Intel, Microsoft, and Linux reshape server market: key takeaways

•

 42 min video

•

 13 min read

YouTube video ID: ksx5g5WvAIw

Source: YouTube video by Asianometry — Watch original video

PDF

In the 1990s, enterprise servers predominantly ran UNIX software on RISC hardware. However, the landscape began to shift dramatically with the emergence of Intel's x86 architecture and two powerful operating systems: Windows NT and Linux. This period marked a fascinating industrial transition where Intel, Microsoft, and Linux collectively challenged and ultimately reshaped the dominance of UNIX in the server market.

Beginnings: Intel's Ambition

Intel, having successfully conquered the consumer PC market through high-volume sales and superior manufacturing yields, sought to replicate this success in the enterprise server space. Enterprise servers, used by corporations to manage and serve client computers, ranged from powerful workstations for intensive tasks to large monolithic systems with multiple CPUs, 64-bit support, virtual machines, and extensive memory for databases, file storage, and complex business applications. The burgeoning internet craze further expanded the market for web servers, though not all required high-end machines.

Vertically Integrated UNIX Dominance

At the decade's outset, the enterprise server market was controlled by vertically integrated vendors offering closed, proprietary systems. These systems typically featured RISC CPUs running a variant of the UNIX operating system. Prominent examples included:

  • Sun Microsystems: SPARC chips and Solaris OS
  • IBM: POWER RISC chips and AIX OS
  • Hewlett-Packard: PA-RISC chips and HP-UX OS
  • Silicon Graphics

These systems were prohibitively expensive, often costing hundreds of thousands of dollars, and their proprietary nature created significant vendor lock-in, making switching nearly impossible.

Windows NT: Microsoft's Entry

To penetrate the server market, Intel needed a strong software partner, which led them to Microsoft and Windows NT. Released in 1993 and led by Dave Cutler, NT was a proprietary OS designed from the ground up for both workstations and servers, distinct from the consumer-oriented Windows. Unlike the consumer Windows, which was a graphical shell over MS-DOS, NT was built for enterprise reliability, power, and enhanced security.

The initial version of NT Server, running on the Intel 386, was immature and lacked features present in competitors like IBM's OS/2, Novell's NetWare, and UNIX variants. It also demanded significant hardware for its time, requiring 16 megabytes of RAM. With only 30,000 units sold in 1993, its launch was a disappointment. However, Microsoft persisted, adding features and steadily growing its market share. By 1994, sales reached 115,000 licenses, demonstrating a clear upward trend.

Pentium Pro: Intel's Performance Wedge

In late 1995, Intel introduced the 32-bit Pentium Pro, their first chip truly designed for the server space. With 5.5 million transistors and speeds of 150-200 MHz, it significantly surpassed the original Pentium. The Pentium Pro surprisingly outperformed the fastest RISC chips in certain integer benchmarks, challenging the perception of x86 as a "desktop" architecture. This performance advantage provided Intel with the necessary leverage to enter the server market.

Combined with Windows NT 4.0, released earlier that year, the Pentium Pro offered a ready-made server standard that could match RISC-based systems' performance at a much lower price. Intel circulated "Standard High Volume Servers" reference models, supporting 1-4 Pentium Pros, 1-2 GB of RAM, and hundreds of GBs of storage, capable of handling up to 2,000 transactions per second. By 1996, Windows NT 4.0 sales grew by 85%, achieving 27% market share, solidifying the "Wintel" presence in servers.

Border Skirmishes and Mismatches

Despite their cooperation, tensions existed between Intel and Microsoft. The "Native Signal Processing" (NSP) affair in 1995, where Intel attempted to offload multimedia compute to the CPU via software, was seen by Microsoft as an encroachment on Windows' turf. Microsoft publicly refused to support NSP, reminding Intel of its leverage.

Furthermore, Windows NT was designed to be portable, running on various architectures like DEC Alpha, PowerPC, and MIPS, not just x86. This raised concerns for Intel that if NT dominated the market, x86 could become just one of many supported hardware platforms.

Another issue was the mismatch in development cycles. Intel released new, faster chips every two years, while Microsoft's Windows releases occurred every three to five years and didn't always fully utilize Intel's latest capabilities. A notable example was the decade-long gap between Intel's 32-bit 386 processor (1985) and Microsoft's 32-bit consumer OS (Windows 95, 1995). During Pentium Pro development, Microsoft's last-minute decision to keep Windows 95's graphics display interface 16-bit, despite earlier assurances of a fully 32-bit OS, led to the Pentium Pro being dinged for subpar graphics performance on consumer desktops.

In the server world, NT 4.0's limitation of 4-8 processors per OS instance imposed a performance ceiling, keeping it out of the high-end market where some UNIX variants supported up to 64 processors. This tension stemmed from Intel's focus on selling the newest chips and Microsoft's prioritization of servicing its entire install base with backward compatibility.

The Rise of Linux

In the late 1990s, a new alternative emerged: Linux.

Stallman and GNU

In January 1984, Richard Stallman, motivated by the fragmentation of the MIT AI Lab and a desire to rebuild a hacker community based on free software, initiated the GNU project. The goal was to create a free operating system compatible with UNIX's design principles (modularity, portability) but without copying its proprietary code. Over the years, the GNU community built various tools, but a crucial component, the kernel, remained elusive.

Linus and Linux Kernel

The missing kernel arrived thanks to Linus Torvalds, a 21-year-old Finnish undergraduate inspired by MINIX. In August 1991, Torvalds announced his hobbyist OS project on Usenet, attracting contributors and eventually leading to the creation of Linux. Torvalds transitioned from coder to project leader, reviewing and merging contributions in a "bazaar-like" approach that enabled rapid code development.

The first workable version of Linux was released in October 1991. In 1992, version 0.12 adopted Stallman's GNU General Public License (GPL). The GPL allowed anyone to read and modify Linux's code, provided that any distributed modifications were also open-sourced. This prevented fragmentation, ensured that company contributions benefited everyone, and was later cited by Torvalds as his best decision.

Distros and Red Hat

Linux, being just a kernel, required "distributions" (distros) to be user-friendly. Distros bundled the Linux kernel with utilities from the GNU Project and other free software. Early notable distros included Slackware and Debian.

Red Hat, founded by Marc Ewing in 1993, emerged as a key commercial distribution. Ewing, frustrated by the time spent managing his Linux box, created a distribution that professionalized Linux. Red Hat provided tools like the Red Hat Package Manager (RPM) for software management and sold CDs, training, education, and customer support services. Red Hat and other commercial distributions like Caldera helped transition Linux from a hacker's tool to a corporate mainstream option.

An OS With Opportunity

Linux adoption grew rapidly. From a few hundred users in early 1992, it reached 1-2 million by 1996 and 3-5 million by mid-1997. While still lagging Windows and Mac in user numbers, its growth rate was explosive.

Price was a significant factor; a Red Hat disc cost around $50, compared to Windows NT licenses ranging from $150 to $1100, and UNIX systems costing $25,000 to $250,000 or more. Beyond cost, Linux offered reliability, with some users describing it as "crash-proof," leading to increased productivity. Sophisticated users valued access to the source code, allowing for customization and innovation (e.g., running Linux on a Palm Pilot, or networking 68 PCs for scientific simulations at Los Alamos).

However, Linux had shortcomings: a limited application ecosystem compared to Windows or Mac (no good Linux versions of Microsoft Office or Visual Basic), and challenges with support and documentation, though companies like Red Hat addressed this. Its association with UNIX also gave it a reputation for being inscrutable. Despite its benefits, a negative bias persisted among executives, who questioned its suitability for critical servers. Yet, many system administrators secretly deployed Linux, prioritizing what "works best."

Intel Meets Linux

In early 1998, Intel began receiving requests from institutions like Fermi National Lab to port its math libraries to Linux. While Intel's prior involvement with Linux was minimal, checks with OEMs revealed an explosion in Linux growth, particularly among Internet Service Providers, with nearly 20% of public web servers running it.

This news reached Andy Grove, prompting Intel to investigate. Intel executives were initially baffled by Red Hat's business model—selling a product whose core IP was freely available. However, Red Hat CEO Robert Young's "ketchup metaphor" (Heinz dominating the ketchup market despite anyone being able to make it) and data on ISP adoption convinced Intel management. Recognizing the risk of relying solely on Windows NT, especially with delays in NT 5.0 (later Windows 2000) and concerns about NT shackling the power of their upcoming 64-bit Itanium chip ("Merced"), Intel saw Linux as a compelling alternative. They could support and contribute to Linux development, optimizing it for their hardware.

Summer of Linux

The summer of 1998 saw major companies embracing Linux:

  • IBM: In June 1998, IBM announced it would include Apache, an open-source web server software (mostly run on Linux), in its e-commerce tools. This marked open source's first significant entry into the enterprise.
  • Oracle: After initially stating no plans to port its database software to Linux, Oracle abruptly reversed course in July 1998, announcing the porting of Oracle8 to Linux. This turnaround was heavily influenced by a "Future of Linux" panel discussion in Santa Clara, which highlighted user demand and the growing momentum of open source. By September 1998, other major companies like Computer Associates, Corel, and Informix also had Linux ports.

Intel and Red Hat Partnership

Intel was present at the "Future of Linux" panel, with Sunil Saxena emphasizing Intel's desire to support Linux and address its weaknesses (scalability, high-end hardware drivers, large memory support). On September 29th, Intel, along with Netscape and VC firms Benchmark and Greylock, announced an investment in Red Hat. Intel committed to technical contributions, a dedicated engineering liaison for the Linux community, and joining Linux International.

This investment had a ripple effect. A Benchmark partner, also a Dell board member, introduced Red Hat to Michael Dell. In early 1999, Dell became the first major vendor to preinstall Linux on workstations and servers, elevating Linux to the same status as Windows and Novell.

Microsoft's Reaction

Intel's investment in Red Hat sent Microsoft into a frenzy. Microsoft executives reportedly "went ballistic," making angry calls to Intel. However, due to an ongoing antitrust lawsuit filed by the US government in May 1998, Microsoft had to temper its behavior. Publicly, they cited Red Hat as evidence against their monopoly claims. Internally, however, Microsoft was deeply concerned. The leaked "Halloween Memos" in October 1998 identified free software like Linux as a "long-term credible threat."

Microsoft reportedly formed a "Linux attack team" to discredit Linux, launching a "Linux Myths" webpage that challenged claims of Linux's performance, reliability, and freeness. In April 1999, Microsoft sponsored Mindcraft benchmark tests that controversially showed Windows NT outperforming Linux as a file and web server.

Port of Choice: Intel's Neutrality

Intel navigated Microsoft's anger by declaring itself "OS neutral," aiming for all operating systems—Linux, NetWare, Solaris, Windows 2000—to run well on its hardware. This "Port of Choice" strategy, formalized later but long-standing, was accepted by OS makers due to Intel's lack of real alternatives in silicon.

Intel continued to innovate its server chips. The Pentium II Xeon (mid-1998) targeted mid-range workstations and servers with faster clock speeds and larger L2 cache. The Pentium III Xeon (March 1999) offered support for 8-way systems via Profusion, breaking the previous 4-way ceiling and addressing performance issues in data centers. While high-end UNIX servers still held advantages in memory support for large databases and CAD systems, these Pentium chips opened the low to middle ground of UNIX servers to massive competition.

The Web Scales Out

Sun Microsystems initially welcomed Linux, porting its Java Development Kit and workstation hardware to it, believing that as long as it ran on SPARC, it was fine. However, this proved problematic. For web serving, email hosting, and firewalls, companies found that scaling out with nodes of Intel silicon running Linux was more financially sensible than scaling up expensive UNIX servers.

This trend accelerated after the Dotcom bubble burst, leading to increased frugality. Amazon, for instance, made a major Linux transition in 2001, cutting costs by an estimated $17 million per quarter. Etrade's 2006 case study highlighted the cost-effectiveness of multiple Linux-Intel boxes compared to a single, expensive Sun server. By 2001, Linux on Intel ("Lintel") began steadily replacing Sun's systems in various applications. Sun's attempts at scale-out with products like Netra failed to compete with the performance-per-dollar of commodity x86.

Going Upscale

After dominating the low-end, Linux set its sights on scaling up. In late 2000, IBM CEO Lou Gerstner announced a $1 billion investment in Linux, dedicating 1,500 employees.

January 2001 saw the release of Linux 2.4, which Red Hat quickly adopted. This kernel significantly closed performance gaps with high-end UNIX servers and NT, particularly addressing multiprocessor support. While Linux 2.2 struggled beyond two CPUs, Linux 2.4 improved scalability to 16 or even 32 processors, enabling Lintel to surge into the middle ground of compute tasks. The economics became highly compelling: a Dell server with four Xeons running Red Hat Linux cost half as much as a comparable Sun Fire UNIX box but performed equally well.

Red Hat and IBM aggressively targeted Sun's Solaris customer base, while HP and SGI began offering Linux solutions to their proprietary hardware customers to retain relationships. In 2002, Linux server hardware revenue grew 63%, while UNIX hardware revenue fell 11%. Although UNIX still generated more revenue due to higher machine costs, the pressure was evident, with the average selling price of a Sun server dropping 31% by 2004.

Opteron and x86-64

Intel initially pushed for 64-bit Linux support for its IA-64 (Itanium) architecture, launching Project Trillian in 1999 to confidentially provide Itanium prototypes for Linux port development. While Itanium didn't replace 32-bit x86, it was adopted by HP for its UNIX systems.

The true 64-bit transition for x86 came from AMD's groundbreaking Opteron chips in mid-2003. Intel later adopted AMD's backward-compatible 64-bit extension with its Nocona Xeon in 2004. This was a critical development, as 64-bit support removed one of the last major architectural barriers for switching from UNIX, allowing x86 systems to address dozens or hundreds of gigabytes of RAM, a capability previously exclusive to higher-end UNIX systems.

Windows' Counter-Strategy

The release of Linux 2.6 in late 2003, an "enterprise-ready" kernel with improved speed, networking, and support for up to 64 processors, coincided with Windows NT 4.0 users' upgrade cycles. IBM, HP, and Dell aggressively wooed these customers to Linux with training, app rewrites, and discounted machines.

Microsoft countered by emphasizing the cost and trouble of switching from Office and other Windows applications. Their 2004 "Get the facts" marketing campaign, which presented research favoring Windows over Linux, was controversial. A more effective strategy was the deep lock-in provided by Active Directory (AD), released with Windows 2000. AD authenticated and authorized users and assets within a Windows network, enabling centralized management, security, and single sign-on. This made switching from Windows a far more complex and costly endeavor than just an OS swap.

Consequently, Windows Server sales continued to grow, matching UNIX revenue in 2005 and surpassing it in 2006. Linux's growth primarily came from displacing UNIX or expanding into new markets like supercomputing, high-performance computing, web, and later, cloud computing, rather than directly eating into Windows' share.

Conclusion

UNIX server sales steadily declined but did not completely disappear. Some customers, particularly in banking and healthcare, still use Solaris, HP-UX, or AIX due to the challenges and risks of porting legacy applications to Linux.

The decline of UNIX was less about technology and more about business models. The vertically integrated UNIX kingdoms struggled against Intel's high-volume economics. Windows NT initially challenged UNIX, but Linux, offering similar capabilities without the vendor lock-in, proved to be an even more formidable competitor. Once Linux matured, UNIX vendors had little choice but to adapt or transition their customers to Linux solutions.

  Takeaways

  • Intel leveraged the Pentium Pro and later Xeon CPUs to offer high‑performance, low‑cost x86 servers that could rival RISC‑based UNIX machines, enabling the “Wintel” foothold in the enterprise market.
  • Microsoft’s Windows NT, despite early hardware demands and processor limits, grew rapidly in the mid‑1990s and, together with Intel’s hardware, established a viable alternative to proprietary UNIX systems.
  • The emergence of Linux, driven by the GPL‑licensed kernel and commercial distributions like Red Hat, provided a free, customizable OS that combined UNIX‑like reliability with dramatically lower licensing costs.
  • Intel’s strategic investment in Red Hat and its “OS neutral” port‑of‑choice policy accelerated Linux adoption across ISPs, Dell, and later large enterprises, while Microsoft responded with aggressive marketing and Active Directory lock‑in.
  • By the early 2000s, Linux on x86 hardware had closed performance gaps, outcompeted UNIX on price‑performance, and reshaped the server landscape, leaving UNIX to survive only in niche legacy environments.

Frequently Asked Questions

Why did Intel invest in Red Hat in 1998?

Intel invested because Linux was rapidly gaining market share among ISPs and offered a low‑cost, open‑source OS that could run on Intel’s x86 chips, reducing reliance on Windows NT and positioning Intel as a neutral hardware provider. The partnership also gave Intel a direct channel to steer Linux development toward better driver and scalability support for its servers.

How did Linux kernel 2.4 improve server scalability?

Linux kernel 2.4 introduced enhanced SMP support, allowing the OS to efficiently use up to 16‑32 processors, which closed the gap with high‑end UNIX systems and enabled commodity x86 servers to handle enterprise workloads. This scalability boost, combined with lower hardware costs, accelerated Linux’s displacement of UNIX in the early 2000s.

Who is Asianometry on YouTube?

Asianometry is a YouTube channel that publishes videos on a range of topics. Browse more summaries from this channel below.

Does this page include the full transcript of the video?

Yes, the full transcript for this video is available on this page. Click 'Show transcript' in the sidebar to read it.

Helpful resources related to this video

If you want to practice or explore the concepts discussed in the video, these commonly used tools may help.

Links may be affiliate links. We only include resources that are genuinely relevant to the topic.

Full transcript is not shown on this page

This page focuses on the summary and original notes. For full verification, refer to the original YouTube video.

PDF