Skip to content

Dead End: Sun Microsystems

Abstract

Sun Microsystems built Java, NFS, ZFS, and the workstation that a generation of scientific computing ran on, and said “the network is the computer” two decades before the cloud arrived to prove it. Its market capitalization peaked around $200 billion in 2000. Oracle bought the company in January 2010 for $7.4 billion. Sun was right about where computing was going and wrong about who would pay for it: every prediction it made came true on commodity x86 hardware sold by somebody else.

Stanford University Network

Sun Microsystems was founded in 1982 by four people: Andy Bechtolsheim, a Stanford electrical engineering PhD student who had designed the SUN workstation (Stanford University Network) as a personal project; Bill Joy, a Berkeley graduate student who had written much of BSD Unix and the vi text editor; Vinod Khosla, the original CEO; and Scott McNealy, who became the long-term CEO and the company’s public face. The founding team’s combination of hardware design (Bechtolsheim), software depth (Joy), and business drive (McNealy/Khosla) was formidable.

The Sun-1 (1982) was a Unix workstation: a complete computer with a high-resolution display, running BSD Unix, designed for engineers and scientists. It competed with Apollo Computer, the Xerox workstations, and eventually Digital Equipment’s VAX. Sun’s advantage was price (the Sun-1 used commodity Motorola 68000 processors rather than custom chips) and software: BSD Unix was technically superior to proprietary minicomputer operating systems for networking and multi-user workloads.

By the mid-1980s, Sun dominated the technical workstation market. Universities, research labs, financial firms, and engineering companies ran their computation on Sun machines. The company motto, “the network is the computer,” was coined by John Gage and described the cloud computing era about two decades before it arrived.

SPARC and Solaris

In 1987, Sun introduced the SPARC architecture, a RISC processor designed in-house that Sun licensed to other manufacturers (Fujitsu, Texas Instruments, Cypress Semiconductor). SPARC was technically excellent: clean RISC design, good floating-point performance, and scalable from workstations to servers. SunOS, the BSD-derived operating system, was replaced with Solaris (1992), a System V Unix derivative that added advanced features including symmetric multiprocessing, the ZFS file system (introduced in 2005), and DTrace (dynamic tracing for performance analysis).

Sun’s hardware margins through the late 1980s and 1990s were wide. A Sun workstation sold for $10,000 to $30,000; the components cost a fraction of that. Sun’s servers (SPARCstation, SPARCserver, and the enterprise-class E-series) carried similar margins. Revenue reached $11.7 billion in fiscal 1999.

The competitive risk was visible. Intel x86 processors were improving at the rate Moore’s Law predicted. By the mid-1990s, Linux running on x86 commodity hardware could perform roughly comparable scientific computing work to Solaris on SPARC at one-tenth the price. Sun’s engineers knew this; Sun’s business model depended on it not mattering to customers before Sun could transition to a new model.

Java

James Gosling led the development of Java at Sun, initially for embedded systems (the Oak language, designed for interactive television set-top boxes, a project that failed to find a market). When Netscape integrated a Java runtime into its browser in 1995 and Sun launched Java as a platform for “write once, run anywhere” cross-platform applications, adoption was immediate. Java applets were everywhere on the late 1990s web. Java became the standard language for server-side web development through J2EE (Java 2 Platform, Enterprise Edition). The JVM became one of the most important pieces of infrastructure in computing.

Sun gave Java away. The JVM was free; the tools were free; the standard libraries were free. Sun’s plan was to benefit from Java’s proliferation by selling the servers Java applications ran on. This worked during the dot-com boom, when Sun sold enormous numbers of E-class servers to internet companies. It stopped working when the boom collapsed.

The Microsoft Legal Battle

Sun licensed Java to Microsoft in 1996 with the requirement that Microsoft implement the full standard. Microsoft extended Java with Windows-specific APIs and produced a JVM that was incompatible with Sun’s standard, allowing Java programs to run on Windows but not portably across platforms. Sun sued Microsoft for breach of contract in 1997; that Java contract dispute settled in 2001. Broader antitrust and patent litigation between the two companies was resolved in an April 2004 settlement in which Microsoft paid Sun roughly $1.6 billion, a resolution in Sun’s favor. The legal dispute consumed years of management attention and produced deep mutual hostility between the companies that persisted long after the case concluded.

NFS

Sun’s most durable technical contribution to computing was neither Java nor SPARC. It was NFS (Network File System), developed at Sun in 1984 and described in the open the following June, when Russel Sandberg, David Goldberg, Steve Kleiman, Dan Walsh, and Bob Lyon presented “Design and Implementation of the Sun Network Filesystem” at the USENIX summer conference in Portland. The protocol was later written up as an IETF specification, RFC 1094 (March 1989), covering NFS version 2. NFS allowed a Unix workstation to mount remote file systems over a network as if they were local disks. The design was simple: stateless server, RPC-based protocol, open specification that any vendor could implement.

Sun made a strategic decision to publish the NFS specification openly rather than keeping it proprietary. The reasoning was that NFS’s value to Sun came from selling the servers that ran NFS, not from licensing the protocol. If every Unix system could mount Sun NFS servers, every organization running Unix was a potential Sun customer for file servers. The strategy worked: NFS became the de facto standard for Unix networked file systems and was implemented by every major Unix vendor, by Microsoft for Windows Services for Unix, and later by Linux. NFS version 4 (2000) and NFSv4.1 with parallel NFS (2010) are still in production use in data centers worldwide: NFS implementations are running today on systems Sun never manufactured.

OpenSolaris and its descendant illumos (fork from 2010, after Oracle’s acquisition) kept Solaris’s technical innovations available after Oracle effectively killed the commercial version. ZFS, the Zettabyte File System designed by Jeff Bonwick and Matthew Ahrens and first shipped in Solaris 10 (2005), brought self-healing checksumming, atomic copy-on-write snapshots, and integrated volume management that no other file system offered at the time. ZFS lives today in OpenZFS, running on FreeBSD, Linux, and macOS, technically superior to many alternatives, another case where Sun’s engineering outlasted Sun’s business.

The Unix Workstation Wars

Through the 1980s and 1990s, Sun competed in a market of technically capable Unix workstation vendors, each with a proprietary processor and operating system:

  • Silicon Graphics (SGI): MIPS R-series processors running IRIX, dominant in 3D visualization, film effects, and engineering simulation. SGI workstations ran the rendering for Jurassic Park (1993) and Terminator 2 (1991). SGI’s workstations cost $15,000–$100,000 and delivered graphics performance Sun could not match. SGI filed for bankruptcy in 2006.
  • DEC (Digital Equipment Corporation): Alpha processors running OSF/1 (later Tru64 Unix), the fastest CPUs available in the mid-1990s. DEC Alpha at 400 MHz in 1994 was significantly faster than any SPARC or x86. DEC was acquired by Compaq in 1998; Compaq was acquired by HP in 2002; Alpha was discontinued in 2004.
  • HP: PA-RISC processors running HP-UX, strong in commercial computing and technical workstations. HP-UX’s enterprise features and Oracle database support made it the choice for large commercial installations.
  • IBM: POWER processors running AIX, dominant in enterprise and database server markets. IBM’s ability to bundle hardware, software, and services gave it a competitive durability that Sun lacked.

Sun’s positioning in this market was price-performance: not the fastest (DEC Alpha), not the best for 3D (SGI), not the most enterprise-proven (IBM), but the best combination of performance, networking capability, and cost. The Sun-4 and UltraSPARC workstations were the default choice for university computer science departments, research labs, and internet infrastructure companies through the 1990s. This installed base of developer mindshare was a significant asset, until it became irrelevant when Linux on x86 could do the same work for less.

Scott McNealy and the Culture of Combat

Scott McNealy was Sun’s CEO from 1984 to 2006, and his personality defined the company’s public posture. McNealy was combative, quotable, and genuinely convinced that networked computing was the future of the industry. He was also willing to turn every competitive situation into a war of rhetoric.

Gage’s slogan was a marketing line that happened to describe a real technical position. Sun was building infrastructure for a world in which computation was a service reached over a network rather than a resource sitting under a desk. That is cloud computing, described in the mid-1980s by a company whose revenue depended on selling the desk-side machine.

McNealy’s statements about Microsoft were quoted across the industry: “Microsoft is like a huge, malicious, spiteful machine. You can’t stop them. You can only slow them down.” The Java dispute consumed years of management attention and produced a hostility that outlasted the litigation.

Sun opened Solaris under McNealy: the bulk of the source went out on 14 June 2005 under the CDDL, and the OpenSolaris distribution followed in May 2008. Jonathan Schwartz, CEO from 2006, pushed the strategy further, keeping OpenOffice (acquired with Star Division in 1999 for $73.5 million) and positioning Sun as the open-source hardware company. The direction was right and the revenue model was missing: open source did not generate income at the scale Sun’s cost base required.

The Dot-com Crash and the Collapse

Sun’s revenue peaked at $18.2 billion in fiscal year 2001. By fiscal 2003 it was $11.4 billion. By 2006 it was $13.1 billion, a recovery that required cutting 15,000 jobs and never fully materialized. The dot-com companies that had bought Sun servers went bankrupt. The enterprise customers who remained were buying Linux x86 servers from Dell and HP for data center expansion, not Solaris SPARC machines.

IBM opened acquisition talks in March 2009 and put roughly $9.40 per share on the table, about $6.85 billion. Sun balked at the price, cancelled IBM’s exclusive negotiating rights, and IBM withdrew. Sun agreed to Oracle’s offer of $9.50 per share, $7.4 billion gross and $5.6 billion net of Sun’s cash and debt, announced on 20 April 2009. The deal closed on 27 January 2010. Sun held out against IBM and got ten cents a share more.

Oracle’s Acquisition and What Remained

Oracle acquired Sun primarily for Java and Solaris. The first major action after closing was a lawsuit against Google for its use of Java APIs in the Android platform, an action that generated over a decade of litigation (Oracle v. Google, ultimately resolved by the Supreme Court in Google’s favor in 2021). Oracle continued developing Solaris and SPARC through 2017, then effectively killed both: SPARC processor development ended, Solaris development was dramatically reduced, and most of the Solaris engineering team was laid off.

Java survived as a thriving ecosystem outside Oracle’s control. OpenJDK (the open-source Java runtime that Sun had released before the acquisition) became the primary Java implementation for most users. The Jakarta EE community (enterprise Java) moved its standards work to the Eclipse Foundation after Oracle failed to transfer the javax.* namespace. Java development continues, managed by an open-source community that works despite Oracle’s legal ownership of the platform rather than because of it.

ZFS was open-sourced and lives on as OpenZFS, shipping on FreeBSD and Linux and available for macOS as a third-party port. DTrace shipped in Solaris, FreeBSD, and macOS, and was ported to Linux, where eBPF has since taken most of the same work. OpenOffice was forked into LibreOffice after Oracle’s acquisition and indifferent stewardship; LibreOffice is now the dominant open-source office suite.

Dead End

Sun’s fundamental strategic error was not building Java or betting on networked computing; those were correct. It was failing to transition away from the proprietary hardware model before commoditization made that model impossible. When Linux on x86 could do 80% of what Solaris on SPARC could do at 20% of the cost, Sun’s only sustainable response was to become a software company. McNealy resisted this framing; the culture of the company was hardware. Jonathan Schwartz, who became CEO in 2006 after McNealy, moved aggressively toward open source (open-sourcing Solaris as OpenSolaris, supporting OpenOffice, embracing Linux) but did so without a clear model for generating revenue from open-source software assets. The transition was correct in direction and incomplete in execution, and the dot-com crash had eliminated the time buffer Sun needed to complete it.

📚 Sources