Steve McGeady, a friend and former colleague at Intel, posted a link to an YouTube video that covered the creating of BiiN a joint venture between Intel and Siemens. I played a significant role in making BiiN happen. It was one of many failures that I had in my career. Watching the video and remembering that time, I thought I would write something about it. Then I get a an idea. I had already written a chapter in my book, The Flight of a Wild Duck, so I asked ChatGPT to create a blog post from that chapter which you can read below.
You can also read the full chapter here.
Adapted from my memoir, The Flight of a Wild Duck.
When I joined Intel in August 1984, I assumed I was joining a semiconductor company that understood where computing was going. What I found was a company that, for a while, wanted to be a minicomputer company. This is the story of the three years I spent on that detour, and what I learned from it.
Arriving at Intel
I came out to California without my family and spent my first few weeks at Intel’s site on Bowers Avenue in Santa Clara. Intel denied having a headquarters, but this sure came close. Gordon Moore and Andy Grove had offices there, and even Bob Noyce, then chairman of the board, still had one. It was a good place to meet the people who ran the company.
I was going to be based in Hillsboro, Oregon, so my boss, Les Vadasz, and I decided I should attend the staff meetings of the Systems Group headquartered there and see if I could help. Much to my surprise, I discovered they were developing personal computers to sell as a “white label” product to companies that wanted their own branded PCs.
My experience at Digital and then at Franklin made the cost side of this something I knew well. I quickly concluded the plan was a serious mistake. It would create friction with Intel’s own microprocessor customers, it was unlikely to succeed, and even if it did, it would not be very profitable. Through my industry contacts, I got pricing from other manufacturers willing to build private-label PCs. Their prices were lower than Intel’s cost to build the same machine. The project was terminated.
The company that didn’t see the PC coming
It is hard to believe today, but Intel did not anticipate the personal computer. By the mid-1970s, Gordon and Andy understood that the microprocessor would matter, and they pushed for a “mainframe on a chip.” But they believed the industry would go on being dominated by expensive, proprietary mainframes and minicomputers from companies like IBM and Digital Equipment.
The first hobbyist personal computer, the Altair 8800, used an Intel 8080 in 1975. Still, Intel showed little interest when Atari asked about the 8080 for its home computers, so Atari chose the 6502 from MOS Technology. Steve Jobs had worked at Atari, and Apple followed it to the 6502. When the Apple II, the TRS-80, the Commodore PET and the Atari VCS all came out in 1977, none of them used an Intel chip.
Meanwhile, Intel put its big bet on an ambitious new architecture, the 432, meant to incorporate the best academic ideas in computer design. Because it was risky, a simpler 16-bit chip, the 8086, was developed in parallel and released in 1978. Then came the critical decision to build an 8-bit-bus version, the 8088. IBM chose it for the PC because it fit the I/O design of the Datamaster, an IBM word processor whose motherboard the PC team wanted to reuse. The 432 did not ship until 1981, and it turned out to be too complicated for both the market and the semiconductor technology of the day.
Lew Eggebrecht, the architect of the IBM PC, told me personally that Intel’s development systems, its compilers and debugging tools, were the most critical factor in his choice. That was the genius of Bill Davidow, who co-managed the microprocessor business with Les. He got Intel to understand that the product was not the device. It was everything that made the device successful: development systems, documentation, support and reference designs. That is how Intel beat Motorola with an inferior chip. Many credit Davidow’s famous Operation Crush campaign for the IBM win. I don’t believe it had anything to do with it.
Had IBM not chosen the 8088, Intel would likely never have led in microprocessors. It probably would not exist today. Until the IBM PC took off, microprocessors were less than 10 percent of Intel’s revenue.
Part of the problem was that Intel’s founders, Andy included, did not really understand computer architecture or even digital logic. They were process people, chemists and physicists. Memory chips, Intel’s main business until 1984, are far simpler than microprocessors and need no software. Management lacked the experience to judge a project as complex as the 432. The exception was Dave House, who came from computer systems and spoke out against it.
Gemini: a minicomputer company after all
Before the PC took off, Andy came to believe that half of Intel’s revenue should come from selling systems. The computer industry was still vertical, and IBM and Digital were very profitable. In 1980, Les was asked to figure out how to get there. Sudhir Bhagwan and Ed Slaughter proposed a family of multiprocessing, distributed, fault-tolerant computers to take on Digital and Tandem. The price tag was $100 million, far beyond what Intel could afford on $350 million in annual revenue. Andy told them to find a partner who would put up 90 percent of the cost for 50 percent ownership.
That partner was Siemens. They needed a highly reliable computer for factory automation and nuclear plants, they had a struggling commercial computer business, and they had a lot of money. The agreement was signed in 1982, and the project was named Gemini, for the two companies. Siemens paid for the computer development and Intel for the chips, with Oregon as the main site.
In 1983, two senior Intel executives, Casey Powell and Scott Gibson, walked out with fourteen others to start Sequent Computer Systems, a competitor with a simpler approach. Sequent later moved to Intel’s 386 and Unix, partnered with Oracle, and was bought by IBM in 1999 for $850 million.
The heart of Gemini was a processor architected by Glen Myers that became the i960. It was the first microprocessor that could execute multiple instructions at once, a 33-bit machine. The extra bit tagged every word in memory as either ordinary data or a protected pointer to an object, and only privileged instructions could set it. A program could not forge a pointer or reach outside its own boundaries. Combined with redundant processors that checked each other, the system could detect and recover from its own failures. It was built for Ada, the language the Defense Department had commissioned. The architecture was elegant and incredibly overengineered. It made the 432 look simple, which was no easy feat. It needed four highly complex chips, a new operating system and compilers for several languages.
When I looked closely, I found three engineering groups and not a single businessperson. I reluctantly concluded I had to take responsibility for the business side, at least for Intel. I had joined Intel hoping to take the company into the consumer market. This was the opposite. But I felt it was my duty, because I knew more about the computer industry than anyone else at my level.
My reluctance was justified. Revolutionary products rarely succeed. I was once again on an overly ambitious project, like the Professional I had led at Digital, and this one was a lot less fun. I moved to Jones Farm in Hillsboro and stayed three years, isolated from mainstream Intel. I wondered whether I had made a significant mistake in taking the job and moving my family to Portland. Intel’s stock fell 40 percent in my first years, then crashed again on Black Monday in 1987, along with my dreams of paying for my children’s university. Fortunately, it recovered in time for my oldest son, Adin, to go to Brandeis University.
We did find some good homes for the i960 as a component, in markets that needed high reliability: military, avionics and workstations. We even tried to get Steve Jobs to use it for NeXT. Boeing used it in the 777, its first fly-by-wire airplane.
BiiN
The original agreement left Intel and Siemens free to compete with each other using the same technology. I knew that without a change, our partner would soon become our competitor. So I proposed a joint-venture company. I spent the better part of a year negotiating it, flying to Germany every month to meet my Siemens counterpart, and we got the approval of both boards. Along the way I worked on the branding, and we named the company BiiN. It was announced in July 1988. By then I had already left the project for corporate business development in Santa Clara, the work that would ultimately lead to Intel Capital.
Both companies wanted a well-known CEO. I don’t think I was ever considered, and I would not have taken the job. Thankfully, I was done with big computer systems. They hired Joe Kroger, who had been the number-two person at Unisys. Joe was a big-company guy and fundamentally a sales guy, and BiiN had nothing ready to sell. He was not dealt a good hand. Siemens had its own internal politics, including a competing effort with Sequent, and Intel was now totally focused on the PC. In 1989, both companies pulled the plug. Six years of work by hundreds of engineers were flushed.
Inside Intel, people joked that BiiN stood for “Billions Invested in Nothing.”
The joke exaggerated, but not by much. When the plug was pulled in October 1989, the trade press reported that the two partners had put about $250 million into the venture through that July, and that winding it down could cost up to another $100 million. Intel took a $35 million write-off for its share of the shutdown. Those figures cover only the joint venture. They leave out the years of chip development Intel paid for on its own under the original Gemini agreement, which never showed up as a separate number anywhere. All told, the two companies probably spent well over $300 million.
Not quite nothing
Still, the joke missed something. Intel never recouped its investment directly, but Gemini had brought in many talented computer scientists and engineers. After BiiN closed, Intel created the Intel Architecture Labs to give them a home, led by Ron Whittier. People like Kevin Kahn, Fred Pollack and Steven McGeady, who had come to Intel for Gemini, went on to play very important roles. IAL was essential to Intel’s evolution, and I would later work closely with it on residential broadband.
The chip at the heart of Gemini also earned its keep. Freed from BiiN, the i960 became one of the best-selling embedded processors of the 1990s. From 1993 to 1995, Intel shipped more than five million a year, more than double any other RISC chip. Nearly half went into HP LaserJet printers, and about a third into networking equipment from companies like Cisco. By Intel’s standards it was never a big earner, a tiny fraction of what the Pentium made, but it stayed in production until 2007.
For Les, the failure must have been incredibly painful. Gemini was his baby, started when everyone at the top believed Intel’s future was selling computers as well as chips. Les and Andy appreciated what I had done in putting BiiN together, and I had shown I could get things done. But I considered it three years of wasted effort.
Looking back, I should have seen that launching a new proprietary computer architecture was impossible just as the PC was transforming the world of computing. Like so many others, including Intel’s top management, I was slow to grasp what was happening in front of us: the restructuring of the whole computer industry, from vertical to horizontal. The very technology Intel was making was the thing driving that change.
As it happened, though, I now found myself in the right place at the right time. I moved to Silicon Valley and began the most significant period of my career, co founding Intel Capital and playing a major role in establishing broadband internet.
Leave a Reply