The One-Way Door
Real Options and Reversibility at Intel in October 1985
In morning in October 1985, inside a fourth-floor executive office at Intel Corporation's headquarters at 3065 Bowers Avenue in Santa Clara, California, President Andrew S. Grove sat across from Chief Executive Officer Gordon E. Moore while their core business disintegrated. For seventeen years, since Moore and Robert Noyce founded the company in July 1968, Intel's identity had been synonymous with Dynamic Random Access Memory, or DRAM. In October 1970, Intel had introduced the 1103 chip, a 1-kilobit semiconductor memory device that rendered magnetic core memory obsolete and made Intel the world's leading memory manufacturer.
By the autumn of 1985, however, Japanese industrial conglomerates—led by Hitachi, NEC, Fujitsu, and Toshiba—had mastered clean-room yields and flooded the global market with 64-kilobit and 256-kilobit DRAM chips. Every time Intel lowered its prices to match the Japanese quotes, Hitachi and NEC cut prices another 10 percent. In 1984, a 64-kilobit DRAM sold for $3.50; by October 1985, it sold for 35 cents, well below Intel's manufacturing cost. Intel's global DRAM market share had collapsed from over 80 percent in the early 1970s to 1.3 percent. For more than a year, Grove and Moore had debated incremental compromises in staff meetings: building specialized niche DRAMs, licensing designs to Korean foundries, or automating Oregon wafer lines to save a few pennies per die.
Looking out the window at the Ferris wheel of the Great America theme park next door, Grove asked Moore what a replacement chief executive would do if the board fired them both. Moore answered immediately that a new chief executive would exit the memory business. Grove proposed that the two of them walk out the door, walk back in, and execute that exit themselves.
Walking out of DRAM and betting Intel's survival on microprocessors—specifically the newly designed 32-bit 80386 chip containing 275,000 transistors—required crossing two irreversible thresholds at once. In corporate strategy and financial economics, commitments divide into two topological classes based on the cost of reversing them once new information arrives. A reversible choice—a two-way door—can be executed, tested against market feedback, and undone for a modest switching cost if the hypothesis fails. An irreversible choice—a one-way door—permanently alters the state space by liquidating physical assets, severing contracts, or forfeiting market architecture in ways that cannot be bought back at the original price.
In October 1985, Intel's leadership confronted a stark balance-sheet asymmetry:
- * The Reversible Illusions (Incremental DRAM Tweaks): Reallocating 10 percent of wafer starts at Fab 5 in Aloha, Oregon, felt safe because management could reverse the schedule next quarter. Yet these reversible adjustments bled $109 million in annual memory operating losses without altering the Japanese cost advantage.
- * First Irreversible Door (Exiting DRAM Manufacturing): Shutting down Intel's memory division meant closing seven of the company's eleven wafer fabrication and assembly plants—including Fab 2 in Santa Clara, Fab 5 in Oregon, and assembly sites in Barbados and Puerto Rico—laying off 7,200 of Intel's 25,400 employees, and taking a $173 million restructuring write-off. Once Intel disbanded its memory process teams and surrendered its customer sockets to NEC and Toshiba, re-entering the DRAM commodity race later would be impossible.
- * Second Irreversible Door (Single-Sourcing the 80386): Up to 1985, IBM and major computer buyers refused to purchase any microprocessor unless its creator licensed the masks to second-source competitors such as Advanced Micro Devices (AMD). For the 80386, introduced on October 17, 1985, Grove decided to refuse all second-source licenses, turning the x86 architecture into a proprietary monopoly. If IBM rejected the single-source terms and switched to Motorola's 68020 chip, Intel would have no memory business left to cushion the blow.
Why had Grove and Moore delayed stepping through the one-way door for fourteen months? In the formal language of real options—pioneered by Stewart Myers at MIT in 1977 and expanded by Avinash Dixit and Robert Pindyck—whenever a divestment is irreversible and future prices are uncertain, holding the uncommitted status quo carries a positive economic premium known as the option value of waiting. Waiting until October 1985 allowed Intel to confirm two facts: first, that Japanese yields on 256-kilobit chips were structurally superior rather than a cyclical blip, and second, that early engineering silicon of the 80386 worked cleanly enough to sustain a standalone company.
Once that informational uncertainty resolved, the option value of waiting fell to zero. In November 1985, Grove walked into the Santa Clara auditorium and announced the termination of Intel's DRAM operations. He reassigned the surviving silicon-process engineers from the closed memory fabs to accelerate sub-micron lithography for the 80386, and he informed IBM that Intel would never license the 386 masks to AMD. Compaq Computer broke ranks first in September 1986 by launching the Deskpro 386 ahead of IBM, forcing the entire personal computer industry to adopt Intel's single-sourced chip. By 1992, seven years after closing its last DRAM line, Intel's annual net profit passed $1.07 billion, and the company became the largest semiconductor manufacturer on Earth.