1993.04
Jensen Huang and two co-founders start NVIDIA
1999.08
GeForce 256 launch — "invented the GPU"
2006.11
CUDA released; the ecosystem moat begins
2012
AlexNet wins with GPU training; deep learning ignites
2016.04
First DGX-1 delivered to OpenAI
2020.04
~$6.9B Mellanox acquisition completed
2022.03
H100 released with Transformer Engine
2024.03
Blackwell architecture announced (GTC)
2025.10
Market cap breaks $5T — first in the world
2026.10
CoreWeave opens NVL72 (Cognition first production customer); Microsoft + NVIDIA launch the RTX Spark PC lineup and open pre-orders at the Oct 7 Windows event (Surface Laptop Ultra from $2,599, shipping Oct 16); DGX Spark 64GB launched (Oct 2, from $4,999); PAIR leaves beta — free and open-source (Oct 13, adds llama.cpp support and direct Hugging Face downloads); RTX Spark laptops ship Oct 16 with a software-ecosystem update (GitHub Copilot, Hermes, OpenClaw, Perplexity, Pi, WorkBuddy and more, simplified install); Nemotron 3.5 Bolt 74B-A9B officially launched (Oct 15: open agentic AI for RTX Spark laptops, fully open commitment, covered by multiple tech outlets); Nemotron Personas UK open dataset released (Oct 14: Locai Labs × NVIDIA, 1M synthetic UK-population personas, CC BY 4.0, EINPresswire corporate release, single source); Nemotron 3.5 Super VL flagship multimodal model launched (Oct 15: 120B total / 12B active parameters, 1M context, OpenMDW-1.1 open license, cross-covered by multiple tech outlets); Nemotron 3.5 family downloadable from Oct 16 (Super VL 120B-A12B and Bolt 74B-A9B on Hugging Face, NVIDIA NIM and ModelScope — open weights + datasets + training recipes, OpenMDW-1.1 commercial license; smaller Bolt E4B (4GB) due Oct 23); German industrial-AI joint declaration of intent signed (Oct 19: Huang and Chancellor Merz on stage at the Invest in Germany summit — intent stage, not a final contract; seed funding of several hundred million euros mainly to rent Blackwell compute on Deutsche Telekom's Industrial AI Cloud; Siemens, Audi, Black Forest Labs and about a dozen other German companies involved; advancing physical-AI open models around the SOOFI sovereign open-model program; NVIDIA official blog + German government portal + Reuters, multiple sources; more detail expected at Huang's GTC Berlin keynote Oct 21); Modal adopts the Vera CPU (Oct 19: Reuters; first publicly named third-party customer; Modal closed a $750M round at a $15.75B valuation the prior week, previously Intel/AMD only; early tests doubled throughput on comparable tasks; NVIDIA expects at least $20B in Vera chip sales this fiscal year, with OpenAI, Anthropic and Oracle also planning adoption)
10-20
GTC Berlin opening-day details on the German industrial-AI partnership: Huang announces investment plans in four German startups (Agile Robots — Munich robot arms/humanoids; Avelios Medical — Munich AI hospital information systems; Mirelo AI — sound/voice generation; Reflex Aerospace — Berlin satellites), part of NVIDIA's ~$10B European commitment; Germany's economics ministry signs an MoU with NVIDIA (details undisclosed); Siemens shows an AI planning agent built with NVIDIA at its Erlangen electronics plant (engineering-preparation workload down 85%, production-planning cost down 20%); NVIDIA runs a record 104-qubit fault-tolerant quantum-computer simulation on the Jülich JUPITER supercomputer
10-21
Huang's GTC Berlin keynote (17:00 Beijing time), three major announcements: DSX POD modular AI-factory reference design (container-prefabricated, on-site deployment cut from 6 months to 14 days, global delivery from early 2027; Dell / HPE / Lenovo / Foxconn / Quanta and others; DSX closed-loop liquid cooling at 45°C warm water and DSX Flex dynamic power); OpenShell agent-security reference architecture (Rubin does the "thinking", Vera CPU runs the agent harness, BlueField-4 runs Sentry hardware-level monitoring; Anthropic / OpenAI / Google / Meta / Microsoft / AWS / SpaceX Super Intelligence and others involved); Kumo open-source models + Agent Knowledge Foundation; Vera Rubin announced in full mass production with the supply chain locked (substrate capacity to grow 2.5× in 3 years, $1.1B KYEC booking); ~$10B European commitment; Agile Robots (Munich) plans to build 1,000–2,000 small humanoid robots next year (disclosed in the keynote)
10-23
Nemotron 3.5 Bolt E4B officially launched (~4GB VRAM, 64K-token context, 35 languages, multi-turn tool use, for local agents on RTX / Jetson / Spark; NVIDIA Robotics posted "Available now" on LinkedIn; on Hugging Face, OpenMDW license)
10-25
Flora open-sourced (coverage wave; paper and model cards dated 10-20): NVIDIA's GEAR lab open-sources three robot-video annotation models fine-tuned from Qwen3.5 (9B segmenter + two 27B describer and reward models), already used in GR00T training; claimed annotation cost $0.72 per video hour (vs $88.49 for GPT-6 Astra, modeled price not a measured bill), 57.1% quality on a 50-episode internal benchmark vs Astra's 53.7%, non-commercial research license
11-05
Cosmos 3 world models open-sourced (four post-trained weights, each a post-trained Cosmos 3 Nano with 16B trainable parameters, OpenMDW 1.1 commercial license: Transfer-Auto, Sim-Auto, Sim-Transfer, Sim-Bimanual; announced on the Hugging Face blog)
11-05
Plan to convert 130+ Brightspeed telecom buildings owned by Apollo into small inference data centers (The Information, Nov 5; widely re-reported Nov 6–7): written proposal submitted to Apollo, covering up to 20 states and ~1GW of power, room for 2,200+ Vera Rubin NVL72 racks, positioned as an "inference network" rather than a training cluster, conversion to start as early as Q1 2027; no final agreement, no confirmed customer list
11-06
Southeast Asian cloud vendors required to join the Fleet Intelligence chip-tracking program to qualify as Cloud Partners (Bloomberg, Nov 6, citing people familiar): the system collects GPU operating data and location via a resident read-only agent to curb smuggling to China; so far applies to new applicants in Southeast Asia, unclear whether it extends to existing partners or other regions, NVIDIA did not respond; cloud vendors also reportedly must reserve at least 51% of compute for non-China customers; ~600K H200s stranded in Hong Kong (Reuters, Nov 6, re-reported): only ~10K-unit batches approved for a few major Chinese internet companies, total foreign-chip approvals possibly below 200K; NVIDIA's 10-Q shows shipped volumes far below US license volumes, data-center revenue share under 1%, and a $400M H200 inventory charge
11-07
NYT investigation: US Commerce approved ~24 Chinese companies to buy or handle H200s, including New Matrix (founded just 17 months ago, Beijing registered address an unfinished bare-shell unit) and Maginfra (registered office empty); New Matrix shareholders are linked to Wise Road, added to the Entity List in 2024, and Maginfra is alleged to be linked to sanctioned Inspur (shared personnel and patents); the approvals cover assembly/handling licenses, no chip delivery confirmed, Commerce did not publicly respond; Congress demands NVIDIA answer on the licensing program by Nov 18
11-09
Board launches independent investigation (announced Nov 8: involving CEO Jensen Huang's relationship with Alice Huang, a former senior executive at Singapore's Megaspeed; Huang has admitted the extramarital affair in an internal letter and apologized, pledging to cooperate and remain CEO; the board says it has "no reason to believe national security or NVIDIA confidential information has been leaked"; China's foreign ministry called the allegations "groundless" on Nov 9); Rosen law firm files a proposed securities class action Nov 9 (alleging misleading disclosures about the relationship and chip export-control risks; proposed class period 2025-10-09 to 2026-11-08); chairs of the House Foreign Affairs, Intelligence and China Select committees demand Huang attend a closed-door briefing by Nov 18 (TechCrunch, YourStory and others cross-reported Nov 9; allegations untested in court)
11-10
Shares fall another ~5.5% to ~$168.9 (Yahoo -5.47% / CNBC -5.54%), after a 4.27% drop on Nov 9 — ~$400B in market value wiped out in two days, turning negative for the year; a shareholder class action has been filed (five minutes after the Nov 9 close, case no. 5:26-cv-13448, alleging NVIDIA and Huang concealed the relationship with Alice Huang, a former senior executive at Singapore's Megaspeed, and related national-security risks); at least five law firms announced investigations the same day; a Delaware senator called for a congressional probe; the three House committee chairs demand a reply from Huang by Nov 11 and a closed-door briefing on Nov 18, or they are prepared to issue a subpoena; Trump said the FBI is investigating the relationship (no FBI confirmation seen); ARK bought ~$17M of NVDA on Nov 9; the board has hired Wachtell Lipton to lead the independent investigation, no findings yet; NVIDIA says it currently has no reason to believe national security or company secrets have leaked; TSMC's board approves up to $500M each in Anthropic and OpenAI (plus a $40.93B capex budget and a 28.6% quarterly dividend increase to NT$9 per share)
11-11
Reuters exclusive: H200 chips reached a PLA-linked university (reviewing 26 procurement records: Northwestern Polytechnical University bought 2 H200s for ¥874,700 on Oct 21 via domestic intermediary Shaanxi Bosuan Network, plus 2 more for ¥796,000; five compute-leasing contracts signed this year, including servers with 8 H200s each, involving A100/A800; the records do not prove NVIDIA supplied directly or knew of violations, and all three parties declined to comment; the Trump administration has applied case-by-case H200 approvals since January, and blacklisted entities should be denied); Baird cuts NVIDIA from Buy to Neutral, slashing its target from $500 to $200 (citing China chip-export curbs and a ~40% slowdown in 2027 hyperscaler capex growth); on a CoreWeave earnings boost, NVDA closed up 2.87% at $173.73, with 266.6M shares traded (~2× the three-month average); Q3 earnings due Nov 17 after market
11-13
CEO search update: FT reports the board has approached Arm CEO Rene Haas about succeeding Jensen Huang, also naming internal candidate Ian Buck; the board states the same day it has not formally approached any candidate, that speculation is premature, and that it is selecting a search firm; CFO Colette Kress continues as interim CEO; the FT report was relayed by Reuters, which says it could not independently verify it (conflicting accounts); shares fall another 2.01% to $162.85 (previous close $166.19, intraday low $158.39, 360M shares traded); HSBC cuts the rating to Hold and slashes its target from $375 to $170; the next tests are the Q3 earnings on Nov 17 and the employee meeting on Nov 18; the Senate Banking Committee summons Huang to testify Dec 3 and Lutnick Dec 10 (NBC, relayed — single-source update)
11-17
Q3 FY2027 earnings smash expectations (see key numbers: revenue $115.84B, +103% YoY; data center $107.6B, +110%; adjusted EPS $2.68; GAAP net income $65.3B, +105%; gross margin 74.8%; Q4 guidance $130B ±3%, gross margin 71–72%; FY2028 revenue growth outlook raised to ~75%; Vera Rubin NVL72 has begun shipping, driving nearly $20B in incremental revenue; Q4 guidance includes no China data-center compute revenue); chairman Harvey Jones opens the call saying the board has no intention of bringing Jensen Huang back; NVDA closes Nov 17 at $155.46 (-0.44%), up nearly 5% after hours