Apr 1993
Milestone
Jen-Hsun Huang, Chris A. Malachowsky and Curtis R. Priem co-found NVIDIA in April 1993, and it is incorporated in California that month (NVIDIA Form 424B4, filed January 22, 1999).
May 1995
Launch
NVIDIA introduces its first product, the NV1 multimedia accelerator, in May 1995 and stops selling it in the first quarter of 1996 (NVIDIA Form 424B4, filed January 22, 1999).
Aug 1997
Launch
NVIDIA begins commercial shipment of the RIVA128, its first 3D graphics processor, in August 1997; product revenue rises from $3.7 million in 1996 to $27.3 million in 1997 (NVIDIA Form 424B4, filed January 22, 1999).
Jan 1999
Milestone
NVIDIA sells 3,500,000 shares of common stock at $12.00 a share in its IPO, $42.0 million before underwriting discounts (NVIDIA Form 424B4, filed January 22, 1999).
Aug 1999
Launch
NVIDIA begins commercial shipment of the GeForce 256 in August 1999, which it calls the world's first GPU, with 23 million transistors (NVIDIA Form 10-K405, filed March 13, 2000).
Jan 2001
Launch
NVIDIA begins commercial shipment of the GeForce3 in January 2001, which it calls the first GPU with full vertex and pixel shader programmability and the foundation for Microsoft's Xbox (NVIDIA Form 10-K405, filed April 27, 2001).
Apr 2001
Milestone
NVIDIA completes its purchase of 3dfx Interactive's graphics assets on April 18, 2001, under an agreement signed December 15, 2000, and has paid about $70 million in total consideration (NVIDIA Form 10-K405, filed April 27, 2001).
Nov 2006
Launch
NVIDIA launches the GeForce 8800 in November 2006, which it calls the first DirectX 10 GPU, with a unified shader architecture and 128 stream processors running at 1.35 GHz (NVIDIA Form 10-K, filed March 16, 2007).
2007
Launch
NVIDIA makes available the first public version of its CUDA software developer kit and C compiler in fiscal 2008, and has a CUDA 1.0 FAQ online by May 21, 2007 (NVIDIA Form 10-K, filed March 21, 2008; NVIDIA Developer Forums, May 21, 2007).
2007
Launch
NVIDIA introduces Tesla, a GPU computing family for high-performance computing with the C870 processor, D870 deskside supercomputer and S870 1U server, and begins shipping the C870 and D870 in the third quarter of fiscal 2008 (NVIDIA Form 10-K, filed March 21, 2008; NVIDIA Tesla product overview, June 2007, nvidia.com).
2012
Milestone
The AlexNet neural network, trained on NVIDIA GPUs, wins the ImageNet computer image recognition competition in 2012, which NVIDIA calls the "Big Bang" moment of AI (NVIDIA Form 10-K, filed February 21, 2024).
Nov 2014
Launch
NVIDIA unveils the Tesla K80 dual-GPU accelerator on November 17, 2014, with 24GB of GDDR5 memory and 480GB/s of memory bandwidth, aimed at machine learning, data analytics and HPC (NVIDIA press release, November 17, 2014).
Apr 2016
Launch
NVIDIA introduces the Tesla P100 at GTC on April 5, 2016, a Pascal GPU with 15.3 billion transistors on 16nm FinFET, NVLink and 16GB of HBM2 memory at 720GB/sec (NVIDIA press release, April 5, 2016).
Apr 2016
Product
NVIDIA unveils the DGX-1 deep learning system on April 5, 2016, with eight Tesla P100 GPUs and up to 170 teraflops of half-precision peak performance; US general availability is June (NVIDIA press release, April 5, 2016).
May 2017
Launch
NVIDIA launches the Volta architecture and its first Volta processor, the Tesla V100, on May 10, 2017; with 640 Tensor Cores, the V100 delivers 120 teraflops of deep learning performance (NVIDIA press release, May 10, 2017).
Mar 2019
Milestone
NVIDIA and Mellanox announce on March 11, 2019 a definitive agreement for NVIDIA to acquire Mellanox for $125 per share in cash, a total enterprise value of about $6.9 billion (NVIDIA Form 8-K, March 11, 2019, Exhibit 99.1).
Apr 2020
Milestone
NVIDIA completes its acquisition of Mellanox on April 27, 2020 for about $7.0 billion, at $125.00 in cash per Mellanox share (NVIDIA Form 8-K, April 27, 2020).
May 2020
Launch
NVIDIA announces on May 14, 2020 that the A100, its first Ampere-architecture GPU, is in full production, with more than 54 billion transistors on a 7-nanometer process and up to 20x the AI performance of FP32 with TF32 (NVIDIA press release, May 14, 2020).
Sep 2020
Milestone
NVIDIA and SoftBank announce on September 13, 2020 a definitive agreement for NVIDIA to acquire Arm in a transaction valued at $40 billion, paid in a combination of NVIDIA shares and cash (NVIDIA Form 8-K, September 14, 2020, Exhibit 99.1).
Feb 2022
Regulatory
NVIDIA, SoftBank Group Capital and SVF Holdco (UK) mutually terminated the share purchase agreement for NVIDIA's purchase of Arm on February 8, 2022 because of significant regulatory challenges; SoftBank kept the $1.25B prepayment, NVIDIA kept its 20-year Arm license, and NVIDIA said it would record a $1.36B charge in the first quarter of fiscal 2023 (NVIDIA Form 8-K, February 8, 2022).
Mar 2022
Launch
NVIDIA announced the Hopper architecture and its first Hopper GPU, the H100, at GTC on March 22, 2022: 80 billion transistors on a TSMC 4N process, a Transformer Engine, and the first GPU to use HBM3, with 3TB/s of memory bandwidth; NVIDIA said the H100 would be available starting in the third quarter (NVIDIA press release, March 22, 2022).
Sep 2022
Product
NVIDIA announced on September 20, 2022 that the H100 was in full production, with partners planning to roll out the first H100 products in October; H100 systems were expected to ship in the coming weeks, customers could begin ordering DGX H100, and Amazon Web Services, Google Cloud, Microsoft Azure and Oracle Cloud Infrastructure would be among the first to deploy H100 cloud instances starting in 2023 (NVIDIA press release, September 20, 2022). NVIDIA's 10-K says it started shipping the H100 in fiscal 2023, which ended January 29, 2023 (NVIDIA Form 10-K for fiscal 2023, filed February 24, 2023).
Oct 2022
Regulatory
On August 26, 2022 the US government told NVIDIA it had imposed a license requirement on future exports of A100 and H100 chips, and systems that use them, to China (including Hong Kong) and Russia; NVIDIA said its third-quarter outlook included about $400M of potential China sales that could be affected (NVIDIA Form 8-K, August 31, 2022). BIS announced its advanced-computing and semiconductor-manufacturing rule on October 7, 2022 and it was published October 13 (Federal Register 87 FR 62186). NVIDIA's 10-K says it offered China customers an A800 product not subject to the new license requirements, and its first-quarter fiscal 2024 10-Q says it sold A800 and H800 products in China (NVIDIA Form 10-K for fiscal 2023, filed February 24, 2023, and Form 10-Q, filed May 26, 2023).
Nov 2022
Milestone
OpenAI released ChatGPT on November 30, 2022 (OpenAI blog post, November 30, 2022). NVIDIA's Data Center revenue was $3.83B in the quarter ended October 30, 2022 and a record $14.51B, up 279% from a year earlier, in the quarter ended October 29, 2023 (NVIDIA Form 8-K, November 16, 2022 and November 21, 2023, Exhibit 99.1). At GTC on March 21, 2023, Jensen Huang said the world was at the iPhone moment of AI (NVIDIA blog, March 21, 2023).
Mar 2023
Commercial
NVIDIA said on March 21, 2023 that DGX H100 systems were in full production and would be coming soon to enterprises worldwide, and named the first wave of recipients as CyberAgent, Johns Hopkins University Applied Physics Laboratory, KTH Royal Institute of Technology, Mitsui and Telconet (NVIDIA press release, March 21, 2023). A May 1, 2023 NVIDIA blog post reported DGX H100 systems as now shipping (NVIDIA blog, May 1, 2023).
May 2023
Milestone
NVIDIA guided second-quarter revenue to $11.00B, plus or minus 2%, against first-quarter revenue of $7.19B (NVIDIA Form 8-K, May 24, 2023, Exhibit 99.1).
Oct 2023
Regulatory
On October 17, 2023 the US government announced an interim final rule adding license requirements for exports to China and Country Groups D1, D4 and D5 of chips above set performance thresholds, naming NVIDIA's A100, A800, H100, H800, L40, L40S and RTX 4090 (NVIDIA Form 8-K, October 17, 2023); the rule was published October 25, 2023 (Federal Register 88 FR 73458). On October 23 the US government told NVIDIA the requirements applied immediately to data-center products with total processing performance of 4800 or more (NVIDIA Form 8-K, October 23, 2023).
Nov 2023
Product
NVIDIA announced the HGX H200 at SC23 on November 13, 2023: the H200 is the first GPU to offer HBM3e, with 141GB of memory at 4.8TB/s, nearly double the capacity and 2.4x the bandwidth of the A100, and systems were expected to begin shipping in the second quarter of 2024 (NVIDIA press release, November 13, 2023).
Mar 2024
Launch
NVIDIA unveiled the Blackwell platform at GTC on March 18, 2024: Blackwell GPUs carry 208 billion transistors, are made on a custom TSMC 4NP process, and join two reticle-limit dies with a 10 TB/s chip-to-chip link into one GPU (NVIDIA press release, March 18, 2024). NVIDIA's GB200 NVL72 specification lists 40 PFLOPS of sparse FP4 and 20 PFLOPS dense per Grace Blackwell Superchip, which is 20 sparse and 10 dense per Blackwell GPU, and 372GB of HBM3E per Superchip (NVIDIA GB200 NVL72 product page, accessed October 2026).
Mar 2024
Product
NVIDIA announced the GB200 NVL72 at GTC on March 18, 2024: a liquid-cooled, rack-scale system combining 36 Grace Blackwell Superchips, which hold 72 Blackwell GPUs and 36 Grace CPUs, acting as a single GPU with 1.4 exaflops of AI performance and 30TB of fast memory, with up to 30x the LLM inference performance of the same number of H100 GPUs (NVIDIA press release, March 18, 2024). NVIDIA's current specification lists 1,440 PFLOPS of sparse FP4 for the rack, which matches the 1.4 exaflops (NVIDIA GB200 NVL72 product page, accessed October 2026).
Dec 2024
Commercial
NVIDIA said on November 20, 2024 that Blackwell was in full production, and its fiscal 2025 10-K says it began shipping production systems of the Blackwell architecture in the fourth quarter of fiscal 2025 (October 28, 2024 to January 26, 2025); on February 26, 2025 it reported billions of dollars of Blackwell sales in that first quarter (NVIDIA Form 8-K, November 20, 2024 and February 26, 2025; Form 10-K, February 26, 2025).
Mar 2025
Product
Blackwell Ultra unveiled at GTC on March 18: GB300 NVL72 delivers 1.5x the AI performance of GB200 NVL72, with partner products expected from the second half of 2025 (NVIDIA press release, March 18, 2025). Each Blackwell Ultra GPU carries 288GB of HBM3e and 15 PFLOPS of dense FP4, 1.5x Blackwell's dense FP4 (NVIDIA technical blog, August 22, 2025).
Apr 2025
Regulatory
NVIDIA disclosed in an April 15 8-K that on April 9 the US government told it a license is required to export H20 chips to China, Hong Kong, Macau and D:5 countries, and said it expected up to about $5.5B of Q1 FY2026 charges; it recorded $4.5B of H20 excess inventory and purchase-obligation charges for the quarter ended April 27 (NVIDIA 8-K, April 15, 2025; Q1 FY2026 results, May 28, 2025).
Jul 2025
Regulatory
NVIDIA said on July 14 that the US government had assured it licenses would be granted to resume H20 sales to China; licenses for certain China-based customers were granted in August, but NVIDIA reported no H20 sales to China-based customers in Q2 FY2026 and no revenue or shipments under those licenses to date. It released $180M of previously reserved H20 inventory on about $650M of H20 sales to a customer outside China (NVIDIA blog, July 14, 2025; Q2 FY2026 results and 10-Q, August 27, 2025).
Nov 2025
Commercial
AWS announced on November 18 that Amazon EC2 P6-B300 instances, with 8 NVIDIA Blackwell Ultra B300 GPUs and 2.1 TB of GPU memory, were generally available through EC2 Capacity Blocks for ML and Savings Plans in US West (Oregon); the launch posts give no price, and the current Capacity Blocks rate is $129.168 per p6-b300.48xlarge instance-hour (AWS What's New and AWS News Blog, November 18, 2025; AWS Capacity Blocks pricing page).
Nov 2025
Milestone
On its November 19 Q3 FY2026 call NVIDIA said it had received Rubin silicon back from its supply chain partners and bring-up was going well, with the Rubin platform on track to ramp in the second half of 2026; it declared Rubin in full production at CES on January 5, 2026 (NVIDIA Q3 FY2026 earnings call, November 19, 2025; CES press release, January 5, 2026).
Jan 2026
Launch
NVIDIA unveiled the Rubin platform at CES on January 5, six new chips including the Vera CPU (88 custom Olympus cores), the Rubin GPU, NVLink 6 (3.6TB/s per GPU) and the ConnectX-9 SuperNIC; a Vera Rubin superchip pairs one Vera CPU with two Rubin GPUs (NVIDIA press release and technical blog, January 5, 2026).
Jan 2026
Commercial
Said at CES on January 5 that Rubin is in full production, with Rubin-based products available from partners in the second half of 2026; the Rubin GPU has 336B transistors, up to 288GB of HBM4 and 50 PFLOPS of NVFP4 inference compute, and AWS, Google Cloud, Microsoft and OCI are among the first clouds to deploy Vera Rubin instances in 2026 (NVIDIA press release and technical blog, January 5, 2026).