#613 1985 · Intel Corporation · Semiconductors
Intel stopped giving away half its best chip's profit to a rival it no longer needed for insurance
the problem
Buyers refused single-vendor chips, forcing Intel to license its own designs to competitors
background
Since the earliest days of mass-market semiconductors, buyers — first the military, then computer makers — refused to build a product around a chip only one company could manufacture, fearing a single factory problem could halt their own production line. "Second-sourcing" became the price of every design win: a chipmaker had to license a rival to build an independent, interchangeable version of its own part before any major customer would commit to it. Intel had played by this rule for a decade, licensing AMD and others to build compatible copies of its 8086 and 80286 processors under a 1982 technology-exchange agreement IBM itself had required.
By 1985 Intel had just booked the worst loss in its history, exiting the memory-chip business it had been founded on, and had sunk four years and roughly $100 million — more than ten times any earlier chip's budget — into its next processor, the 386. Breaking with second-sourcing risked alienating IBM, the industry's gatekeeper, and AMD, Intel's own licensed partner, at the exact moment Intel could least afford to lose a customer.
what everyone would do
The safe move, especially coming off the worst year in company history, was to keep doing what every chipmaker had always done: license a second manufacturer so buyers were never dependent on one vendor's factory. Refusing to do that for your flagship next-generation chip risked losing both your biggest customer and your own licensed partner at the moment you could least afford either.
what they saw
Customers never actually wanted a second company legally entitled to make the chip — they wanted a guarantee they'd never be stranded without supply. Second-sourcing was only ever a proxy for that guarantee, adopted back when any single company's fabs were too unreliable to trust alone. Intel's own multiple factories, now running an identical, tightly disciplined process, could supply that same guarantee internally — the redundancy a license used to buy could now come from Intel's second and third factory instead of from a rival's.
the move
Intel refused to license any competitor to manufacture the 386, ending a practice every major chipmaker had followed for its entire history. It bet that its own newly disciplined "Copy Exactly" manufacturing process, running in duplicate across multiple internal factories, could give customers the same guarantee against a production failure that an external second source used to provide — without Intel having to hand a rival the design or split the margin.
why it works
Because Intel's duplicate fabs made a single-factory failure survivable, customers no longer needed an external second source as insurance against that specific risk — the risk the whole convention existed to cover had quietly moved in-house. Once one credible buyer (Compaq) was willing to build around Intel alone, the gatekeeper withholding demand for a second source (IBM) had nothing left to hold out for: it either adopted the chip on Intel's terms or watched a rival ship it first. That shifted the standoff from persuasion to market pressure, and once Compaq's machines sold, Intel kept every dollar of margin a licensee would otherwise have taken, right as PC volumes were about to explode.
the payoff
IBM initially balked, wary that a 386-based PC would undercut its own higher-end machines, and did not adopt the chip. Compaq broke ranks and released the Deskpro 386 in 1986, the first PC built around the chip without IBM's blessing; IBM returned to the table seven months later but never regained its former leadership of the PC platform. Intel kept the entire margin on the most important processor generation of the PC era instead of splitting it with a licensee: the PC platform's own market share grew from 55% in 1986 to 84% by 1990, and by 1992 Intel held roughly 83% of the microprocessor market and over $1 billion in annual profit. AMD sued for breach of the 1982 agreement; the dispute ran through arbitration and courts for nearly a decade and cost both companies over $100 million before settling in 1995, by which point Intel had already captured years of exclusive high-margin leadership.
where it breaks
It requires your own reliability to be genuinely improved, not merely asserted — customers only tolerate a sole source once real years of near-flawless supply have removed their fear, and moving too early just recreates the shortage the convention was built to prevent. It also requires an alternative buyer willing to defect from whichever customer benefits from the old assurance; without a challenger ready to move first, the incumbent gatekeeper can simply refuse to adopt the product and starve it. And ending an existing license relationship carries real legal exposure if the terms aren't clean — Intel spent nearly a decade and over $100 million in litigation with AMD over this exact move, even though it ultimately won the market.
what came after
The industry's decades-old second-sourcing convention collapsed within a few years as competitors followed Intel's example; sole-sourcing a flagship chip became the semiconductor industry's new default rather than the exception. Harvard Business School and the Computer History Museum both later built dedicated case studies and oral-history panels around the decision as a turning point in computing history.
references
- [1]Intel's disruptive move with the 386InfoWorld, 2009infoworld.com
- [2]Intel and the x86 Architecture: A Legal PerspectiveHarvard Journal of Law & Technology Digest, 2011jolt.law.harvard.edu