The Kitchen Table Deal

In the spring of 1993, the San Jose area was transforming. The dot-com fever was still a few years away, but the region's appetite for technological ambition was already reaching a boil. Along the winding roads of Silicon Valley, venture capitalists were writing checks for ideas that ranged from promising to preposterous. Semiconductors, software, networking—everywhere you looked, engineers with borrowed dreams were betting everything on the future. It was in this crucible that three men sat down together in a booth at a Denny's restaurant near the intersection of Great Oaks Boulevard and Saratoga-Sunnyvale Road, ordered a few pitchers of cheap beer, and decided to start a company that would eventually redefine what computing meant.

Two of them had never met in person before that night.

Jensen Huang was thirty years old, a soft-spoken engineer from Taiwan who had spent the previous decade climbing the ladder at LSI Logic and Advanced Micro Devices. He carried himself with a quiet intensity, a quality that those who worked with him would come to recognize as both his greatest strength and his most disarming feature. Jensen did not raise his voice when he disagreed with someone; he simply stated, with a certainty that bordered on the physical, why the other person was wrong. It was an unsettling habit, and one that he deployed with increasing confidence as his career progressed. By 1993, he had been poached from AMD by a graphics startup called Diamond Multimedia, where he served as a director of business development—a role that gave him a broad view of an industry in the early, chaotic stages of figuring out what visual computing could become.

Chris Malachowsky was a decade younger, a twenty-year-old engineering student at the University of Florida who had dropped out to chase the gravitational pull of Silicon Valley. He arrived in California with the restless energy of someone who believed, with the unearned confidence of youth, that he already knew everything worth knowing. Malachowsky had spent time at Sun Microsystems and Sega, accumulating a reputation as a gifted hardware engineer with a talent for making things work that were not supposed to work. He had a particular gift for coaxing performance out of silicon through sheer ingenuity of design, a skill that would prove essential in the years ahead.

Curtis Priem, the third member of what would become NVIDIA's founding triumvirate, was the most understated of the three. A former engineer at IBM and Sun Microsystems, Priem brought the credibility of large-company experience and the pragmatism of someone who had seen how promising ideas died in the chasm between conception and execution. He was the one who understood, perhaps more deeply than the others, the economic realities of building a semiconductor company from scratch—the astronomical cost of chip fabrication, the ruthless economics of time-to-market, the brutal selection pressure of an industry where a single miscalculation could mean tens of millions of dollars evaporated in a year.

The Denny's meeting was not the product of careful orchestration. It was the culmination of months of loose conversations that had been building toward a conviction none of them could fully articulate yet. Jensen and Malachowsky had known each other through the tightknit graphics engineering community, a world where everyone seemed to know everyone else and where the boundaries between collaboration and competition blurred with alarming frequency. Curtis Priem had connected with Jensen through a mutual contact, and the three of them had been circling the same idea for months: that the future of computing was visual, that the rendering of three-dimensional environments in real time was not merely a feature of future systems but the defining capability around which next-generation computing would be built.

They talked through the night. The details of that conversation have become the stuff of Silicon Valley legend, embellished with each retelling, but the essential thrust is clear enough. The three men agreed on a foundational premise: that the personal computer was on the verge of an evolutionary leap, from a tool for productivity and calculation into an immersive visual medium. They believed—correctly, as it turned out—that as screen resolutions increased and software demanded richer visual experiences, the burden of rendering graphics would overwhelm the general-purpose processors of the day. A specialized chip, a dedicated graphics processing unit, would be needed to shoulder that burden. The question was not whether such a chip would exist, but who would build it first, and whether a startup could survive long enough to see its bet pay off.

What made their conviction remarkable was its audacity. In 1993, the personal computer market was dominated by Intel processors and Microsoft software. The dominant graphics standard was the VGA interface, a two-dimensional technology that had been designed for displaying text and simple images on corporate monitors. Three-dimensional graphics existed almost exclusively in the realm of expensive workstations manufactured by companies like Silicon Graphics Inc., whose $50,000 and $100,000 machines rendered the kind of photorealistic imagery that would later appear in movies like Jurassic Park. The notion that ordinary consumers sitting in front of $2,000 PCs might one day experience immersive 3D environments was, to most industry observers, a fantasy.

But the three men sitting in that Denny's booth saw something else. They saw the trajectory of Moore's Law, the observation made famous by Intel cofounder Gordon Moore in 1965 that the number of transistors on a semiconductor chip doubled approximately every two years. They saw the implications of that relentless doubling: processors would grow fast enough to support increasingly sophisticated software, and consumers would demand increasingly sophisticated visuals. The bottleneck would not be compute power or software ambition. It would be the graphics pipeline—the specialized hardware that translated mathematical descriptions of three-dimensional scenes into the stream of pixels that filled computer monitors. And that pipeline, they believed, was not being built well enough by anyone currently in the market.

The competitive landscape in early 1993 offered both encouragement and caution. The market for graphics accelerators—specialized chips designed to offload certain graphics tasks from the CPU—was fragmenting, with dozens of small companies competing for position. 3dfx Interactive had just been founded the previous year, focused on a technology called texture mapping that promised to bring richer surfaces to 3D environments. S3 Graphics, a California company, was gaining traction with its Trio64 chip, which offered accelerated 2D performance at a price point that appealed to PC OEMs. Matrox, based in Montreal, was pursuing a similar strategy. ATI Technologies, operating from the suburban sprawl north of Toronto, was rapidly building its presence in the OEM market with increasingly capable graphics solutions.

Each of these competitors was solving a specific, immediate problem: making existing software run faster and look better on the growing installed base of Windows-based PCs. The solutions were incremental, evolutionary. No one was thinking about the problem from first principles—about what a graphics chip designed from the ground up for real-time 3D rendering might look like, or what it might enable. The engineers at 3dfx came closest with their Voodoo Architecture, which would arrive in 1996 and redefine consumer gaming, but even they were optimizing for the game titles of the moment rather than the broader vision of visual computing as a platform.

NVIDIA's founders chose a different path. Rather than chasing the existing market, they decided to build for the market they believed was coming—three to five years out. Their chip, which would eventually carry