Nvidia

• 1. Introduction

  ›

  Nvidia was founded at a Denny's roadside diner on Berryessa Road in East San Jose, where three engineers sat down in late 1992 and agreed to build a company around a bet that graphics processors would change computing forever. Jensen Huang, Chris Malachowsky, and Curtis Priem each walked away from stable, well-paying jobs at companies like AMD, Sun Microsystems, and IBM to pursue that vision. What began with $40,000 in the bank and a shared desk in Priem's townhouse in Fremont, California, grew into the first company in history to surpass $4 trillion and then $5 trillion in market capitalization. The questions worth asking are not simply how a graphics card company became the most valuable on earth, but what happened in the decades between that Denny's booth and the armored cars now delivering Nvidia chips to the world's biggest data centers.

• 2. The Founding Deadlock

  ›

  Curtis Priem resigned from Sun Microsystems effective the 31st of December 1992, and that act of nerve broke a standoff that might have killed the company before it started. Jensen Huang's wife, Lori, would not let him quit his job at LSI Logic unless Malachowsky quit Sun at the same time. Malachowsky's wife, Melody, held the same condition in reverse. Priem, who had no such arrangement to honor, simply handed in his notice first, putting pressure on the other two to not leave him, as Priem later put it, to "flail alone." Huang officially joined Priem on February 17, which happened to be Huang's 30th birthday. Malachowsky followed in early March. The three men had spent their planning sessions at Denny's discussing a vision of graphics-based computing so compelling that Huang walked away from running his own division at LSI Logic. As Huang later explained their early logic: "Video games was our killer app - a flywheel to reach large markets funding huge R&D to solve massive computational problems." As for the company's name, it came from Priem's search through words containing the letters "NV", derived from "next version" of a chip series he had worked on at Sun. Both Priem and Huang have each claimed credit for landing on Nvidia, drawn from "invidia", the Latin word for envy. The venture capital that followed, including money from Sequoia Capital and Sutter Hill Ventures, brought the total to $20 million in early funding.

• 3. Thirty Days from Collapse

  ›

  Shoichiro Irimajiri, Sega's president, visited Jensen Huang in person to deliver the news that Nvidia would not be supplying the graphics chip for the Dreamcast console, a contract the company had been working on for roughly a year. Rather than simply walking away, Irimajiri persuaded Sega's management to invest $5 million in the company anyway. Huang would later reflect that this money was all that kept Nvidia alive, and that Irimajiri's "understanding and generosity gave us six months to live." The immediate problem was that Nvidia's first graphics accelerator, the NV1, had been designed around quadrilateral primitives rather than the triangle-based approach that Microsoft chose when it built its Direct3D API into the DirectX platform. That mismatch killed the NV1's prospects. In 1996, Huang cut Nvidia's headcount from 100 employees down to 40 and redirected everything toward building a triangle-optimized accelerator called the RIVA 128. By August 1997, when the RIVA 128 shipped, the company had enough money left for exactly one month of payroll. That precariousness gave rise to Nvidia's unofficial motto, which Huang reportedly recited at the start of internal presentations for years: "Our company is thirty days from going out of business." Nvidia sold roughly a million RIVA 128 units within four months of release, and used that revenue to fund development of the RIVA TNT in 1998, which began to establish the company's standing in the graphics market. When Nvidia went public on the 22nd of January 1999, the $5 million that Irimajiri had persuaded Sega to invest turned out to be one of his best decisions as the company's president. After he left Sega in 2000, the company sold its Nvidia stock for $15 million.

• 4. The GPU Takes Shape

  ›

  In late 1999, Nvidia released the GeForce 256, which it called the NV10, and this was the first product the company expressly marketed as a GPU. Running at 120 MHz with four-pixel pipelines, it introduced onboard transformation and lighting to consumer-level 3D hardware and outperformed existing products by a wide margin. The success of that product led Microsoft to award Nvidia the contract to develop graphics hardware for the Xbox game console, a deal that included a $200 million advance. In December 2000, Nvidia reached an agreement to acquire the intellectual assets of 3dfx, a pioneer in consumer 3D graphics that had led the field through much of the 1990s. The acquisition was finalized in April 2002. In 2001, Standard and Poor's added Nvidia to the S&P 500, choosing it to replace Enron. That same stretch of years brought a wave of smaller acquisitions: Exluna in July 2002, MediaQ for approximately $70 million in August 2003, iReady in April 2004, and ULI Electronics in December 2005. In February 2008, Nvidia bought Ageia, the developer of the PhysX physics engine, with plans to fold that technology into its GPU products. Then in July 2008, the company took a write-down of approximately $200 million after reporting that certain mobile chipsets and GPUs had what it described as "abnormal failure rates" due to manufacturing defects. A class action lawsuit followed in September 2008, and in September 2010 Nvidia reached a settlement covering repairs or replacements for owners of affected laptops. On the 10th of January 2011, Nvidia signed a six-year, $1.5 billion cross-licensing agreement with Intel that ended all litigation between the two companies.

• 5. CUDA and the AI Pivot

  ›

  In the early 2000s, Nvidia invested over a billion dollars to build CUDA, a software platform and programming interface that let developers run massively parallel programs on GPU hardware rather than conventional central processors. That investment looked like a bet on an uncertain future for most of the decade. Then in 2009, researchers at Google Brain used Nvidia GPUs to create deep neural networks capable of machine learning, and Andrew Ng determined at the time that GPUs could increase the speed of deep learning systems by roughly 100 times. Nvidia describes 2009 as the "big bang" of deep learning for exactly this reason. The CUDA platform sat at the center of that shift, because it gave programmers a way to direct GPU cores toward the matrix operations that underpin machine learning. By 2025, Nvidia controlled more than 80% of the market for GPUs used in training and deploying AI models, and its chips powered over 75% of the world's TOP500 supercomputers. The company's data center business grew so dominant that in January 2024, analysts at Raymond James Financial estimated individual H100 GPUs were selling in the $25,000 to $30,000 range, while individual units on eBay were fetching over $40,000. Nvidia's first DGX-1 supercomputer, built in April 2016 from an eight-GPU cluster, was gifted to OpenAI in August 2016 and reportedly reduced certain processing jobs from six days to two hours. The GPU Technology Conference, which Nvidia launched in 2009 in San Jose with a focus on GPU computing challenges, drew roughly 59,000 registrants when it converted to a digital format in 2020, reflecting how broadly the AI field had grown around Nvidia's hardware.

• 6. Trillion-Dollar Milestones

  ›

  In May 2023, Nvidia crossed $1 trillion in market valuation during trading hours. In 2023, it became the seventh U.S. company to reach that level. On the 1st of March 2024, it became only the third company in U.S. history to close with a market capitalization above $2 trillion, and it needed just 180 days to get there from $1 trillion, compared to more than 500 days for Apple and Microsoft. On the 18th of June 2024, Nvidia surpassed both Microsoft and Apple to become the world's most valuable company, with a market cap exceeding $3.3 trillion. On the 10th of July 2025, Nvidia closed for the first time above $4 trillion, becoming the first company in history to reach that level, at which point it was worth more than the combined value of all publicly traded companies in the United Kingdom. On the 29th of October 2025, it became the first company to reach $5 trillion. In January 2025, Nvidia also experienced the largest single-day loss in market capitalization ever recorded for a U.S. company, shedding $600 billion in one session after DeepSeek, a Chinese AI startup, released an advanced AI model built at substantially lower cost and computing requirements than existing alternatives. By early 2026, Nvidia's market cap stood at $3.66 trillion, itself worth more than double the combined value of AMD, ARM, Broadcom, and Intel. The September 2023 reunion at the original Denny's booth on Berryessa Road, where Denny's CEO Kelli Valade joined Huang to install a plaque marking the corner booth as the birthplace of a $1 trillion company, captures how much distance had been covered from that first meeting in late 1992.

• 7. Competition, Controversy, and China

  ›

  In December 2006, both Nvidia and AMD received subpoenas from the United States Department of Justice regarding possible antitrust violations in the graphics card industry. In 2022, the SEC charged Nvidia with failing to disclose that cryptomining was a significant driver of revenue growth from chips sold for gaming. Nvidia agreed to pay $5.5 million to settle the civil charges without admitting or denying the findings. In June 2024, the Justice Department opened an antitrust investigation into Nvidia's conduct in the AI industry. On the 26th of September 2023, French competition authorities searched Nvidia's French offices as part of a suspected anti-competitive practices investigation in the graphics card sector. The company did not publicly comment. On the review front, in December 2020, Nvidia told YouTube reviewer Steven Walton of Hardware Unboxed it would no longer supply him with Founders Edition graphics card samples, citing his editorial focus on rasterization over ray tracing. The move drew heavy criticism from other technology reviewers before Nvidia reversed course two days later. Senior PR manager Bryan Del Rizzo acknowledged in a subsequent apology that "to withhold samples because I didn't agree with your commentary is simply inexcusable and crossed the line." China has become an increasingly difficult market. The H20 chip, developed in late 2023 specifically to comply with U.S. export restrictions, featured 96 GB of HBM3 memory and demonstrated over 20% faster performance than the H100 in large language model inference tasks despite sharply reduced computational power at 296 TFLOPs versus the H100's 1979 TFLOPs. In August 2025, Chinese government directives led Nvidia to halt H20 production, and Jensen Huang denied allegations that the chip contained security backdoors. The company was simultaneously developing the B30A, a Blackwell-architecture chip intended as the H20's successor for China.

• 8. What Comes Next

  ›

  At GTC 2025, Huang projected that AI-driven infrastructure would push Nvidia's data center revenue to $1 trillion by 2028. The Blackwell architecture, named after mathematician David Blackwell and announced on the 18th of March 2024, underpins the current product generation, and by November 2024, Morgan Stanley reported that the entire 2025 production run of Blackwell chips was already sold out. At the GTC conference in March 2026, Nvidia announced the Vera Rubin platform as Blackwell's successor, pairing the Rubin GPU with the Vera CPU and targeting what the company describes as "agentic AI" and reasoning-capable computing at scale. In October 2024, Nvidia released NVLM 1.0, an open-source family of multimodal large language models with a flagship version carrying 72 billion parameters. In December 2025, the Nemotron 3 family followed with Nano, Super, and Ultra tiers built on a hybrid mixture-of-experts architecture, with Ultra reaching 500 billion parameters. Nvidia announced in December 2025 plans to build a major research and development campus in Kiryat Tivon, Israel, projected to employ more than 10,000 people across a 22-acre site with construction beginning in 2027 and initial operations expected in 2031. In January 2026, the company launched Earth-2, an open-source weather forecasting platform. The Inception Program, which Nvidia created to support AI and data science startups, had grown by August 2021 to more than 8,500 member companies across 90 countries, with cumulative funding among those members reaching $60 billion. For a company that survived on Sega's generosity and thirty-day payroll windows in the 1990s, the scale of its open-source commitments, its chip architecture roadmap, and its $60 billion startup ecosystem suggest that the original bet on parallel computing still has considerable room to run.