In August of 1963, Thomas Watson Jr. , the president of IBM, the largest computer company on Earth, wrote a memo that became one of the most famous documents in the history of American technology. He could not understand how IBM, with its thousands of engineers and billions in revenue, had lost its leadership in computing to an operation of 34 people, including the janitor. Watson meant it as an indictment of his own organization.
But the man who had built the machine, a quiet, almost pathologically private engineer named Seymour Cray working out of a converted laboratory in the farming town of Chippewa Falls, Wisconsin, offered a response that outlived everything else in the story. The smallness was the point. The machine those 34 people built was not a little faster than anything IBM had. It was three times faster than the fastest computer on the planet, a record it held for five straight years.
It weighed 12,000 pounds and contained 400,000 transistors and more than 100 miles of hand-soldered wiring, cooled by Freon pumped through a cabinet shaped like a plus sign. It could perform three million floating-point calculations every second, a number that belonged in science fiction in 1964. Control Data sold a fully configured system for roughly seven million dollars. Over the next five years, more than a hundred shipped to weapons laboratories, nuclear research centers, and universities from Los Alamos to CERN.
This was the CDC 6600, the first true supercomputer. The company that built it rose from a rented warehouse in Minneapolis with $600,000 in seed money to a global enterprise employing 60,000 people with revenues approaching four billion dollars a year. Then it fell, not because a competitor destroyed it and not because the technology failed, but because the company that had beaten IBM with 34 people decided it would rather become IBM instead. William Norris was born in 1911 on a cattle farm outside Red Cloud, Nebraska.
He studied electrical engineering at the University of Nebraska and graduated in 1932 into the worst year of the Great Depression. In 1933 he found work selling X-ray equipment for Westinghouse in Chicago. Then the war came. The Navy made Norris a codebreaker, rising to Lieutenant Commander in Washington, where his team built machines to track and break German U-boat communications.
By the end of the war Norris had learned two things: computing machines could win wars, and the people who built them were worth more than the institutions that employed them. In January of 1946, Norris and a fellow cryptographer gathered 40 of their Navy codebreaking colleagues and founded Engineering Research Associates in St. Paul, Minnesota. ERA built some of the first commercial scientific computers in America, but government funding problems drained the company, and it was sold to Remington Rand, which merged it into Sperry Rand, which buried it in a corporate structure that valued sales quotas over engineering.
Among the engineers Norris had hired was Seymour Cray. By the time he arrived at ERA in 1950, he was already the best digital circuit designer most of his colleagues had ever seen. He avoided publicity, avoided meetings, avoided small talk. He dug a tunnel under his house in Chippewa Falls as a hobby and attributed his best design breakthroughs to elves who visited him there.
By 1957, both Norris and Cray had watched Sperry Rand suffocate ERA’s engineering culture under layers of corporate management. Norris decided to leave before the decline was complete and intended to take the best engineers with him. They followed, nearly all of them. On July 8th, 1957, Control Data Corporation filed its incorporation papers in Minneapolis.
The starting capital was $600,000, raised by selling 600,000 shares of stock at one dollar each. It was the first publicly financed computer company in American history. They set up operations in an old warehouse across the Mississippi River. Seymour Cray did not join on day one.
He stayed behind at Sperry Rand for several months, honoring his commitment to the Naval Tactical Systems Project. When he arrived at Control Data, he was given the title of chief design engineer and, far more important, the freedom to design whatever he wanted. The first major product was the CDC 1604, a fully transistorized computer at a time when most machines still ran on vacuum tubes. It shipped in January of 1960 and was by credible accounts the fastest computer available at its price point anywhere in the world.
Within three years of its founding, Control Data had 10,000 employees and $160 million in annual sales. But Cray was already restless. He wanted to build something that had never existed before: the fastest computer on Earth. Getting from the 1604 to the machine Cray had in mind required a fight that nearly broke the company apart.
Control Data’s board wanted commercial products for banks and insurance companies, the safe middle of the market where IBM lived. Cray wanted to build a scientific computer so fast that nothing else would come close. The argument escalated until Cray walked into Norris’s office and told him flatly that either the management structure changed or he was leaving. Norris made a decision that would define the company’s next decade.
He let Cray go home. Norris authorized Cray to establish a separate laboratory in his hometown of Chippewa Falls, Wisconsin, 185 miles from Minneapolis, far enough that no executive would drop by uninvited and close enough that Norris could drive up when it mattered. The lab was deliberately small. Cray staffed it with a handpicked team that eventually numbered 34 people in total.
Among them was Jim Thornton, an electrical engineer from St. Paul who became co-architect of the machine they were about to build. Cray designed on quarter-inch ruled paper pads by hand, without computer-aided design tools, using Maxwell’s equations to calculate the precise electrical path lengths on every circuit board so signals would arrive in perfect synchronization. He arrived at the lab early each morning, worked until mid-afternoon, then disappeared to sail on Lake Wissota or dig the underground tunnel beneath his property.
Colleagues learned that Cray would often return to the lab after dark, having solved a problem during the digging or the skiing that had resisted him at the workbench. Heat was the fundamental enemy. Four hundred thousand transistors packed into a cabinet small enough for electrical signals to stay in sync generated enormous waste heat. Cray designed a Freon-based refrigeration system that pumped coolant through channels in the plus sign-shaped frame, a physical architecture driven entirely by thermodynamics.
The shape was a heat exchanger with a computer inside it. Meanwhile, Control Data was burning through cash, and more than one board member questioned whether Norris had made a mistake in letting his best engineer disappear into the Wisconsin woods with no deadline and no oversight. Cray kept working. In 1964, they delivered the machine that changed everything.
The CDC 6600 was an ambush. When the first unit shipped to the Lawrence Livermore National Laboratory, it was three times faster than the IBM 7030 Stretch and roughly ten times faster than anything else commercially available. Cray and Thornton had stripped the central processor down to 65 instructions. The processor handled nothing but mathematics.
Every input and output task was managed by ten separate peripheral processors, small 12-bit machines that ran independently and simultaneously, feeding the central processor a continuous stream of data so it never had to wait. Thornton’s scoreboard, a piece of hardware logic that tracked every instruction flowing through the processor and released conflicting instructions automatically, allowed the 6600 to begin executing an instruction before the previous one had completed. The technique, now called out-of-order execution, would not appear in mainstream commercial processors for another 30 years. The 6600 operated so quickly that electrical signals could travel only about a foot during each clock cycle, which is why the cabinet was shaped the way it was, and why Cray had insisted on doing the work in a small lab where he could control every connection by hand.
When Control Data publicly announced the 6600’s specifications, Watson’s memo arrived before the first unit had even shipped. IBM’s response was the Advanced Computing Systems project, which consumed years of effort, employed hundreds of engineers, and never shipped a single machine. IBM also pre-announced products that existed only on paper, designed to convince potential Control Data customers to wait. When Control Data announced the 6600, IBM unveiled the System/360 Model 91 with specifications suggesting it would match or exceed the 6600’s performance.
IBM shipped a handful of Model 91 systems, but the machine never achieved its advertised benchmarks, and the project was quietly discontinued after fewer than 20 units were delivered. The announcement had served its purpose. Potential customers who might have ordered a 6600 had waited for the IBM machine instead. Norris refused to accept it.
In December of 1968, Control Data filed an antitrust lawsuit against IBM, alleging that IBM had fraudulently announced products it had no intention of building in order to destroy Control Data’s sales pipeline. In 1973, IBM settled. The terms included IBM’s entire Service Bureau Corporation, a timesharing operation with roughly 20,000 customers worldwide, handed to Control Data outright. The settlement was valued at more than $600 million.
A company that had started with $600,000 in stock had forced IBM to the table and walked away with a piece of IBM itself. But by the time Norris won the lawsuit, Seymour Cray had already left. The Cold War made Control Data. The market for machines like the 6600 arose because the United States government needed to simulate nuclear detonations, predict weather patterns for military operations, and decrypt the communications of its adversaries.
In its first five years, Control Data sold more than 100 6600 systems, and the majority of those buyers were governments, militaries, and government-funded laboratories. The Atomic Energy Commission alone accounted for a substantial fraction of early orders. The National Security Agency became one of the largest single customers. France and the United Kingdom purchased 6600 systems for their own nuclear weapons programs.
When the CDC 7600 arrived in 1969, it extended the reign. It ran at 36 megahertz, nearly four times the 6600’s clock speed, and was roughly ten times faster overall. Control Data sold 75 of them. For a full decade, from 1964 to the mid-1970s, Control Data held the title of the company that built the fastest computer on Earth.
But the CDC 8600, Cray’s most ambitious design, attempted to run four processors in parallel with an aggressive cooling system that proved unreliable. Norris, facing the simultaneous demands of funding the 8600 and a competing vector processor project, told Cray the company could not afford both paths. Cray, who had never accepted being told to choose between his design and someone else’s, chose neither. In 1972, Seymour Cray walked out of Control Data and founded Cray Research.
Norris did not try to stop him. He invested $250,000 of his own money in Cray’s new company. Within four years, Cray Research delivered the Cray-1 to Los Alamos, and within a decade it owned the supercomputer market Control Data had created. The strange truth about Control Data’s peak is that it arrived after its greatest engineer had already gone.
By 1981, the company employed roughly 60,000 people worldwide. Annual revenues exceeded $3. 5 billion. Control Data had become the fourth largest computer company in the world and the dominant employer in the Twin Cities.
But this was William Norris’s company now, and Norris had a vision that extended far beyond supercomputers. The peripherals division reached $1. 5 billion in annual sales and controlled 50 percent of the market. The company ran Cybernet, one of the first commercially available computing networks in the country.
It operated 69 vocational schools. It owned Ticketron, which generated $120 million in annual revenue. And then there was Plato, a computer-based educational platform that offered interactive lessons with touch-sensitive screens, graphical displays, and early forms of online messaging and discussion forums. Norris believed it would transform education worldwide.
It did not. The terminals were expensive. The maintenance was complex. Schools could not afford the infrastructure.
Plato saw almost no adoption in the education market and the hundreds of millions Control Data invested never returned. Norris also opened a manufacturing plant on the north side of Minneapolis in one of the most economically depressed neighborhoods in the city, hiring minorities, people with limited education, people with criminal records. He established mobile health clinics on the Rosebud Indian Reservation. Doing good was good business, Norris said, and he believed it.
But by the late 1970s, Control Data had acquired 88 companies across peripherals, data services, and financial services. The company that had beaten IBM with a small team in a Wisconsin farm town was now running vocational schools, ticketing systems, agricultural ventures, financial operations, and social programs. While the supercomputer market it had created was being captured by the company its own best engineer had founded. In 1976, Seymour Cray delivered the Cray-1, the fastest computer in the world, built by a small team under Cray’s direct supervision, sold to the same government laboratories that had been Control Data’s core customers.
Over 80 Cray-1 systems eventually sold, and every unit was a sale Control Data did not make. Control Data tried to answer. In August of 1983, it spun off a subsidiary called ETA Systems with a mandate to build a supercomputer capable of ten billion floating-point operations per second, roughly 40 times the speed of the Cray-1. The ETA-10 achieved 9.
1 billion operations per second, just short of its target. Seven liquid-cooled units and 27 air-cooled variants were sold, but the software lagged behind the hardware. By the time ETA had a marketable product, massively parallel machines built from hundreds of cheap commercial processors were delivering comparable performance at a fraction of the price. The era of the single monolithic hand-wired supercomputer was closing.
In April of 1989, Control Data shut down ETA Systems entirely. Robert Price flew from Minneapolis to ETA’s facility in St. Paul and delivered the news in person to an auditorium of engineers who had been told only that a mandatory meeting had been called. The announcement took less than an hour.
Seven hundred fifty employees were told their positions no longer existed. The liquid nitrogen cooling tanks were drained. The remaining machines were donated to high schools. Control Data permanently and irrevocably exited the supercomputer business it had created 25 years earlier.
The personal computer finished what the Cray-1 had started. Control Data’s peripherals business depended on the mainframe ecosystem. As personal computers displaced mainframes, the market contracted year after year. Norris was pushed out.
In January of 1986, William Norris stepped down as chairman. In 1985, Control Data posted a loss of $568 million and defaulted on a $300 million loan. In 1986, it lost another $265 million. In 1989, the company lost another $680 million.
Over five years, total losses approached $1. 3 billion. They sold the disk drive division to Seagate for $450 million. They sold Ticketron to the Carlyle Group, which flipped it to Ticketmaster within two years.
They spun off the financial services arm, Commercial Credit, and watched a financier named Sandy Weill build that single piece into Citigroup. They laid off 3,100 workers in 1989 alone. On May 27th, 1992, Control Data Corporation announced that it would cease to exist as a unified company. The computer operations were spun off into Control Data Systems, which was acquired by Syntegra, a subsidiary of British Telecommunications, for $340 million in the year 2000.
Everything else was gathered under a new name, Ceridian Corporation. The company that had reached revenues of $3. 7 billion and employed 60,000 people was carved into pieces whose combined sale prices did not approach what the whole had been worth at its peak. The name Control Data was retired.
Seymour Cray never saw the final chapter. On September 23rd, 1996, Cray was critically injured in a rollover accident on Interstate 25 near Colorado Springs. He died on October 5th, one week after his 71st birthday. He had never stopped designing.
His last company had filed for bankruptcy the year before, but Cray had already founded another venture and was sketching a massively parallel machine when the accident ended it all. A CDC 6600 sits today in the Computer History Museum in Mountain View, California, its plus-sign cabinet preserved behind glass, its 100 miles of internal wiring still intact. Another unit survives in the collection of the Science Museum in London. NCAR ran serial number 7 for over 11 years, from late 1965 to 1977, modeling atmospheric systems that no other machine could handle.
CERN used its 6600 systems to analyze particle collisions that would eventually help confirm the standard model of physics. The machines are more than 60 years old now, treated with the reverence usually reserved for works of art. Control Data did not die because it lost to IBM. It died because it stopped being the company that beat IBM.
The clarity that had produced the 6600, the relentless stripping away of everything that was not essential to speed, was replaced by the sprawl of a conglomerate that tried to be a computer company, a financial services firm, a ticketing operation, a school system, and a social enterprise all at once. The machine had succeeded because it did one thing and did it faster than anything else alive. The company failed because it tried to do everything and lost the one capacity that had made it extraordinary. The 6600 was proof that concentration of talent and purpose could defeat unlimited resources.
The company’s death was proof that diffusion of talent and purpose could squander any advantage, no matter how vast.
