Dead End: BeOS

Jean-Louis Gassée and the Exit from Apple

Jean-Louis Gassée joined Apple France in 1981 and rose to become Apple’s Vice President of Product Development in 1985 — the role vacated when Steve Jobs was pushed out by John Sculley. Gassée was brilliant, opinionated, and French in the most theatrical sense: he wore turtlenecks before Jobs made them a uniform, collected mathematics, and ran Apple’s hardware division with genuine technical conviction.

The conflict with Sculley was strategic. Sculley wanted to expand the Macintosh market through aggressive price cuts. Gassée argued the opposite: maintain high margins, protect the premium brand. The Macintosh LC and IIci were Gassée’s machines — expandable, affordable by Apple standards, but not cheap. By 1990, Sculley had won the argument internally, and Gassée departed in February of that year.

Six months later he founded Be Inc. in Menlo Park, California. The founding team included Steve Sakoman, an Apple hardware engineer who had led the Newton project before it shipped, and a group of engineers who shared a conviction that the operating systems of 1990 — DOS, early Windows, even System 7 — were fundamentally inadequate for what multimedia computing would require. BeOS was not a patch on existing architecture. It was a clean-sheet design.

The Architecture of a Media OS

Gassée called BeOS “the Media OS.” The phrase was not marketing. It was an architectural thesis.

Every significant design decision in BeOS flowed from a single premise: that audio and video processing required deterministic, low-latency access to hardware, and that existing operating systems — built around batch processing or interactive text — were structurally incapable of providing it. Windows 95 could crash while playing audio because audio processing competed with the scheduler on equal terms. BeOS solved this at the kernel level.

Pervasive multithreading was the mechanism. BeOS ran every subsystem — the file system, the graphics layer, the network stack, even individual UI elements — in its own thread pool. An application that blocked waiting for a disk read did not stall the UI. A heavy computation in one window did not affect audio playback in another. The system could sustain eight simultaneous audio streams with less than a millisecond of latency on 1997 hardware. Developers who ported audio workstation software from other platforms reported that code they had written defensively — full of hacks to compensate for unpredictable scheduler behavior — simply ran clean on BeOS without modification.

Symmetric multiprocessing was built into BeOS from the first release. The BeBox (1995), Be’s custom hardware, ran two PowerPC 603 processors. SMP was not an afterthought or an enterprise add-on; it was the assumption the entire OS was designed around. Windows NT had SMP. Consumer Windows did not until Windows XP in 2001, six years after the BeBox shipped.

The Be File System (BFS), designed primarily by Dominic Giampaolo, was 64-bit and journaling in 1995 — when most consumer file systems were neither. Its defining feature was file attributes: every file in BFS could carry arbitrary typed metadata stored directly in the file system. An MP3 had title, artist, album, and genre stored as database attributes, queryable through the file system itself with SQL-like syntax. This was not a layer on top of the file system; it was the file system. The Tracker application browser was effectively a database query interface. Users could create virtual folders that aggregated all files matching an attribute query — all files with artist=“Miles Davis”, regardless of physical location — in 1998, seven years before Apple added Spotlight.

The API was C++ based and genuinely elegant. Developers who worked in it in the late 1990s still describe it with unusual affection. The message-passing system between application components was clean, the threading model was explicit and manageable, and the absence of thirty years of backward compatibility cruft made the entire development experience lighter than Windows or even UNIX.

The BeBox and the Port to Power Mac

The BeBox hardware launched commercially in 1995. It was a dual-PowerPC tower with an unusual feature that illustrated Be’s engineering culture: the GeekPort, a hardware I/O port that exposed direct access to analog inputs, digital I/O lines, and power — a built-in interface for hardware experimentation that no other consumer computer of the era offered. The BeBox was aimed at developers and power users. It was not cheap, it was not widely distributed, and it never achieved significant sales volume.

Be made a strategic decision in 1996 that proved both necessary and ultimately fatal: they would exit the hardware business and port BeOS to existing PowerPC hardware — specifically, to Apple’s Power Mac line. This was the only realistic path to a user base large enough to attract developers. The hardware margins were not there, and competing with Apple and the Wintel world on hardware economics was not a fight Be could win.

The Power Mac port (1996) was technically successful. BeOS ran well on the machines millions of Mac users already owned. Be offered BeOS as a dual-boot alternative to System 7 and the early System 8. The strategy was to establish BeOS as a specialized “media boot”: users would boot into BeOS to run audio production or video editing software, then return to the Mac for everything else.

The plan assumed those media applications would exist. They did not, at sufficient scale, because of the next event in BeOS’s history.

The Meeting Apple Chose Not to Take

By 1996, Apple was in genuine crisis. The Macintosh operating system, now called System 7, was aging. The company had spent hundreds of millions of dollars and five years on Copland, a next-generation OS designed to bring protected memory and preemptive multitasking to the Mac. Copland was cancelled in August 1996 — too late, too over-engineered, too fractured by committee decisions to ship. Apple needed an operating system to buy.

The two candidates were BeOS and NeXT.

Gassée knew this. He had been cultivating a relationship with Apple for months. BeOS was technically compelling — demonstrably more modern than anything Apple had — and it was designed for the same hardware Apple was already shipping. The case for BeOS was coherent: Apple’s existing users could run it on their existing machines, the media focus aligned with Apple’s creative professional customer base, and Be’s team knew the hardware intimately.

Negotiations in December 1996 broke down over price. Gassée reportedly wanted $300 million for Be Inc. Apple, under CEO Gil Amelio, was offering substantially less — some accounts put Apple’s opening position around $100 to $125 million. Gassée declined to negotiate to Apple’s range. He believed BeOS was worth more. He may have been right about the technology’s value in the abstract. He was wrong about what Apple would pay.

Meanwhile, Steve Jobs had been meeting separately with Apple’s board. NeXT’s price was $429 million — more money than Be was asking. But Jobs came with NeXT. The board voted to acquire NeXT not just for the operating system but for its founder, who had been building, at NeXT and Pixar, the skills and credibility he would need to run Apple again. Jobs returned to Apple as an “advisor” in January 1997. By July 1997, Amelio was out and Jobs was effectively running the company.

BeOS was not NeXT. NeXT came with a CEO.

Intel and the Last Years

Be ported BeOS to Intel x86 processors with Release 4 in November 1998. The port was technically clean and opened BeOS to the enormous installed base of Windows PCs. Be’s strategy shifted again: BeOS would become a dual-boot media operating system for Windows users, available in shrink-wrap at retail.

The market response was poor. Retail sales of alternative operating systems in 1998 required crossing a psychological threshold that most Windows users were unwilling to cross. Developers faced a familiar dilemma: port your application to an OS with a small user base, or focus on Windows and Mac where the customers were. The chicken-and-egg problem that had constrained BeOS on PowerPC followed it to x86.

The crucial OEM deal that might have broken this cycle collapsed under pressure that was never definitively proven but was widely suspected. In 1999, Compaq had agreed to preinstall BeOS on certain machines — a dual-boot configuration that would have put BeOS in front of mainstream consumers at point of sale. The deal was cancelled. Be Inc. filed a complaint with the European Commission in 2000, alleging that Microsoft had pressured OEMs not to preinstall competing operating systems. The complaint named the Compaq deal specifically. The EC action was absorbed into the broader Microsoft antitrust proceedings but never produced a specific remedy for Be.

BeOS Release 5 shipped in March 2000. Be also released BeOS Personal Edition — a free version that could be downloaded and run from within Windows, without repartitioning, by users curious enough to try it. The Personal Edition represented a genuine attempt to lower the adoption barrier, and it worked in the limited sense that downloads were significant. They did not translate into an application ecosystem. Without major software vendors — Adobe, Macromedia, Microsoft Office — BeOS remained a demonstration of technical possibility rather than a working computing environment for most users.

Shutdown and Aftermath

Be Inc. ceased operations in August 2001. Palm Inc. acquired Be’s intellectual property for $11 million — a price that measured precisely how far the company’s fortunes had fallen from Gassée’s $300 million ask five years earlier. Palm wanted BeOS technology for a new generation of mobile and handheld devices, believing the OS’s low-latency, media-capable architecture was well-suited for what would become the smartphone era. The products never shipped. Palm’s own trajectory was erratic, and the BeOS technology was not integrated into any Palm product before Palm itself was acquired by Hewlett-Packard in 2010.

The engineers scattered. Dominic Giampaolo went to Apple. Several former Be employees joined Danger Inc., the startup that created the Sidekick smartphone — a device whose innovative UI and cloud-synchronization model influenced the smartphone design language before Danger was acquired by Microsoft in 2008. Others joined companies across the industry. The concentration of multimedia and threading expertise that Be had assembled dispersed but did not vanish.

Haiku: The Continuation

BeOS did not simply end. In 2001, a community of developers began an open-source reimplementation named OpenBeOS, later renamed Haiku — after Be’s internal project codenames, which used Japanese words. The goal was binary compatibility with BeOS R5 applications on modern hardware.

Haiku achieved its first self-hosting build in 2002 and released Haiku R1/Alpha 1 in 2009. As of 2026, Haiku remains in active development and has achieved Release 1 (2019), with ongoing releases maintaining compatibility with BeOS applications while updating the hardware support layer and porting modern software. It runs on both 32-bit and 64-bit x86 hardware, supports Wi-Fi, modern graphics, and USB. The BFS file system, the Tracker file browser, the multithreaded architecture, and the C++ API all survive intact.

Haiku is a niche operating system used by enthusiasts and developers interested in clean OS design. It is not a commercial competitor to any modern platform. But it is the most faithful preservation of an OS architecture that its creators did not get to finish, maintained by a community that believes the original design decisions were correct.

Why It Failed

BeOS did not fail because it was bad. This matters because most computing dead ends fail for technical reasons. BeOS failed for structural ones.

The Apple negotiation was the single most consequential moment. A deal at $150 million would have kept Be intact and given BeOS the most plausible path to adoption it was ever going to get. Gassée’s unwillingness to meet Apple’s range cost Be its best customer, its best distribution channel, and its best argument to the OEM and developer ecosystem that BeOS had a serious future.

The application poverty was a consequence, not a cause. Developers did not port to BeOS because the user base was small. The user base was small because there was no hardware deal. The hardware deals did not close because the OEM ecosystem was, allegedly and demonstrably, unfavorable to alternatives. This is the standard platform death spiral, and BeOS ran through it without any of the structural advantages — no installed base, no existing developer loyalty, no deep-pocketed corporate parent — that might have broken the cycle.

Gassée himself understood the design better than the business. He was an excellent product thinker and a poor negotiator when his leverage was weakest. The $300 million ask for a company with no consumer distribution, no major application, and a hardware business it had already exited was not a negotiating position grounded in the deal Apple needed. He was not wrong that BeOS was worth the price. He was wrong that Apple could not get what it needed for less — and he was wrong that NeXT would not suffice.

The lesson BeOS teaches is the same lesson OS/2 teaches, and the same one the Amiga teaches: in platform markets, being technically correct is a necessary condition but not a sufficient one. Distribution, developer relationships, and timing determine survival. BeOS had the best technical ideas and the worst timing. It arrived at exactly the moment when Microsoft’s grip on OEM distribution was most unbreakable and Apple was most desperate — and desperate buyers choose the option that comes with a CEO, not the one that comes with better threading.

📚 Sources

• Jean-Louis Gassée at Apple — Folklore.org (Andy Hertzfeld)
• BeOS Technical Overview — Be Inc. Developer Documentation (archived)
• Dominic Giampaolo: Practical File System Design with the Be File System (Morgan Kaufmann, 1999)
• Be Inc. vs. Microsoft — European Commission Complaint (2000)
• Haiku Project — Official Site
• Apple Acquires NeXT — Press Release (1996)
• Be Inc. — Wikipedia
• BeBox — Wikipedia