Seymour Cray Left Control Data to Build the World's Fastest Computers

A $1 Billion Company Built on a Handshake and a Resignation

In 1972, Seymour Cray walked away from Control Data Corporation, the company where he had designed the CDC 6600 and 7600, the fastest computers in the world. He had sold his CDC stock back to the company for $1 million and moved to Chippewa Falls, Wisconsin, to start Cray Research. The first machine, the Cray-1, shipped in 1976 and sold for $8.8 million. By 1985, Cray Research held 70 percent of the supercomputer market. The company’s rise was not a story of marketing genius or venture capital. It was a story of one engineer’s refusal to build anything but the fastest machine possible, and the business structure he created to protect that obsession.

Cray was born in 1925 in Chippewa Falls, a small town in northern Wisconsin. He studied electrical engineering at the University of Minnesota and joined Engineering Research Associates (ERA) in 1950, a company founded by former Navy codebreakers. ERA was acquired by Remington Rand in 1952, and Cray worked on the UNIVAC 1103, one of the first commercial computers. In 1957, he left with several colleagues to found Control Data Corporation, where he led the design of the CDC 1604, the first commercially successful transistorized computer. The 1604 was a breakthrough, but it was the CDC 6600 that made Cray famous.

The 6600, released in 1964, was ten times faster than any computer on the market. It used a technique called functional parallelism, with separate units for arithmetic, logic, and memory access. The machine was so fast that IBM, the dominant player in computing, launched a project called the “Superior” series to catch up. Cray’s design was so far ahead that the 6600 was still in use a decade later. But Cray was not interested in corporate management. He moved to a secluded lab in Chippewa Falls, away from Control Data’s headquarters in Minneapolis, and refused to attend meetings. He worked on the 7600, the follow-up to the 6600, but Control Data’s management wanted him to focus on smaller, cheaper machines. Cray left in 1972.

The resignation is the part of the story that gets retold as temperament — the genius who would not sit in meetings. It was also a financing decision. Cray had already sold his CDC stock back to the company for $1 million, and that money became the founding capital for Cray Research. The structure he built around the new company was deliberately thin: a small team, a single product line, and a customer list short enough to fit on one page. That is what allowed a company selling a handful of machines a year to reach $120 million in revenue by 1980 and $420 million by 1985. The handshake culture and the resignation were not quirks surrounding the business. They were the business.

The Cray-1 and the Birth of the Supercomputer Industry

Cray Research was founded in 1972 with $1 million of Cray’s own money and additional funding from investors. Cray’s goal was to build the fastest computer in the world, regardless of cost. The Cray-1, released in 1976, was a 64-bit vector processor with a clock speed of 80 MHz. It was the first computer to use vector processing, a technique that allowed the machine to perform operations on arrays of data simultaneously. The Cray-1 was also famous for its distinctive circular design, with the cooling system integrated into the chassis. It sold for $8.8 million, and the first customer was Los Alamos National Laboratory.

The Cray-1 was a commercial success, but Cray Research’s business model was unusual. The company did not sell many machines; it sold a few, very expensive ones. By 1980, Cray Research had sold about 60 Cray-1s, generating $120 million in revenue. The company went public in 1976, and its stock price soared. Cray became a millionaire many times over, but he continued to live in Chippewa Falls and work on the next machine. He was not interested in the business side of the company; he left that to John Rollwagen, who became CEO in 1981.

Cray’s next machine, the Cray X-MP, was released in 1982. It was a multiprocessor version of the Cray-1, with two or four processors working in parallel. The X-MP was faster than the Cray-1, and it was used for weather forecasting, nuclear weapons simulations, and oil exploration. By 1985, Cray Research had sold over 100 X-MPs and had a market share of 70 percent in the supercomputer market. The company’s revenue reached $420 million in 1985, and its stock was a Wall Street favorite.

The Cray-2 and the Limits of Speed

In 1985, Cray Research released the Cray-2, the fastest computer in the world. It used a new cooling system with liquid fluorocarbon, and it was capable of 1.9 gigaflops. But the Cray-2 was a difficult machine to build, and it was expensive. It sold for $17 million, and only about 25 were sold. Cray was already working on his next design, the Cray-3, which he promised would be 12 times faster than the Cray-2. But the Cray-3 was delayed, and Cray Research’s management was losing patience.

In 1989, Cray left Cray Research to found Cray Computer Corporation, a new company based in Colorado Springs. He wanted to build the Cray-3 using gallium arsenide chips, a new technology that promised higher speeds than silicon. But gallium arsenide was difficult to manufacture, and the Cray-3 was delayed repeatedly. Cray Computer Corporation went public in 1990, but the company struggled. The Cray-3 was finally released in 1993, but only one was sold. Cray Computer Corporation filed for bankruptcy in 1995.

Seymour Cray died in 1996 after a car accident. He was 71. Cray Research, the company he founded, was acquired by Silicon Graphics in 1996 for $740 million. The Cray brand was later acquired by Tera Computer Company, which renamed itself Cray Inc. In 2019, Cray Inc. was acquired by Hewlett Packard Enterprise for $1.3 billion. The Cray name is still used for HPE’s supercomputers.

The Business of Building the Fastest Machine

Cray Research’s success was not just about technology. It was also about the business model. Cray sold a few, very expensive machines to government labs, universities, and large corporations. The company did not spend much on marketing; its reputation was built on benchmark results. Cray Research’s customers were willing to pay a premium for speed because their work depended on it. Weather forecasting, nuclear simulations, and oil exploration required massive computational power, and Cray was the only company that could deliver it.

The mechanism is worth spelling out, because it is the opposite of how most hardware companies grew. A conventional vendor spreads risk across many product lines and many buyers. Cray Research concentrated it: one architecture at a time, a customer base measured in dozens, and a price that captured nearly the whole value of being first. That works only when the buyer’s alternative is waiting. National laboratories and defense programs could not wait, so they paid $8.8 million for a Cray-1 and $17 million for a Cray-2. The model produced extraordinary margins and a 70 percent share, but it also meant that a single delayed machine, or a single customer deciding to skip a generation, moved the whole company.

But the supercomputer market was small and cyclical. Cray Research’s revenue peaked at $800 million in 1987, and then declined as the market matured. The company faced competition from Japanese manufacturers like Fujitsu and NEC, which were subsidized by their government. Cray Research also faced competition from parallel processing startups like Thinking Machines and Intel, which were building machines with thousands of microprocessors. Cray was skeptical of massively parallel processing, and he preferred to build machines with a few very fast processors. His approach was vindicated for a time, but eventually, the industry moved toward parallel processing.

Cray Research’s decline was also due to management missteps. The company invested heavily in the Cray-2 and the Cray-3, which were expensive and difficult to sell. The company’s stock price fell, and it became a takeover target. In 1996, Silicon Graphics acquired Cray Research, and the Chippewa Falls lab was eventually closed.

What Seymour Cray’s Career Teaches About Technical Leadership

Seymour Cray was not a businessman. He was an engineer who happened to found a company. He was notoriously shy and avoided public speaking. He preferred to work alone, and he designed his machines on paper before building them. He was also stubborn; he refused to build machines that he did not believe in. When Control Data’s management wanted him to build smaller machines, he left. When Cray Research’s management wanted him to focus on the Cray-2, he left to build the Cray-3.

Cray’s approach was to build the fastest machine possible, regardless of cost. He did not care about market share or revenue growth. He cared about speed. This obsession led to some of the most important computers in history, but it also led to business failures. The Cray-3 was a technical marvel but a commercial flop. Cray Computer Corporation went bankrupt. But Cray’s legacy is secure. He is remembered as the father of the supercomputer, and his machines are in museums.

The lesson for founders is that technical leadership and business leadership are different skills. Cray was a brilliant engineer, but he needed a John Rollwagen to run the company. When he tried to run his own company, he struggled. Cray Research succeeded because Cray had a CEO who could manage the business while Cray focused on the technology. Cray Computer Corporation failed because Cray did not have that support.

The Counter-Evidence: When the Obsession Was the Problem

The tidy version of the Cray story casts him as the pure engineer whose vision was betrayed by managers. The record is messier. Cray’s own technical bets were not uniformly right. He was skeptical of massively parallel processing while startups like Thinking Machines and Intel were building machines with thousands of microprocessors, and the industry eventually moved in that direction. The Cray-2, his own design, was the fastest computer in the world and still sold only about 25 units at $17 million apiece. Speed, the thing he optimized for, did not by itself produce a durable business.

The gallium arsenide gamble is the clearest case. Cray promised the Cray-3 would be 12 times faster than the Cray-2, and he chose a material that promised higher speeds than silicon but proved difficult to manufacture. The machine was delayed repeatedly, released in 1993, and sold exactly one unit. Cray Computer Corporation filed for bankruptcy in 1995. This was not a case of managers failing to understand an engineer. It was a case of an engineer choosing a harder path than the market could wait for, and the market declining to wait.

There is a further caveat about the handshake-and-resignation framing. Cray Research went public in 1976, the same year the Cray-1 shipped, and its stock price soared. Whatever the culture inside the lab, the company was financed and valued like a public corporation, and its 70 percent share in 1985 was defended by a CEO, Rollwagen, who took over in 1981. The romance of the lone engineer is real, but it was underwritten by ordinary capital markets and an ordinary management hierarchy. Strip those away, as Cray Computer Corporation did, and the obsession had nothing to catch it.

The Cray Legacy in Modern Computing

Cray’s influence is still felt in modern computing. Vector processing, which Cray pioneered, is used in GPUs and AI accelerators. The Cray-1’s circular design was a precursor to modern cooling systems. Cray’s emphasis on speed over cost is still a guiding principle for supercomputer designers. The world’s fastest supercomputers today, like Frontier and Fugaku, are built on the same principles that Cray pioneered: parallelism, vector processing, and advanced cooling.

Cray Research’s business model also influenced the industry. The company sold a few, very expensive machines to a small number of customers. This model is still used by companies like Nvidia, which sells high-end GPUs to data centers. The supercomputer market is still small and cyclical, but it is essential for scientific research and national security.

Seymour Cray’s story is a reminder that the most important innovations often come from individuals who are willing to ignore conventional wisdom. Cray did not care about the market; he cared about the machine. That obsession led to the creation of an industry.

What This Means for the Reader

If you run a technical company, the Cray record offers a test rather than a slogan. The question is not whether you have a visionary engineer. It is whether the organization around that engineer can absorb a delayed machine, a customer who skips a generation, and a market that peaks at $800 million in 1987 and then declines. Cray Research could, for a while, because it had a CEO and public capital. Cray Computer Corporation could not, because it had neither in sufficient measure.

The second test is about which kind of speed you are buying. Cray optimized for the fastest machine possible regardless of cost, and that produced the Cray-1 at $8.8 million and the X-MP, which sold over 100 units by 1985. It also produced the Cray-2 at $17 million and about 25 units, and the Cray-3 at one unit. The difference between those outcomes was not the quality of the engineering. It was whether the customer’s work depended on being first, badly enough to pay for it.

For anyone building hardware or infrastructure today, the durable lesson is structural. Put the obsession in the lab and the business discipline in the front office, and make sure the two report to each other. Cray’s best years came when that arrangement held. His worst came when it did not.

Frequently asked questions

What was the first computer Seymour Cray designed? Cray designed the CDC 1604 at Control Data Corporation in 1959. It was the first commercially successful transistorized computer.

How much did the Cray-1 cost? The Cray-1 sold for $8.8 million in 1976. The first customer was Los Alamos National Laboratory.

Why did Cray Research decline? Cray Research faced competition from Japanese manufacturers and parallel processing startups. The company also invested heavily in expensive machines like the Cray-2 and Cray-3, which were difficult to sell. It was acquired by Silicon Graphics in 1996.