This article is the tenth chapter of The Entrepreneur's Journey, a collection of stories about startup companies and the entrepreneurs who built them. To continue reading about key startup themes and lessons learned, check out the entire series here in The Seraf Compass, or purchase the book on Amazon in paperback or Kindle format.
Time and money have an interesting way of changing the landscape we live in. This is especially true in the world of computers. In 1965, while he was CEO of Intel, Gordon Moore published a paper which predicted the number of transistors on a computer chip would double every two years for at least the next ten years.
Moore’s prediction held true and continued, albeit at a slightly reduced pace, till around 2012. His conclusions became known as Moore’s Law, and provided a roadmap for the rapid growth in computing power. With more transistors on each chip, the speed of computing increased commensurately. As difficult as it might be to believe, the main processor in an Apple iPhone from 2020 is millions of times faster than the CPU processor in an IBM PC from the early 1990s.
Starting with my early teen years, I was a first hand witness to this rapid evolution in technology. My connection with computers began at the middle school I attended in Baltimore, Maryland. The school had a Hewlett Packard computer we could program using punch cards. During math class, I would write simple programs in BASIC, and then place my stack of cards into a basket in the back of the classroom. The next day the math teacher would hand me a printout containing a listing of my program along with the resulting output. If I had a single error in my program, I would have to fix it, resubmit my punch cards, and wait an entire day before I received a new printout. I was fascinated with computers and loved writing programs, but the inability to have immediate results frustrated me.
By the time I was in high school, personal computers hit the scene. Now, I was able to write programs and see immediate results. I was a natural at writing software. I spent as much time as I could in the computer lab whenever I had a break in my daily schedule.
In 1979, immediately prior to my freshman year in college, Brown University launched its Computer Science (CS) Department. Brown became one of the first universities to put together a comprehensive course of study in CS. Under the leadership of Andy van Dam, Brown became a pioneer in the field of computer graphics. When it came time for me to choose a major, CS was a given, and Andy influenced me to focus my research in computer graphics.
While studying at Brown, I spent two years working for Jim Head at the Planetary Geology department. As part of his research on planetary systems, Jim worked on a number of NASA interplanetary missions. My project for Jim focused on developing 2D and 3D models of the planet Venus. Working in a dark room filled with computer equipment, I would spend days at a time cranking out code.
I had access to data collected from NASA’s Pioneer Venus mission in addition to data collected from the Arecibo Observatory in Puerto Rico. Combining these datasets, I created the first three dimensional flybys of the planetary surface of Venus. Seeing Jim’s eyes light up the first time he saw one of these flybys was one of the highlights of my time in college.
In the early 1980s, real time 3D graphics required powerful computer systems costing hundreds of thousands of dollars. Only large corporations and research universities could afford such equipment. My research at Brown was made possible because I was working at a university with the resources to acquire and maintain expensive computer graphics hardware. A world with inexpensive, widely adopted 3D graphics was still over a decade away.
My early post college career allowed me to continue creating powerful computer graphics software for the avionics and pharmaceutical industries. I ventured into the world of computer hardware a few years after meeting Bill Poduska in 1986. At the time, I was working for a drug discovery software company called Polygen. Bill was one of the most prolific tech entrepreneurs in the country. With a PhD in Electrical Engineering from MIT, Bill is a brilliant engineer and a great business leader. And, he is one of the most gregarious, friendly individuals you will ever meet.
In the 1970s and 1980s, Bill co-founded two major computer companies, Prime Computer and Apollo Computer. Both companies became very successful, publicly traded companies. When I first met Bill, he was introducing his third startup, Stellar Computer. For this startup, Bill assembled a world class team of software and hardware engineers along with experts in manufacturing, finance, sales and marketing.
Stellar’s mission was to build the world’s first graphics supercomputer at an affordable price. This was a goal near and dear to my interests. Their initial product, called the GS1000, helped scientists and engineers solve challenging problems in areas as varied as weather modeling, drug discovery, nuclear simulation and engineering analysis. These are tasks that were beyond the scope of all but the most expensive computers at the time.
Bill’s enthusiasm for Stellar was infectious. After hearing him speak at a local tech event, I wanted to learn more about the company. Based on what I heard from Bill, I felt the GS1000 would be a perfect fit for the software I was developing. And so Polygen became one of Stellar’s first customers. I was one of the early engineers given the opportunity to utilize this new generation of graphics supercomputers.
A year or so after Polygen bought its first GS1000, Stellar reached out to me and asked if I might join their company to help further the development of their computers. After interviewing with Bill’s management team, I was all in. How could I turn down their offer to change the world of high performance computing?
I joined Stellar in 1989 at a time when the company had two main competitors, Silicon Graphics (SGI) and Ardent Computer. Stellar, SGI and Ardent sold workstations to scientists and engineers with prices ranging from $50,000 to well over $100,000. Our target customer had funding for such a large expense because the problems they were trying to solve had urgent, real world implications. Designing a new airplane engine or modeling a nuclear simulation must be done right or the consequences are severe. Companies like Boeing and Airbus, and national labs, including Lawrence Livermore and Argonne, needed our sophisticated computers to do their job.
Stellar, SGI and Ardent had brisk businesses because the personal computers of the time from companies such as IBM, Dell and Compaq didn’t have enough horsepower to take on these complex and demanding computing challenges. All of the PC companies used CPU chips designed by Intel and AMD to power their machines. In the late 1980s and early 1990s, those chips were quite limited in the tasks they could handle. So SGI, Ardent and Stellar all chose to build custom chips and develop proprietary software to deliver the performance their customers needed.
When Bill Poduska launched Stellar, he knew the company would face many serious challenges. The technical hurdles to move his vision from paper to reality were daunting. However, Bill’s successful track record at his earlier startups, Prime and Apollo, allowed him to recruit some of the best engineers in the world and to raise financing from top tier VCs. In short order, Bill pulled together the human and financial resources he needed to get an amazing initial product to market.
Even with an incredible amount of talent and money at their disposal, Stellar faced an uphill battle typical of situations when small hardware startups compete against large, highly-profitable companies. In the beginning, Stellar saw their main competition come from similarly sized companies in Ardent and SGI. In time, the competition came from industry giants such as Intel and IBM.
As it would turn out, a lack of perspective about Moore’s Law and a few more years of exponential growth in transistors on a computer chip led to one of the biggest mistakes made by Stellar and its competitors. They completely misunderstood the impact of massive financial resources on their market. In 1985 (the year Stellar was founded) Intel released its Intel386 chip. This chip could perform up to 4 Million Instructions Per Second (MIPS). That was more than enough speed to efficiently perform basic productivity tasks such as spreadsheets and word processing. But if you wanted to model a complex weather system, these chips would take forever, if they could do it at all. Stellar’s first CPU performed at 25 MIPS alongside a co-processor that did an additional 40 Million Floating Operations Per Second (MFLOPS). For most scientific applications, the GS1000 was 10 to 100 times faster than a PC.
Just four years later, Intel had leveraged its massive R&D organization to introduce a new chip called the 486. Weighing in at roughly the same size, it ran five times as fast at 20 MIPS. At that time Stellar introduced the GS2000 which could still perform at a rate of 25 MIPS but with double the rated MFLOPS at 80. Because of the massive resources Intel could throw at R&D, the speed differential between Stellar and the ubiquitous PC was rapidly shrinking.
Designing custom hardware is expensive and time consuming. This is especially true with extremely complex, specialized and difficult to manufacture hardware like computer CPUs. Massive resources allow tremendous leaps forward. For a small startup like Stellar, it was hard to keep pace with well-funded companies like Intel. In 1989, Intel had revenues of $3.1 billion and spent $365 million in R&D. Stellar’s revenues were less than one percent of Intel’s. And, their R&D budget was less than five percent of Intel’s.
Although it was not clear at the time, in hindsight, the end result was inevitable. It was only a matter of time before the massive resources thrown at inexpensive, commodity PCs with Intel CPUs would make them competitive with the speed of the specialty high performance graphics supercomputers from SGI, Ardent and Stellar. When Stellar launched in 1985, Bill and his engineering team were confident they could maintain a significant performance gap over the likes of Intel. But in the end, Moore’s Law and a huge R&D budget allowed Intel to close the performance gap and put companies like Stellar and Ardent out of business. This was a case where David could not defeat Goliath.
Even after merging their two companies, Stellar and Ardent were unable to compete. The combined company, called Stardent, shut down towards the end of 1991 having burned through more than $200 million in venture capital. SGI was able to hold on longer, but it too ran into serious headwinds. Ultimately, SGI went bankrupt in 2009 and had a firesale for their remaining assets.
At its core, Stellar was a computer hardware company with all of the financial and logistical challenges faced when producing a physical product. Hardware companies sell physical atoms as contrasted with the intangible bits of software products.
Hardware companies, like Stellar, have different economics and cost structures than software companies. They have longer, more complex design processes. They are less nimble and adaptable in terms of redesign if they are caught out of step. They have long supply chains. They consume capital for inventory. And their sales cycles are typically longer if pilots and capital budgeting processes are involved.
As a result, hardware companies have thinner margins which leave less room for error in their business plans. But, if you can overcome these hurdles to success, you will have built a huge moat that will keep the competition at bay. And, your investors will be very happy.
In our theme on “bits vs. atoms”, we will dive into the stories of two more hardware companies, Powerhouse Dynamics and Mobius Imaging. Both companies had an accomplished CEO with many years of hardware experience. They knew the challenges they would face before they took a leap of faith and launched their businesses. Although both companies ended up being acquired, the financial outcome for their investors was vastly different.
My experience at Stellar was an incredible learning opportunity for me. It was my first time working for a hardware company. I had a front row seat to witness how difficult it is to be successful building physical products. I learned my lesson well, and quickly returned to the software industry to continue my career selling bits.