Founding and Early Dominance in the Graphics Processing Market

In the winter of 1993, three engineers sat in a Denny’s diner in Santa Clara, California, sketching out a business plan on a napkin: Jensen Huang, a 30-year-old former chip design lead at LSI Logic; Chris Malachowsky, a 36-year-old Sun Microsystems engineer with deep expertise in graphics system architecture; and Curtis Priem, a 32-year-old IBM researcher who had previously designed the graphics accelerator for IBM’s groundbreaking RS/6000 workstation. The name they chose for their startup, NVIDIA, was derived from the Latin word *invidia*, meaning “envy” — a deliberate nod to their ambition to build products so desirable that competitors would covet their capabilities. At the time, the personal computer graphics market was dominated by 2D display controllers from vendors such as S3 and ATI, with 3D graphics still a niche, software-rendered capability reserved for high-end workstations from SGI and IBM that cost tens of thousands of dollars. The three founders identified an unfilled gap: the rapidly growing PC gaming market had no affordable, hardware-accelerated 3D graphics solution, a gap they believed could be addressed with purpose-built graphics processing units (GPUs) that offloaded 3D rendering workloads from the central processing unit (CPU). The first seed round of funding came from Sequoia Capital and Sutter Hill Ventures, which invested $2 million in exchange for 30% of the company, giving the fledgling team enough runway to begin developing their first product, the NV1.

The NV1, launched in 1995, was a bold but commercially flawed product. Unlike the industry’s emerging standard of triangle-based rasterization, the NV1 used a proprietary quadratic surface rendering technology that promised smoother 3D surfaces but was incompatible with the DirectX API Microsoft had launched the same year to standardize PC game development. The chip also integrated 2D graphics and audio functionality, a feature that added cost but provided little value to PC gamers who already had separate sound cards. Only 47,000 units of the NV1 were sold, and by late 1995, NVIDIA’s bank account held less than $500,000, enough to cover only three months of operating expenses. Huang was forced to lay off half of the company’s 40 employees, and the team narrowly avoided bankruptcy after securing a $5 million bridge loan from existing investors. This near-death experience forced the leadership to make a critical strategic pivot: abandon the quadratic surface rendering roadmap, fully commit to the triangle rasterization standard supported by DirectX, and focus exclusively on 3D graphics performance rather than feature integration.

The turnaround came in 1997 with the launch of the RIVA 128, a $149 chip that delivered 1 million triangles per second of rasterization performance — double the throughput of ATI’s contemporary Rage II, the market leader at the time. Unlike ATI’s part, the RIVA 128 integrated 2D and 3D functionality on a single chip, eliminating the need for users to purchase a separate 2D display controller. Huang made a deliberate bet on the DirectX ecosystem, working closely with Microsoft to ensure the RIVA 128 was a launch partner for DirectX 5, the first version of the API to support hardware-accelerated transform and lighting (T&L) operations. This partnership gave NVIDIA a critical first-mover advantage in the Windows gaming ecosystem, as game developers optimized their titles for DirectX-compatible hardware first. NVIDIA also built a robust channel partner ecosystem, working with add-in board (AIB) vendors such as Diamond Multimedia, ELSA, and ASUS to build reference designs for RIVA 128-based graphics cards, and securing pre-installation deals with major PC OEMs including Dell, HP, and Compaq. By the fourth quarter of 1997, NVIDIA had turned profitable for the first time, and its global GPU market share rose from less than 2% in 1996 to 10% in 1998, with annual revenue reaching $160 million that year. The success of the RIVA 128 was amplified by the release of *Quake II* in 1997, which was one of the first games to fully leverage DirectX 5 and hardware T&L, and became a de facto benchmark for 3D graphics performance. NVIDIA’s close collaboration with id Software, the developer of *Quake II*, to optimize the game for RIVA 128 hardware created a viral marketing effect: PC gamers seeking the best *Quake II* experience chose NVIDIA cards, driving sales of over 3 million RIVA 128 units in 1998.

The company cemented its lead in the high-end gaming market with the 1998 launch of the RIVA TNT (short for “TwiN Texel”), the first consumer GPU to support 32-bit color depth and dual-texture rendering, enabling more realistic lighting and texture effects in games. The RIVA TNT delivered 50% higher 3D performance than the RIVA 128 at a price point of just $199, and its support for OpenGL, the open standard for 3D graphics used in professional applications, allowed NVIDIA to begin encroaching on the professional visualization market that had previously been dominated by SGI and 3dfx. By 1999, NVIDIA’s market share had risen to 25%, making it the second-largest GPU vendor behind ATI, but still well ahead of 3dfx, whose Glide API-based Voodoo2 graphics card required users to pair a separate 2D display controller, adding cost and complexity for end users.

The watershed moment for NVIDIA’s long-term dominance came in August 1999, with the launch of the GeForce 256. At the product’s launch event, Huang made a now-iconic declaration: “We are not selling a graphics card. We are selling a graphics processing unit, a standalone processor dedicated to 3D graphics.” This was the first time the term “GPU” was used to describe a consumer graphics product, a deliberate branding move to distinguish NVIDIA’s hardware from the generic “graphics accelerator” products sold by competitors. The GeForce 256 was the first single-chip GPU to integrate all core 3D rendering functions — vertex processing, geometry transformation, rasterization, texture mapping, and pixel shading — on a single die, eliminating the need for the CPU to handle any graphics-related calculations. Its geometry processing performance was 10 times faster than the fastest x86 CPUs of the era, and its 120 MHz core clock and 4.8 GB/s memory bandwidth delivered 2.5x higher 3D performance than ATI’s Rage 128 Pro at a comparable price point of $199. The GeForce 256 was also the launch partner for DirectX 7, which introduced hardware-accelerated vertex and pixel shaders, enabling game developers to create far more realistic visual effects such as dynamic lighting, reflective surfaces, and realistic water simulation. The launch of *Half-Life* later that year, which was optimized for the GeForce 256 and became one of the best-selling PC games of all time, drove demand for the chip to unprecedented levels: NVIDIA sold 8 million GeForce 256 units in 1999, more than double its total GPU sales in 1998, and annual revenue surged to $270 million, a 69% year-over-year increase.

NVIDIA’s early dominance was not built on hardware alone, but on a deliberate ecosystem strategy that created high switching costs for both developers and end users. The company invested heavily in driver optimization, releasing monthly driver updates that improved performance for new games and fixed bugs for existing titles, a practice that was rare among GPU vendors at the time. This “NVIDIA drivers make old cards run new games” reputation built deep trust with the gaming community, and the phrase “NVIDIA cards age well” became a common refrain among PC enthusiasts. The company also formalized its partnership with Microsoft, becoming a founding member of the DirectX advisory board and ensuring that all future GeForce products were launch partners for new versions of the DirectX API. For professional users, NVIDIA launched the Quadro line of GPUs in 2000, based on the same silicon as the GeForce 2 but with optimized OpenGL drivers and certification fr