From 3D graphics to AI infrastructure: how NVIDIA rewired computing
NVIDIA began with a bet that personal computers would become powerful entertainment machines. More than three decades later, the company sits at the centre of accelerated computing, artificial intelligence, data-centre infrastructure, gaming graphics, robotics and industrial simulation.
Jensen Huang, Chris Malachowsky and Curtis Priem founded NVIDIA on 5 April 1993. NVIDIA’s own corporate timeline says the founders believed the PC would become a consumer platform for games and multimedia at a time when more than two dozen graphics-chip companies were already competing for the market.
As of 18 September 2026, Huang remains NVIDIA’s founder, president and chief executive, while Malachowsky remains with the company as an NVIDIA Fellow. The business they created is now better understood as an accelerated-computing platform company than simply a graphics-card maker.
This article traces how NVIDIA moved from early 3D graphics to programmable GPUs, CUDA, deep learning, RTX and full-stack AI infrastructure.
NVIDIA started in a crowded graphics market
NVIDIA’s first years were not a straight path to dominance. The company was entering a young market in which many competitors were trying to define how PCs should render 3D graphics.
Its first product, the NV1, arrived in 1995. It combined 2D and 3D graphics capabilities, but its architecture did not become the industry standard. NVIDIA then changed direction and aligned more closely with Microsoft’s Direct3D approach.
The RIVA 128, introduced in 1997, became a much more important commercial breakthrough. NVIDIA’s historical material says more than one million units shipped in its first four months, establishing the company with PC manufacturers and add-in-board partners.
That success gave NVIDIA the scale and credibility to keep pushing into increasingly sophisticated consumer graphics.
GeForce 256 gave the GPU a name
In 1999, NVIDIA introduced the GeForce 256 and marketed it as the first graphics processing unit, or GPU. The company defined the GPU around the integration of transformation, lighting, triangle setup and rendering on one processor.
The name proved far more durable than the individual product. “GPU” became a standard computing term and eventually described hardware used for workloads far beyond games.
NVIDIA also completed its initial public offering in 1999 and expanded its professional graphics business with Quadro. Gaming, workstation graphics and increasingly programmable hardware became the core of its growth.
GeForce turned NVIDIA into a major PC gaming brand
Successive GeForce generations increased performance and introduced programmable shaders, higher-resolution rendering and more advanced visual effects. NVIDIA also supplied graphics technology for Microsoft’s original Xbox and expanded into notebook GPUs.
The company strengthened its position in 2000 by acquiring assets from 3dfx, one of the most influential names in early consumer 3D acceleration.
By this stage, NVIDIA was already a major graphics company. The more important transformation, however, was still ahead: using the same parallel-computing characteristics that made GPUs good at graphics for general computation.
CUDA changed what developers could do with a GPU
NVIDIA introduced CUDA in 2006. The Compute Unified Device Architecture gave developers a programming platform for using NVIDIA GPUs for non-graphics workloads.
That decision changed the strategic value of the GPU. Researchers could use the hardware for scientific simulation, engineering, numerical analysis and other tasks that benefited from thousands of operations running in parallel.
CUDA also gave NVIDIA something more defensible than raw chip performance: a software ecosystem. Developers could write applications against NVIDIA libraries and tools, making hardware and software increasingly difficult to separate.
Over time, CUDA became central to high-performance computing and machine learning.
AlexNet connected GPU computing with the modern AI boom
In 2012, the AlexNet neural network used GPU acceleration to achieve a major breakthrough in the ImageNet image-recognition competition. NVIDIA identifies the event as one of the moments that helped ignite the modern deep-learning era.
The fit between GPUs and neural networks was important. Training large models involves enormous numbers of matrix operations, a workload that maps naturally to highly parallel processors.
NVIDIA already had both the hardware and CUDA software environment needed by researchers. As machine learning expanded into language, vision, recommendation systems and generative AI, demand for GPU computing expanded with it.
This gradually shifted NVIDIA’s centre of gravity. Gaming remained important, but data-centre computing became increasingly central to the company’s identity.
RTX brought AI and ray tracing back into consumer graphics
NVIDIA launched the Turing architecture and GeForce RTX platform in 2018. RTX introduced dedicated hardware for real-time ray tracing and AI-assisted graphics workloads.
Ray tracing models how light behaves to create more realistic reflections, shadows and global illumination. The technique had existed for decades in offline rendering, but dedicated RT hardware made it more practical in real-time games.
NVIDIA also used Tensor Cores for AI workloads such as DLSS, which applies machine learning to image reconstruction and performance enhancement in supported games.
RTX demonstrated that NVIDIA’s AI work and gaming business did not need to become separate worlds. Technologies developed for one could reinforce the other.
Mellanox pushed NVIDIA deeper into the data centre
NVIDIA completed its acquisition of Mellanox Technologies in 2020 in a transaction valued at US$7 billion.
Mellanox brought high-performance networking and interconnect technology. That became strategically important as AI training systems grew from individual accelerators into clusters containing thousands of GPUs.
Those GPUs need to exchange data extremely quickly. Networking therefore becomes part of the computing problem rather than simply an external connection between servers.
By adding networking to GPUs and software, NVIDIA moved closer to supplying complete AI infrastructure rather than individual chips.
NVIDIA became a systems and platform company
Modern NVIDIA spans far more than GeForce. Its portfolio includes data-centre accelerators, CPUs, networking, systems, software libraries, AI models, robotics platforms, professional graphics and simulation technology.
DGX systems package NVIDIA computing and networking for AI workloads. Grace CPUs extend the company into server processors. Omniverse combines graphics, simulation and digital-twin technology for industrial applications. DRIVE targets automotive and autonomous systems, while Jetson serves robotics and edge computing.
The common thread is accelerated computing: moving workloads away from general-purpose CPU execution when specialised parallel hardware can perform them more efficiently.
Jensen Huang has led NVIDIA since the beginning
Leadership continuity is unusual in a technology industry defined by constant executive change. Jensen Huang has served as president and chief executive since NVIDIA’s founding in 1993.
Before NVIDIA, Huang worked at LSI Logic and Advanced Micro Devices. Chris Malachowsky also brought semiconductor experience and remains part of NVIDIA’s executive history.
That continuity allowed one architectural idea — parallel processing — to be developed across several computing eras rather than abandoned when individual markets changed.
Gaming still matters in the AI era
NVIDIA’s rise in AI infrastructure can make it easy to overlook the business that created the company’s original identity.
GeForce remains a major gaming and creator platform. RTX continues to combine conventional raster graphics, ray tracing and AI-assisted rendering. Technologies developed for AI have fed back into graphics through DLSS and other neural rendering techniques.
The relationship also works in the opposite direction. Decades of investment in programmable graphics and high-throughput parallel hardware created much of the technical foundation that later made NVIDIA useful for AI.
How NVIDIA became what it is today
NVIDIA’s history is not simply a story of a graphics-card company stumbling into artificial intelligence. The more consistent thread is that the company repeatedly expanded what its parallel processors could do.
NV1 and RIVA established the graphics business. GeForce helped define the modern GPU. CUDA opened the hardware to general computing. Deep learning created a huge new use for that architecture. Mellanox added the networking needed to scale thousands of accelerators, while RTX, Omniverse and robotics extended the platform into new markets.
As of 18 September 2026, NVIDIA describes itself around accelerated computing and AI infrastructure. The company still sells gaming GPUs, but the strategic unit is no longer an individual graphics card. It is the combination of processors, networking, systems and software.
That evolution is why future TechnologyBlog.co.za coverage of NVIDIA products should be read in the context of a much broader company than the one founded to make 3D graphics faster in 1993.
Primary sources checked include NVIDIA’s official corporate timeline and NVIDIA Newsroom executive biographies. Information reflects public material available on 18 September 2026.