Nvidia Timeline

Nvidia 35

1993 Nvidia's early strategy — the multimedia chip In its early months, Nvidia targets a single chip that handles 2D graphics, video playback and 3D graphics simultaneously. Nvidia partners with chip manufacturer SGS-Thomson and signs a contract with Sega to supply graphics chips for the Sega Saturn game console. The Sega contract provides crucial early revenue and engineering direction. A wrong bet on proprietary 3D geometry will nearly destroy the company three years later.
1993 Nvidia founded in a Denny's diner Jensen Huang (30), Chris Malachowsky and Curtis Priem found Nvidia at a Denny's restaurant in East San Jose on 17 April 1993. Huang, a former AMD and LSI Logic engineer, articulates their mission: to build a chip that accelerates graphics for personal computers. They incorporate with $40,000 in seed funding from friends and family, followed by $20 million from Sequoia Capital and others. The San Jose Denny's later installs a plaque commemorating the founding meeting.
1995 NV1 — first chip, first near-death Nvidia releases its first product — the NV1 multimedia chip — in 1995. It uses quadratic texture mapping rather than the triangle-based rendering (DirectX) that Microsoft and the games industry adopt as standard. The NV1 is not a commercial success. Sega cancels its chip contract. Nvidia faces bankruptcy with just 3 months of cash remaining. Sega provides a $5 million emergency payment — money that allows Nvidia to complete its next chip, the RIVA 128.
1997 RIVA 128 — Nvidia is saved Nvidia releases the RIVA 128 in August 1997 — built from scratch in just four months, adopting DirectX compatibility and triangle-based rendering. The RIVA 128 delivers the fastest 3D graphics performance available and sells one million units in four months — the fastest-selling graphics card in history at the time. It saves Nvidia from bankruptcy and establishes the company as a serious player in the PC graphics market.
1999 Nvidia IPO — goes public at $12 per share Nvidia goes public in January 1999 at $12 per share. The same year, Nvidia releases the GeForce 256 on 31 August 1999 — coining the term 'GPU' (Graphics Processing Unit). The GeForce 256 is the first chip to perform transform and lighting calculations in hardware rather than software. It delivers 60% better gaming performance than its predecessors. The GPU concept — a massively parallel processor for visual computation — will prove one of the most consequential hardware innovations of the 21st century.
1999 GeForce 256 — the first GPU Nvidia releases the GeForce 256 on 31 August 1999, coining the term GPU (Graphics Processing Unit) to describe a chip that handles geometry and lighting calculations previously performed by the CPU. The GeForce 256 delivers dramatic gaming performance improvements and establishes Nvidia's GeForce brand as the leading gaming GPU line. The GPU concept — a massively parallel processor for visual computation — will prove one of the most consequential hardware innovations of the 21st century.
2000 GeForce 2 and 3dfx defeated The GeForce 2 GTS launches in April 2000, cementing Nvidia's position as the dominant gaming graphics chip maker. Nvidia defeats rival 3dfx — previously the market leader — comprehensively; 3dfx declares bankruptcy in December 2000 and Nvidia acquires its intellectual property. Nvidia also announces the GeForce 3, the world's first GPU to support Microsoft's DirectX 8 programmable shaders — a capability that will prove crucial when GPGPU computing emerges.
2001 Nvidia wins Xbox graphics chip contract Nvidia wins the contract to supply the graphics processor for Microsoft's first Xbox game console — a landmark deal that gives Nvidia enormous revenue, engineering experience and credibility in the console market. The Xbox NV2A chip is closely related to the GeForce 3. The relationship deepens Nvidia's ties with Microsoft, which proves important when CUDA is developed and when AI computing platforms emerge.
2002 Nvidia acquires 3dfx — eliminates main rival Nvidia completes its acquisition of 3dfx Interactive's intellectual property and trademarks in April 2002, formally eliminating its only significant rival in dedicated graphics processing. ATI (later acquired by AMD) remains as a competitor, but Nvidia's technology lead is now substantial. The acquisition also gives Nvidia valuable GPU architecture patents that strengthen its competitive moat.
2004 SLI returns — two GPUs in one system Nvidia revives SLI (Scalable Link Interface) — a technology allowing two Nvidia GPUs to work together in a single PC for double the gaming performance. SLI is the first hint that GPUs can work in parallel as a computing platform. The parallel processing architecture of multiple GPUs working together anticipates the massively parallel computing model that will underpin deep learning a decade later.
2006 CUDA launched — general-purpose GPU computing Nvidia launches CUDA (Compute Unified Device Architecture) in 2006 — a programming platform that allows developers to use Nvidia GPUs for general-purpose computing beyond graphics. CUDA is the crucial strategic move that transforms Nvidia from a gaming chip company into an AI infrastructure company. Researchers immediately recognise that GPUs are ideal for the massively parallel matrix calculations required by neural networks. Without CUDA, the deep learning revolution could not have happened as it did.
2007 Tesla computing platform — Nvidia enters data centres Nvidia introduces the Tesla computing platform in 2007 — GPUs designed specifically for scientific computing, oil and gas exploration, financial modelling and other high-performance computing applications. The Tesla (named after Nikola Tesla, not the car company) is the first Nvidia product aimed at the data centre rather than the consumer. Tesla GPUs are adopted by universities and research labs for molecular dynamics, climate modelling and other computationally intensive workloads.
2008 Nvidia mobile GPU defect — $196 million charge Nvidia discloses in July 2008 that a manufacturing defect in its mobile GPU chips — used in hundreds of thousands of laptops from Dell, HP and Apple — is causing widespread failures. Nvidia takes a $196 million charge and replaces affected GPUs. The defect, caused by improper underfill material in the flip-chip bonding process, is one of the most expensive product defects in Nvidia's history and temporarily strains relationships with major laptop manufacturers.
2009 Fermi architecture — first serious HPC GPU Nvidia releases the Fermi GPU architecture in 2009 — its most significant architectural advancement since CUDA. Fermi doubles the number of CUDA cores, adds error-correcting memory essential for scientific computing, and introduces concurrent kernel execution. Fermi establishes Nvidia's Tesla GPUs as the standard computing platform for HPC clusters worldwide. The world's fastest supercomputers begin incorporating Nvidia GPUs alongside traditional CPUs.
2011 Tegra — Nvidia enters mobile chips Nvidia's Tegra system-on-chip — combining CPU, GPU and multimedia processors — begins to gain traction in smartphones, tablets and automotive systems. The Tegra 2 powers several Android tablets. While Nvidia never becomes a dominant mobile chip supplier (Qualcomm wins that battle), the Tegra programme gives Nvidia experience in mobile power efficiency and leads directly to the Drive platform for autonomous vehicles.
2012 AlexNet — deep learning revolution begins on Nvidia GPUs At the ImageNet competition in October 2012, Alex Krizhevsky's AlexNet neural network — trained on two Nvidia GTX 580 GPUs — achieves a top-5 error rate of 15.3%, dramatically better than any previous approach. The result triggers the deep learning revolution. Every major AI research lab immediately adopts Nvidia GPUs. Jensen Huang later calls this moment 'the big bang' of the AI era. Nvidia's CUDA ecosystem, built patiently since 2006, has no serious competitor.
2014 Maxwell architecture — efficiency leap Nvidia's Maxwell GPU architecture (2014) dramatically improves energy efficiency — delivering twice the performance per watt of its predecessor. As deep learning adoption accelerates in research labs and companies, Nvidia's GPU revenues grow rapidly. Google, Microsoft, Facebook and Amazon all build GPU clusters for AI training. Nvidia's data centre business grows from near-zero to hundreds of millions of dollars annually. The strategic importance of CUDA's ecosystem lock-in becomes clear.
2016 Pascal architecture — P100 delivered to Elon Musk Nvidia's Pascal GPU architecture launches in 2016 with the P100 — the most powerful computing chip ever made at the time. Pascal is designed from the ground up for deep learning: its NVLink interconnect allows multiple GPUs to work together as one. Jensen Huang delivers a P100 personally to Elon Musk for Tesla's Autopilot development. Nvidia's stock rises 225% in 2016 — the best performance of any S&P 500 stock that year.
2017 Volta architecture — tensor cores invented Nvidia releases the Volta GPU architecture in 2017, featuring dedicated Tensor Cores — processing units designed specifically for the matrix multiplication operations at the heart of deep learning. The V100 GPU is 12 times faster than its predecessor for AI training. Cloud providers immediately offer V100 instances. The V100 becomes the standard computing unit for AI research globally. Nvidia's data centre revenues surpass gaming revenues for the first time — a strategic milestone.
2018 Nvidia market cap crosses $100 billion Nvidia's market capitalisation crosses $100 billion for the first time in 2018, driven by AI, gaming and data centre growth. Nvidia is now the world's most valuable semiconductor company — surpassing Intel. The milestone validates the decade-long bet on CUDA and GPU computing. However, a cryptocurrency mining boom and bust causes Nvidia's stock to fall 50% by late 2018 — a temporary setback before the AI-driven surge resumes.
2018 RTX — real-time ray tracing comes to gaming Nvidia launches its Turing GPU architecture in September 2018, featuring dedicated hardware for real-time ray tracing — a lighting simulation technique that produces photorealistic game graphics. The RTX 2080 is the first gaming GPU to support ray tracing in real time. The RTX brand replaces GTX as Nvidia's gaming flagship designation and becomes one of the most recognised names in PC gaming.
2020 Ampere architecture — A100 becomes AI standard Nvidia launches the Ampere GPU architecture in 2020, with the A100 data centre GPU delivering 20 times the AI performance of the previous V100. The A100 becomes the standard chip for training large language models — GPT-3, BERT, and every major AI model trained between 2020 and 2023 uses A100s. Cloud providers offer A100 instances within months. Nvidia's data centre revenues grow from $3 billion in 2019 to $10 billion in 2021. The A100 is arguably the most strategically important chip in AI history.
2020 Nvidia announces $40 billion Arm acquisition — ultimately fails Nvidia announces its intention to acquire Arm Holdings — the British chip designer whose architecture powers virtually every smartphone in the world — from SoftBank for $40 billion in September 2020. The deal faces opposition from regulators in the US, UK, EU and China. After 18 months of regulatory battle, Nvidia abandons the acquisition in February 2022, paying a $1.25 billion break fee. The failed deal costs Nvidia significant management bandwidth during a critical AI growth period.
2021 Nvidia Drive — autonomous vehicle platform Nvidia's Drive platform for autonomous vehicles gains major traction: Mercedes-Benz signs a deal to use Nvidia Drive as the computing platform for its next generation of vehicles. Nvidia positions itself as the AI computing stack for the automotive industry — not just chips but the full software platform for perception, mapping and planning. The automotive segment grows into a multi-billion dollar business as car manufacturers invest heavily in autonomous and assisted driving.
2022 Hopper architecture — H100 defines the AI era Nvidia's H100 GPU, based on the Hopper architecture announced in 2022 and ramping through 2023, becomes the most in-demand chip in the world. The H100 delivers 30 times the AI inference performance of the A100. Every major AI company — OpenAI, Google, Microsoft, Meta, Amazon, Anthropic — builds its AI infrastructure on H100 clusters. Nvidia supplies chips costing tens of billions of dollars. The H100 is both Nvidia's greatest product and the choke point of the global AI build-out.
2022 ChatGPT launches — Nvidia demand explodes OpenAI launches ChatGPT in November 2022, triggering a global AI arms race. Every company rushes to develop large language models — all requiring Nvidia H100 GPUs. Nvidia's data centre revenues go from $3.8 billion per quarter in early 2023 to $18.4 billion per quarter by late 2023 — the fastest revenue growth of any large company in history. H100 GPUs have waiting lists of 6–12 months. A single H100 costs $30,000–$40,000 on secondary markets. Nvidia's gross margins exceed 70%.
2023 Jensen Huang — rock star CEO Jensen Huang's public profile transforms in 2023 from respected technology executive to global celebrity. His GTC keynotes attract hundreds of thousands of online viewers; he is mobbed for autographs at trade shows; his black leather jacket becomes a cultural symbol. Time magazine names him one of the world's 100 most influential people. His ability to communicate complex AI computing concepts clearly to investors, customers and the public makes him one of the most effective technology CEOs in history.
2023 Nvidia joins the trillion-dollar club Nvidia's market capitalisation crosses $1 trillion for the first time on 30 May 2023 — making it the sixth company in history to reach this valuation, joining Apple, Microsoft, Alphabet, Amazon and Saudi Aramco. Jensen Huang rings the Nasdaq opening bell. Nvidia's quarterly revenue guidance of $11 billion — nearly double analyst expectations — triggers the most value added in a single day in stock market history.
2024 Blackwell architecture — worlds most powerful AI chips Nvidia releases the Blackwell GPU architecture in 2024 — the B100 and B200 chips deliver four times the AI training performance of Hopper. Demand for Blackwell exceeds supply by several times. Customers including Microsoft, Google, Amazon and Meta pre-order tens of billions of dollars of Blackwell GPUs. Nvidia's market cap surpasses $3 trillion in June 2024, briefly making it the world's most valuable company. Jensen Huang's personal net worth exceeds $100 billion.
2024 NIM and software — Nvidia beyond hardware Nvidia launches Nvidia Inference Microservices (NIM) — a software platform for deploying AI models in production. The move signals Nvidia's ambition to become a software and platform company, not just a chip maker. NIM, combined with the CUDA ecosystem, Nvidia's networking (Mellanox, acquired 2020) and its DGX systems, makes Nvidia the end-to-end AI computing stack. Jensen Huang declares: 'We are a full-stack computing company.'
2024 Nvidia becomes most valuable company AI chip demand drives $3 trillion valuation Nvidia briefly surpassed Microsoft and Apple to become the world's most valuable company in June 2024, driven by extraordinary demand for its AI chips. Its market cap reached $3.3 trillion.
2024 Nvidia briefly becomes worlds most valuable company On 18 June 2024, Nvidia's market capitalisation reaches $3.34 trillion — briefly surpassing Microsoft and Apple to make Nvidia the world's most valuable publicly traded company. The milestone, achieved just 31 years after Huang, Malachowsky and Priem met in a Denny's diner, is driven by insatiable demand for AI computing infrastructure. Nvidia's stock has risen over 700% in two years. The company's dominance of AI computing is compared to Intel's 1990s PC dominance — but more complete.
2025 Nvidia GB200 NVL72 — the AI supercomputer in a rack Nvidia's GB200 NVL72 system — 72 Blackwell GPUs connected via NVLink in a single rack — becomes the standard building block for AI data centres in 2025. A single NVL72 rack delivers 1.4 exaflops of AI performance and costs approximately $3 million. Microsoft, Google, Meta and Oracle order thousands of racks. The NVL72 is not just a chip but a complete computing system — Nvidia has vertically integrated from silicon to rack to software stack.
2025 Rubin architecture announced — next generation Nvidia announces the Rubin GPU architecture (named after astronomer Vera Rubin) for 2026, continuing its annual GPU architecture cadence. Nvidia also announces NVLink Fusion — allowing Nvidia GPUs to be combined with non-Nvidia CPUs and custom chips — an attempt to expand its ecosystem even as AMD, Intel and custom chips from Google, Amazon and Microsoft attempt to reduce dependence on Nvidia. The AI chip race intensifies but Nvidia's software moat via CUDA remains formidable.
2026 Nvidia revenue exceeds $200 billion annually Nvidia's annual revenues surpass $200 billion in fiscal year 2026 — a figure that would have been inconceivable when Jensen Huang was sketching chip designs at a Denny's in 1993. The company's data centre business alone generates more revenue than Intel's entire company. Nvidia employs approximately 36,000 people — a remarkably small workforce for a company of its market cap and revenue — reflecting its fabless model and software-heavy value creation.