This is the story of a European supercomputer that never quite happened. It's a tale of grand vision, high ambition and, ultimately, failure. It's also a story that has a lot to teach European entrepreneurs, and those who wonder why there is no European equivalent of Silicon Valley.
The Advanced Computer Research Institute - or ACRI, as it was commonly known - was created in the French city of Lyon on November 2, 1989. Its founder, Jacques Stern, is one of Europe's most accomplished entrepreneurs. In 1964, he left a safe career as a French government engineer to start up SESA, Europe's first systems engineering firm. By 1982, Stern had built SESA into a $150 million-a-year company employing 1,700 people, and French President François Mitterrand - who was reportedly close friends with Stern - lured Stern away to become chairman and CEO of Groupe Bull, France's giant, government-controlled computer company. Seven years later, Stern was ready for a new challenge and a new start-up: ACRI.
The market niche that Stern wanted to tap lay in scientific computing between supercomputers - like the Cray 3 which cost about $10 million - and high-speed engineering workstations - like those from Sun and Silicon Graphics which cost about $100,000. Scientists and engineers, Stern quite rightly believed, would be only too happy to pay $1 million or so for a machine capable of tackling CAD/CAM, real-time, discrete simulation and other scientific data processing operations. The ACRI 1, as his machine was to be called, would work as either a stand-alone computer or as a server supporting high-speed workstations. Best of all, thanks to new RISC microprocessors becoming available off-the-shelf, the ACRI machine could achieve two to four times what Cray could on some common computing tasks. Over 10 years, ACRI hoped to sell 300 to 400 machines, generating perhaps $500 million in revenues.
But grasping market opportunity was only part of Stern's vision. He also wanted to transform the European computer industry. By developing a high-performance computing industry in Europe, he would nurture the growth of related hardware, component, and software industries. And he would create a new computing culture of risk taking and entrepreneurship. "Most people in Europe don't understand that very big companies can't innovate in new markets because it's just too risky," Stern explains.
Unlike the stereotypical American entrepreneur, who cares only about money and machines, Stern is clearly a social capitalist. He believes in industries and jobs, not just in money. One senior American staffer, charmed by Stern into moving to France, recalls that his first meeting with Stern "almost had a hypnotic effect on me. After two hours, I was almost in tears. He'd convinced me of this great future for Europe and that he wanted me to be part of it."
With hindsight, it is tempting to argue that Stern's vision and dedication were also a key part of his and ACRI's downfall for they helped to create two fatal flaws in the venture. While ACRI's management focused on building a company, the product sometimes came to seem almost secondary.
Meanwhile, Stern himself worked in Paris most of the week, in the realms of high politics and high finance, leaving the engineering and day-to-day management to others far less experienced in complex engineering projects.
Stern's own final judgement on the project is noble, ironic, and telling. "I am ultimately responsible for this failure and I did not fully realise all the difficulties that we would face," Stern says. "But," he adds, "there is no political or financial will in Europe" to support big information technology ventures.
A lot of jealous US entrepreneurs would give an arm and a leg to experience a similar failure of will. Over the five and a half years of its life, Stern Computing, ACRI's parent company, spent more than $120 million, nearly all of it in French or European Union grants.
In the end, no product ever came to market. By contrast Convex, a US supercomputer-maker which would have been a rival to ACRI.- if ACRI's product had ever reached the market - brought its first machine to market in three years for just $35 million.
Today, with ACRI in bankruptcy, some of the underlying reasons for its failure are already being shrouded by the second-guessing that comes with hindsight, and the mutual recriminations of disappointed executives. But given the record of its subsidies and its spending "a failure of political and financial will" seems to have been the least of ACRI's problems.
"We should have been able to bring three machines to market with the money we had," says one ex-ACRI engineer. "The waste and incompetence were unbelievable. The top managers just wouldn't listen to the engineers. They acted like they knew what they were doing but they didn't have a clue."
Unlike many successful US start-ups, ACRI's management did not appoint a product architect with real executive power. This just isn't the way things are done in France, especially if the architect isn't French and isn't a graduate of one of the handful of grandes écoles that educate and define France's elite. Over the five and a half years of its existence, ACRI had three different product architects. Product and engineering focus dissipated. But spending did not slow.
"They spent way too much money too early," remembers Knut Lonsted, a former ACRI project manager. "Their idea was to hire everybody quickly and push like a bulldozer." Jacques Stern was convinced that it was necessary to staff up quickly to get the machine to market fast. He calculated that there was a window of only four years, 1994-1998, in which the first generation ACRI machines would outperform its rivals.
But ACRI was also encouraged to add staff by the requirements of ESPRIT funding. ESPRIT, the European Union's main programme of subsidy for R&D in information technology, sets staffing milestones for grant recipients; companies must hire some fixed number of people to qualify for further aid. This seems a perverse incentive for cash-hungry start-ups, which should be trying to conserve capital during development years, when no revenues roll in.
ACRI, for its part, did not seem to need much encouragement in hiring. The company did hire some French engineers and managers, but due to a deep lack of home-grown expertise in the areas of supercomputer architecture, compiler technology, high-speed circuit boards, mechanics, and cooling systems, it also hired a mélange of American, British, Israeli and Scandinavian personnel. ACRI handed out salaries of up to $200,000 a year to attract the "right talent."
Employment peaked at around 160 people in 1993-94. A director of operations was recruited years before there were any operations to manage. ACRI managers hired a quality assurance team and then, when it realized there was not yet a need, it broke the group up and its members were given other tasks to perform - not all of them very meaningful. Cost-cutting does not seem to have been high on the managerial agenda.
One senior American mathematician who came to ACRI to help design the product remembers attending a technical conference on behalf of the firm in the United States, where he was to deliver two papers. But ACRI sent five Frenchmen to accompany him, all flown out from France on business class and all staying in luxury hotels. None of them had any papers to deliver and at that time, in the early 1990s, there wouldn't be a product for sale for at least three years, so there was no need to shmooze potential customers. The American estimates that ACRI blew $50,000 on that jaunt alone.
ACRI occupied prestigious offices in a tower in central Lyon, and equipped them well. An insider claims that the company spent an estimated $1 to $1.5 million on a "quickturn box" to perform high-speed simulations on the machine design and then didn't have the opportunity to use it for two years, at which point its lease expired. ACRI formed a special-ised 20-person compiler company in Belgium when a team of two or three based in Lyon might have sufficed. It opened a company in Boston to recruit more American personnel.
To keep up with increased spending, ACRI executives had to plead for more money from Paris and Brussels every six months. Miraculously, they kept getting it, in large part because of Stern's influence and friends within the French government. "They seemed to think that the money would just keep coming in," remembers one ex-employee. "Jacques Stern was used to that. For seven years as head of Bull he'd just call the president and ask for more."
While Stern worked most of the week in Paris, day-to-day management of ACRI in Lyons fell to Tor Bloch, the chief operating officer. Bloch had gained his supercomputer expertise running the computer centre at CERN (Centre for European Nuclear Research). He had also worked for Stern in a marketing role at Groupe Bull. But he had no engineering management experience and had never worked for a technology start-up.
"He [Bloch] was unsuited for the job at hand," argues Adrian Wise, an experi-enced supercomputer sales and marketing man who was appointed vice president of sales and marketing at Stern Computing in 1993. "For a year and a half, there was never a management review of the entire project. Board meetings just took place at Stern's level and addressed how to get more money."
Stern, however, knew Bloch well and trusted him implicitly. "It's extraordinary," recalls Knut Longsted. "Jacques works with you for a while and then, when he feels sure about you, he gives you every bit of his trust and confidence."
Few other ACRI managers seem to have been so trusting. ACRI was extremely hierarchical, unlike Silicon Valley management structures where everybody is their own secretary and senior staff even take turns cleaning the toilets to save money. "There were six layers of management at ACRI," claims Peter Bird, senior architect on the project from 1992-1993. "Ford Motor Company only has five."
This hierarchical structure made it hard for Tor Bloch and other senior executives to integrate and manage all the project's diverse elements. It also meant that a team of French and European managers with little engineering experience held the top posts while the experienced nonFrench architects and engineers reporting to them often had little input into making crucial technical decisions.
Peter Bird remembers that "I had no more authority than some French managers who had no knowledge and had only thought about things for five minutes. I got very depressed. The French board's off-the-cuff remarks pulled greater weight than my ten years of research."
Bloch shared Bird's frustrations with ACRI's engineering management. "It was not so easy to link hardware and software development," Bloch said. "There were too many changes of architect ... We didn't run like a US start-up. But many of the Americans asked for big salaries and wanted to travel."
Bloch traces the lack of direction back to Hanan Potash, chief architect and head of development from 1989-1990. Bloch claims that Potash revelled in creating management hierarchy. "The architecture didn't gel," Bloch says, "it never came together."
Jacques Stern, by contrast, argues that it is "unfair to blame Hanan Potash. His impact on the architecture was not very great." Potash himself, in a phone interview, still remembers his days in Lyon with bitterness. He claims Bloch gave everyone "a very unclear agenda. And when people didn't live up to expecta-tions, he fired them." More significantly, Potash questions Bloch's fundamental understanding of computer design and architecture.
ACRI's original engineering strategy had been to stick closely to off-the-shelf components. In theory, this would give ACRI the best of two worlds. It would gain supercomputing speed by arranging those components in clever ways - and com-plementing them with quick software. It would also save money and development time by using tried-and-tested parts. In practice, however, the lure of flashy technology proved too great for ACRI to resist.
The first controversial decision concerned the central processor for ACRI's machine. Instead of chips made from silicon, ACRI decided to switch to gallium arsenide technology for the machine's central processor. Gallium arsenide makes very fast chips, and they are available off-the-shelf. But gallium arsenide chips are fragile and prone to overheating, which creates a variety of design and cooling problems.
The decision to go with gallium arsenide perplexed many of ACRI's engineers. "Aside from Convex, every company using gallium arsenide has gone bust," explains one. "It's brittle, it breaks, it requires toxic chemicals for its fabrication. Many of us warned against gallium arsenide but it made no difference." The decision also raised concerns among the project's backers in the French government. In an urgent communique to Bloch from the French government's directeur de l'informatique (directorate of computing) dated 18 July 1993, an official wondered, "why must ACRI take such a risk for a machine which in three years will be in competition with CIMOS [silicon] technology?"
But, in addition to switching to gallium arsenide, ACRI also decided to package the chips with a fancy new noise-reducing packaging technique called Atab. According to Adrian Wise, a senior ACRI executive "found out about this at a conference and just decided to use it." Vitesse, which was supplying ACRI's gallium-arsenide chips, couldn't do the special packaging. So the chips were shipped to an IBM plant for packaging and then onto Lyon. Over an eight-month period in 1994 as the chips arrived at ACRI, every single one of them was broken and unusable.
Other engineering headaches concerned the way in which the chips were connected. "For the machine to work," recalls ex-ACRI software engineer Kevin Kissell, "billions of bytes of information had to flow through its core every second. We needed a great amount of data bandwidth and so the basic connectors joining the boards became critical."
But after three architects, a bevy of designers and developers and some ill-advised technological chopping and changing, the machine's core had become extremely complicated. "The CPU board had several thousand components..." wrote former ACRI architect Robert Hyerle in a memoir of the project. "Boards were 26-layer, super-duper stuff with all sorts of constraints. As we added con-straints, they just got more expensive and took longer to manufacture." Hyerle went on, "The board designer mentioned to me that he had never worked on a board like this, where he had no idea what it did. He was just translating netlists and driving CAD tools."
As the engineering problems multi-plied, delivery dates slipped. The company couldn't even manage to mount a prototype demonstration. Bloch and Stern went back to the well of public funding again in early 1995 and failed. An earlier whirlwind tour of the US, Europe, and Japan had failed to raise any private venture capital. In the first quarter of 1995, the money ran out, some employee paychecks stopped arriving and Stern Computing filed for bankruptcy.
The liquidation turned out to be messy, as liquidations often are. Two offers were made to buy what was left of ACRI, one for FFr3 million by ACRI's vice president of marketing and sales, Adrian Wise, and another for around FFr1 million from France's Compagnie des Signaux et des Equipements Electroniques, whose president, Yazid Sabeg, was also a former director of Stern Computing.
The liquidator accepted the lower offer. Compagnie des Signeaux offered jobs to around 40 former ACRI staffers in hopes of salvaging some ACRI technology to apply to its core business of high-speed signal processing. Tor Bloch joined Compagnie des Signeaux to work on marketing, but left within seven months.
Many of ACRI's creditors - including one German supplier, one British supplier and universities in Britain, Spain, and Crete - are still waiting to be paid. Bloch says of the creditors, "It's a sad part of the story, but they got a lot of projects and money out of this along the way."
For Bloch, some of the big lessons learnt from ACRI concern, not surprisingly, financial risk. "This was Jacques Stern's project - he was CEO and president. I wouldn't have taken such big risks on the financing end as he did. I became largely an implementor."
But Stern, in turn, was encouraged both by the strength of his own vision and by the incentives built into the programmes which subsidised his efforts - in particular those provided by the European Union's ESPRIT programme of subsidies for IT projects. Not only did ESPRIT require ACRI to hire minimum numbers of people to be eligible for grants, it also required the company to take on development partners in other European countries - hence ACRI's partners in Britain and Germany. Both requirements seem to be motivated largely by the politics of jealousy. With unemployment high on the political agenda everywhere in Europe, no government wants its part-ners to get more job-creating subsidies than it can grab for itself.
But government intervention can cripple entrepreneurs. "These kinds of partnerships make more sense for large companies than for start-ups," reflects Jacques Stern. And here, perhaps, lies the real lesson of ACRI: Entrepreneurs do not create jobs - at least not on purpose. They create products, and from these they make their fortunes. Jobs are a by-product. To confuse creating jobs with being entreprenerial, as so many European institutions seem to do, is to ensure that the region will have neither.
Richard Evans is a London-based investigative journalist, who has written about both high-tech and low-tech.