving with fingers the input problem the Newton had handed to a stylus.

Dead End: Announced Too Early, Shipped Too Early

The convenient explanation is that 1993 hardware could not do what the Newton attempted, but the Newton’s own history contradicts it. Rosetta ran a neural network on the same class of ARM chip in 1996 and worked. The recognition problem was solvable within the product’s own lifetime, and Apple solved it. What was not survivable was the order of events: Sculley named the category in January 1992, promised a price and a year, and the team shipped in August 1993 with the flagship feature unfinished. The reviews, and then the comic strip, were written against version 1.0 and never revisited.

The rest was market position. At $699 and up, the Newton competed as a computer while doing the work of an organizer. In March 1996 Palm shipped the Pilot 1000 for $299: no fax, no soup architecture, a simplified alphabet the user had to learn instead of a recognizer that had to learn the user. Palm sold a million units in eighteen months. Apple sold roughly 200,000 Newtons in total.

When the iPhone arrived in 2007 the missing pieces had arrived with it:

  • A capacitive multi-touch screen, replacing the resistive panel and the stylus
  • An ARM11 processor rated at 620 MHz, run at 412 MHz, with a UNIX operating system behind it
  • Wi-Fi and EDGE cellular data, so the device was connected by default
  • A battery good for a day of real use
  • A software distribution channel with a payment system attached (the App Store, 2008)

The Newton was right about what a handheld should do and wrong about how long that would take. Its epitaph is a phrase a cartoonist made up about a machine he had never used.

📚 Sources