In the mid‑1980s Microsoft quietly produced an obscure, hardware‑tied edition of OS/2 that, by the account of a long‑time Microsoft engineer, may have sold only eleven copies — eight of which were returned — leaving it to claim the dubious title of the company’s worst‑selling shipped product.
Background
The context: why cards like the Mach series existed
Personal computing in the 1980s was a period of explosive technical change and uneven budgets. Businesses were beginning to invest in PCs but accounting rules and depreciation schedules often discouraged wholesale replacement. Incremental hardware upgrades — devices that let an organization
extend an existing PC’s life — became an attractive middle ground.
A family of add‑on “Turbo” upgrade cards answered that need by replacing or augmenting the on‑board CPU and bundling extra functionality on a single expansion card. Microsoft’s Mach line followed that pattern: the Mach 10 was a CPU‑replacement card centered on a 9.54 MHz 8086, and the Mach 20 scaled the idea up with an Intel 80286 and daughterboard expansion options intended to give XT‑class machines AT‑class capabilities at a fraction of the cost of a new system.
Where OS/2 fits in the picture
OS/2 began life as a joint Microsoft–IBM project intended as the next major PC operating environment, with protected mode, better multitasking, and an ambition to supplant DOS‑based Windows on the enterprise desktop. Although OS/2 later became primarily an IBM product, in its early life Microsoft explored multiple distribution strategies and specialized builds; one of those experiments was an OS/2 image tailored to run on the Mach 20 card. The logic was straightforward: if the Mach 20 could make an XT behave like an AT, then shipping an OS/2 image tuned to that niche hardware could make the upgrade more compelling.
The hardware: Mach 10, Mach 20 and what they offered
Mach 10 — the first experiment
The Mach 10 replaced an 8088 in an IBM PC/XT with an adapter and a ribbon cable to a full‑length card carrying a 9.54 MHz 8086 processor and local cache RAM. The result was a noticeable CPU speed boost in certain workloads, plus a built‑in mouse port — a handy convenience at a time when expansion slots were in short supply. The card solved a real pain point (slot scarcity) by consolidating functions onto a single board.
Mach 20 — a much more ambitious upgrade
The Mach 20 pushed the idea into AT territory. Its headline features included:
- An Intel 80286 at higher clock rates than stock XTs, giving genuinely better CPU performance for protected‑mode operating systems.
- An 80287 floating point coprocessor socket for accelerated math workloads.
- Daughterboard connectors for memory expansion (the Mach 20 Memory Plus option could present several megabytes of RAM local to the card).
- A Disk Plus daughterboard that allowed connection of additional floppy drives without consuming a motherboard slot.
- Design choices meant to save on expansion slots — a crucial selling point in the PC/XT era.
Contemporary reviews put the Mach 20’s retail price in the roughly several‑hundred‑dollar range (InfoWorld reported $495), which was a lot for a card but far less than an IBM PC AT. Still, that price didn’t include the additional memory and storage work necessary to run ambitious operating systems well.
OS/2 for Mach 20: the product and the claim
What the product was
Microsoft produced a special build of OS/2 intended to run on the Mach 20 platform. The idea was to let Mach 20 owners run OS/2’s protected mode features and, in theory, enjoy OS/2’s more modern multitasking and system services without buying a new AT‑class machine.
The sales anecdote
Raymond Chen — a veteran Microsoft developer and author of the Old New Thing column — relayed an anecdote based on conversations with former support staff: according to the memory of a technical support specialist, only eleven boxed copies of “OS/2 for Mach 20” were ever sold, and eight purchasers returned their copies, leaving three final non‑returned units. Chen explicitly frames this as recollection and labels it as legend rather than a corporate ledger. That exact “11 sold, 8 returned” tally has been repeated by multiple tech retrospectives and enthusiast outlets, which typically cite Chen (or reuse the same anecdote) while noting its oral‑history nature.
Verifying the claim: what can and cannot be confirmed
Firmly verifiable points
- Microsoft shipped Mach‑series hardware and did produce a Mach‑specific edition of OS/2, as documented by period reviews and later recollections. Contemporary trade press and retrospectives confirm the hardware’s capabilities, price range, and the existence of an OS/2 image targeted at the card.
- The Mach 20’s design choices — local memory, daughterboard expansion, 80287 socket — are documented in reviews and period literature. Those specs are corroborated by multiple independent sources and by archived reviews quoting InfoWorld and other contemporaries.
The weak link: the sales numbers
- The specific sales number (eleven sold, eight returned) originates from recollection reported by Raymond Chen. There is no known public sales ledger, shipping manifest, or Microsoft published figure to independently confirm the exact quantity. Several modern write‑ups repeat Chen’s account verbatim but flag it as anecdotal or legendary. That means the numerical claim is plausible but not definitively proven in public records. Treat this as oral history rather than audited fact.
What historians and collectors still look for
Enthusiasts and archivists have searched for surviving boxed copies, original floppies, and documentation that would definitively prove packaging and shipment volumes. The absence of scanned invoices or shipment records online prevents turning Chen’s remark from a colorful anecdote into a documented sales figure. Preservationists consider such items “final boss” collectibles precisely because they can resolve these kinds of historical ambiguities if discovered.
Why the Mach 20 + OS/2 combination was especially fragile
Technical mismatches and performance bottlenecks
A Mach 20 card provided a faster CPU and local memory, but the rest of an XT platform — notably the 8‑bit ISA bus and slow MFM disk subsystems common in XTs — remained a major bottleneck. OS/2’s protected‑mode features and multitasking model placed far more demand on memory and disk performance than DOS programs did. Running OS/2 effectively required:
- Additional memory (OS/2 1.x required significantly more RAM than a typical XT could support without the Mach memory daughterboard).
- Faster disk I/O than the original XT floppy/MFM setups could reasonably provide.
- Careful driver support to reconcile timing, bus semantics, and nonstandard devices attached through the Mach card.
The combo of a niche OS and a niche hardware add‑on multiplied friction: buyers had to purchase the Mach 20, the Mach memory expansion (if they didn’t already have sufficient RAM), and possibly other upgrades to make OS/2 usable. By the time those costs and complications were accounted for, the economic case versus buying an inexpensive new AT or a different upgrade path often disappeared.
Support burden and documentation gaps
Specialized OS builds demand specialized drivers and unique installation steps. Documentation that assumed a level of familiarity with low‑level hardware and manual configuration would have been a barrier for many buyers. Support teams faced an outsized workload for a tiny user base: every edge case was expensive to troubleshoot when few customers existed to share knowledge and when the hardware itself was uncommon. That explains in part why returns could pile up quickly.
Preservation, provenance, and the collector’s angle
Rarity and archival value
If Chen’s recollection is even close to accurate, surviving copies of OS/2 for Mach 20 are extraordinarily rare and would be prized artifacts for software historians. The combination of fragile floppy media, niche hardware requirements, and low distribution volume makes this an archetypal preservation challenge: even if a copy exists in a personal archive, it might be unreadable without specialized hardware, and the Mach 20 card itself is itself a scarce artifact.
How collectors and archivists approach this class of artifact
- Hunt for boxed copies, original installation floppies, printed manuals and invoices.
- Preserve the hardware needed to boot the image (a Mach 20 card or an emulator that accurately reproduces its behavior).
- Document oral histories with dates, job titles, and cross‑references to period documents so that anecdote can be evaluated against physical evidence.
Critical analysis: lessons for product teams and the modern parallels
What the Mach 20 episode teaches product managers
- Niche hardware + niche software multiplies risk. Combining two narrow markets reduces addressable customers exponentially unless the value proposition is compelling and crystal clear. The Mach 20 + OS/2 pairing created multiple friction points customers had to cross before success was possible.
- Engineering effort and support costs don’t scale down with customer count. Building a specialized OS image is costly; supporting idiosyncratic hardware setups is expensive; when only a handful of customers exist, per‑customer support costs can make the entire effort uneconomic.
- Expectation management and documentation are decisive. Many returns likely stemmed from buyers who didn’t fully appreciate the hardware and upgrade steps required. Clear marketing, accurate system requirements, and straightforward installation guidance can dramatically reduce returns — and yet these are often the last things to be prioritized on short‑lived niche projects.
Modern echoes
The structural lessons still apply today. Consider OS ports for unusual SoCs, vendor‑specific firmware, or heavily customized enterprise builds: when a software vendor builds for a tiny piece of hardware, the support burden and fragility of that bet remain the same. Modern teams mitigate risk by:
- Using virtualization and emulation to broaden testing coverage.
- Publishing detailed compatibility matrices up front.
- Pricing support and lifecycle management into the product business case.
The Mach 20 story is a reminder that clever engineering alone doesn’t guarantee market success; distribution economics, support, and clarity of value matter just as much.
Cross‑verification of key claims (short checklist)
- The Mach 10 and Mach 20 were real Microsoft‑branded upgrade cards and offered CPU replacement and daughterboard expansion features.
- Microsoft produced a Mach‑specific OS/2 image. Multiple retrospective accounts and period press corroborate that such a build existed.
- The “11 sold, 8 returned” figure originates with Raymond Chen’s Old New Thing column and is presented there as recollection rather than a formally audited number; it has been echoed by other outlets but not proven by primary sales documentation publicly available. Flagged: anecdotal / unverified in public records.
- Mach 20 pricing and the real‑world requirements for OS/2 (notably memory and disk performance) are documented in contemporaneous reviews and explain why the combination underperformed in practice.
Conclusion
The Mach 20 and its accompanying OS/2 image present a compact, illuminating case study in the risks of combining two niche bets: specialized hardware and a specialized operating system. The oft‑repeated tally — eleven boxes sold and eight returned — comes to us as a credible but unverifiable oral history from former Microsoft support staff, preserved by a veteran engineer and amplified by retrospective tech coverage. Whether those exact numbers are ultimately provable or not, the more important story is structural: a technically clever product can still fail spectacularly if the supporting ecosystem, documentation, and economic case aren’t there.
For historians and collectors, the Mach 20 OS/2 artifact remains a tantalizing prize: a rare intersection of hardware ingenuity, corporate experimentation, and the messy realities of real‑world computing in the pre‑PC‑mainstream era. For product teams today, it stands as a cautionary tale — a reminder that integration costs, support overhead, and customer expectations are as decisive as the silicon and code.
(Primary recollection and the figures discussed are documented in a first‑hand account by a Microsoft developer. For readers who want to examine the original recollection and contemporaneous reporting, the Old New Thing column and later retrospective coverage reproduce the anecdote and the hardware details; they should be treated as oral history and corroborated against physical artifacts where possible.
Source: Softonic
The worst product in Microsoft's history: it sold 11 units and 8 were returned - Softonic