NVIDIA Corporation
Semiconductors and artificial intelligence infrastructureFounded 1993- Updated June 2026
Reviewed by Swet Parvadiya
Last reviewed: June 2026
CorpDigest
NVIDIA Corporation
Company History
Founded 1993 in Santa Clara, California
Last reviewed: 2026-06-03 · By Swet Parvadiya
A Denny's restaurant in San Jose, February 1993: Jensen Huang, Chris Malachowsky, and Curtis Priem sketch the outline of what will become NVIDIA on a paper tablecloth. All three had come from Sun Microsystems and LSI Logic. They saw an opportunity in 3D graphics for PCs — the hardware didn't exist yet in a form that consumer computers could afford, and the software that would drive demand for it (video games, CAD software) was beginning to appear. Their founding bet was that graphics processing would become a specialized, dedicated hardware function rather than a general-purpose CPU task.
The early years were difficult. NVIDIA's first three product generations were commercial failures or near-failures, and the company came close to running out of capital before the NV10 — the GeForce 256, released in 1999 — established the GPU as a commercially viable category. That product introduced hardware transform and lighting, shifting visual processing from the CPU to dedicated graphics hardware in a way that game developers immediately adopted. The Xbox gaming console, which used an NVIDIA graphics processor, and the subsequent consumer graphics card market built the revenue base that gave NVIDIA time to pursue longer-term bets.
The 2020 acquisition of Mellanox for $6.9 billion added high-speed networking hardware to NVIDIA's portfolio, and what seemed like an adjacent but peripheral expansion turned out to be essential to the data center architecture that AI clusters require. By 2022, when AI training workloads began consuming enormous quantities of H100 GPUs, NVIDIA's Mellanox-derived InfiniBand and Ethernet networking allowed it to sell not just the GPUs but the complete cluster interconnect fabric — a move that dramatically expanded the revenue opportunity per data center deployment.
Jensen Huang co-founded NVIDIA in 1993 and has remained CEO through every major era of the company: early PC graphics, GeForce, CUDA, data-center acceleration, autonomous systems, and generative AI infrastructure. He helped recover from the NV1 misstep, pushed the company toward the GeForce 256, and later championed CUDA as a platform investment when GPUs were still associated mainly with games. Huang's leadership style is product-centered and theatrical, but the substance is architectural: he repeatedly tries to move NVIDIA up the stack from chips to systems, software, and developer ecosystems. After the 2020 Mellanox acquisition, his AI factory vision became commercially decisive because networking made large GPU clusters usable across large volumes. By FY2026, NVIDIA's $215.9 billion revenue base reflected the long arc of that strategy. Huang's lasting influence is the belief that a chip company can become an infrastructure platform if it owns the developer workflow, earns customer trust, and keeps extending the roadmap before rivals finish copying the last generation.
Chris Malachowsky co-founded NVIDIA in 1993 and became a central figure in the company's technical foundation. His contribution was especially important during the period when NVIDIA had to recover from early product misalignment and compete against companies such as 3dfx, ATI, S3, and Matrox. He helped shape the engineering discipline behind NVIDIA's graphics processors and supported the company's transition from consumer graphics into broader accelerated computing. Over time, Malachowsky moved into senior technology and advisory roles, remaining part of NVIDIA's long-term technical culture rather than becoming a public-facing CEO figure. His influence shows up in the company's engineering bias: product cadence, platform thinking, and respect for developer ecosystems. He also helped maintain continuity between the original graphics mission and the later CUDA/data-center strategy. While Jensen Huang became the public strategist, Malachowsky helped make the company technically credible enough for that strategy to survive multiple industry cycles. His legacy is quiet but durable: NVIDIA still behaves like engineering execution is a strategic asset.
Curtis Priem co-founded NVIDIA in 1993 and served as one of the company's original architects. His role was especially important in the early years, when NVIDIA was trying to define a path through a crowded graphics market and recover from architectural choices that did not align with emerging standards. Priem helped establish the company's technical ambition around specialized processors for visual computing. He later stepped back from day-to-day executive prominence, but his influence remained in NVIDIA's willingness to pursue hard architecture problems rather than commodity graphics. Priem's lasting contribution is the idea that graphics chips could become a serious computing platform. That belief looked narrow in the 1990s, when the market was mostly gaming and multimedia, but it became central to the company's later move into CUDA, scientific computing, and AI infrastructure. His founder legacy is visible whenever NVIDIA treats rendering, simulation, and AI as variations of parallel computation. He also represents the less-public technical founding layer behind NVIDIA's later market mythology.
NVIDIA acquired Mellanox to add high-speed networking, InfiniBand, Ethernet, switches, adapters, and data-center interconnect expertise to its GPU business. The goal was to control more of the AI and high-performance computing cluster stack.
NVIDIA acquired core assets of 3dfx Interactive, including patents and engineering talent, after the Voodoo graphics pioneer collapsed. The purpose was to remove a rival, resolve litigation, and strengthen NVIDIA's graphics intellectual property.
NVIDIA acquired AGEIA to bring PhysX physics simulation technology into its GPU ecosystem. The goal was to make gaming and simulation more realistic while giving developers another reason to improve for NVIDIA hardware.
NVIDIA acquired Icera to strengthen its mobile modem and baseband capabilities as it tried to compete in smartphone and tablet processors through Tegra. The deal was meant to fill a connectivity gap against Qualcomm and integrated mobile rivals.
NVIDIA acquired Cumulus Networks to add open networking software to its data-center portfolio after the Mellanox deal. The goal was to strengthen software-defined networking for modern cloud and AI infrastructure.
NVIDIA acquired Excelero to improve high-performance block storage technology for large-scale enterprise and AI workloads. The goal was to reduce bottlenecks around data movement and storage access in accelerated computing environments.
NVIDIA acquired Run:ai to improve orchestration, scheduling, and utilization of AI infrastructure. The software helps teams allocate expensive GPU resources more efficiently across users, models, and workloads.
NVIDIA was founded in April 1993 by Jensen Huang, Chris Malachowsky, and Curtis Priem at a Denny's diner on East Capitol Expressway in San Jose, California — a meeting-spot the founders chose because it was cheap, open late, and within driving distance of all three. Huang had been director of CoreWare at LSI Logic. Malachowsky and Priem had been engineers at Sun Microsystems, where they had worked on graphics hardware for the GX series. Their thesis was that PC graphics would converge on dedicated 3D accelerators, replacing the software rasterization and 2D framebuffers that dominated at the time. Initial seed funding of roughly $20 million came from Sequoia Capital and Sutter Hill Ventures, with Don Valentine of Sequoia famously telling Huang on a single phone call that he would invest if Huang made the company successful — and that he would 'kill' him if he didn't. The company was incorporated as NVIDIA Corp on April 5, 1993, and initially headquartered in Sunnyvale before moving to its current Santa Clara campus.
NVIDIA's first product, the NV1 in 1995, was a commercial failure. It used quadratic texture mapping rather than triangular polygons, was incompatible with the emerging Microsoft Direct3D triangle-based standard, and combined graphics with audio and game-pad ports in a way the market rejected. By 1996 NVIDIA was nearly bankrupt and laid off roughly half its staff. The recovery began with NV2, a Sega Saturn project that was canceled, and culminated in RIVA 128 in 1997 — a triangle-based 3D accelerator that achieved competitive performance at an aggressive price. The follow-up RIVA TNT in 1998 established NVIDIA as a top-tier graphics vendor competing with 3dfx Voodoo. The 1999 launch of GeForce 256, marketed as 'the world's first GPU,' integrated transform-and-lighting on-chip and pulled NVIDIA decisively ahead. The company went public on the NASDAQ in January 1999 at $12 per share and acquired the assets of bankrupt rival 3dfx Interactive in late 2000, consolidating the consumer-3D market and inheriting the engineering talent that had built the original Voodoo cards.
CUDA — Compute Unified Device Architecture — launched in November 2006 alongside the GeForce 8800 GTX, the first consumer GPU built with general-purpose programmability in mind. CUDA exposed the GPU's parallel cores to programmers through a C-like API, letting researchers write non-graphics workloads — fluid simulation, molecular dynamics, image processing, and eventually neural networks — that ran on commodity GeForce hardware. At the time the bet looked speculative: GPUs were a $4 billion gaming market, and the incremental die area dedicated to CUDA features cost gaming performance for benefits no one was monetizing. Huang nonetheless committed billions of R&D dollars over the following decade to making every NVIDIA GPU CUDA-capable and to building the software stack — cuDNN, TensorRT, NCCL, RAPIDS — that AI researchers would adopt. The 2012 AlexNet result, in which a convolutional neural network trained on two GeForce GTX 580 GPUs won the ImageNet competition, validated the bet. By the early 2020s essentially every meaningful AI training and inference workload ran on CUDA, making the software stack the most-cited moat in the company's competitive positioning.
NVIDIA's revenue mix shifted dramatically across roughly four years. In FY2020 (calendar 2019), the gaming segment generated approximately 50% of revenue against a data-center segment that was a fraction the size. By FY2024, data center had become the largest segment by far, generating $47.5 billion of $60.9 billion in total revenue, driven by hyperscaler purchases of A100 and then H100 GPUs for large-language-model training and inference. By FY2025, data-center revenue exceeded $115 billion of $130.5 billion total, and gaming had become a single-digit-percentage line item. The transformation reflected the November 2022 launch of ChatGPT, which triggered an industry-wide build-out of AI compute capacity, and the H100's status as the dominant chip for training frontier models. NVIDIA's market capitalization rose from roughly $300 billion in late 2022 to peak around $5.7 trillion in 2025 — briefly the largest company in the world by market value. The Blackwell B100/B200 generation launched in 2024 and the GB200 NVL72 rack-scale system extended the franchise into 2025.
Five decisions shaped NVIDIA's trajectory more than any others. First, the 1997 pivot from quadratic to triangle-based rendering after NV1 failed — a pivot that nearly bankrupted the company but produced RIVA 128. Second, the 1999 introduction of the term GPU and the integration of transform-and-lighting in GeForce 256, which redefined the category. Third, the 2006 launch of CUDA, made over the objections of investors who did not see general-purpose computing as a graphics-company business. Fourth, the 2017 launch of the Volta V100 with Tensor Cores specifically optimized for matrix multiplication used in deep learning, signaling that NVIDIA recognized AI as a primary workload before competitors did. Fifth, the 2020 acquisition of Mellanox for $6.9 billion, which gave NVIDIA the InfiniBand networking that turns individual GPUs into multi-thousand-node training clusters. The aborted 2020 Arm acquisition would have been a sixth, but regulatory opposition from the UK, EU, and US blocked the deal in February 2022, and NVIDIA paid SoftBank a $1.25 billion break fee.