# Nvidia
> The US GPU designer whose chips run more than 90% of global AI model training, making it the world's most strategically contested semiconductor company.

**Meta:** type: reference · date: 2026-07-03 · heads:  · 4 takes · 2 lenses · 1 regions

## What it is

Nvidia (NVDA, NASDAQ) is a US semiconductor and software company headquartered in Santa Clara, California. It designs the graphics processing units (GPUs) and supporting software stack, chiefly CUDA (Compute Unified Device Architecture), that run more than 90% of AI model training worldwide. As of mid-2026, it is the most valuable publicly traded company by market capitalisation, surpassing US$3.4 trillion.

## History

Jensen Huang, Chris Malachowsky, and Curtis Priem founded Nvidia on 5 April 1993, with US$40,000 in seed capital, to build dedicated graphics accelerators. The company shipped its first chip, the NV1, in 1995. In 1999 it released the GeForce 256, marketed as the world's first GPU for its on-chip hardware transform-and-lighting capability, and listed on NASDAQ the same year.

The pivotal move came in 2006, when Nvidia released CUDA, giving programmers a general-purpose parallel-computing model for GPU hardware. When the deep-learning wave broke around 2012, CUDA was already the standard toolkit. Successive architectures extended the lead: Turing (2018), Ampere (2020), Hopper (2022), and Blackwell (2024-2025). The Hopper-generation H100, released in 2022, became the most contested industrial commodity of the decade. CUDA now covers an estimated 4 million developers and 3,000-plus optimised applications. Huang has served as president and CEO continuously since 1993.

## Current state

In fiscal year 2026 (ending 25 January 2026) Nvidia reported record annual revenue of US$215.9 billion, up 65% year-on-year, driven almost entirely by data-centre sales. In Q1 FY2027 (ending 26 April 2026) revenue reached US$81.6 billion, up 85% year-on-year, with data-centre revenue of US$75.2 billion, up 92%. The Blackwell B200 and GB200 Grace Blackwell systems are now the primary shipment; the successor Vera Rubin architecture is in production ramp.

Demand from hyperscalers and frontier model builders continues to run ahead of supply. GPU cloud operators such as [RunPod](/ko/n/runpod-100m-series-a-2026) and serving platforms such as [Together AI](/ko/n/together-ai-series-c-800m-2026) remain dependent on Nvidia allocation. The GPU memory bottleneck has made high-bandwidth memory (HBM) a strategic input; Micron's record HBM quarters (see [마이크론, 사상 최대 분기 실적; HBM 매진, 마진 약 81% 근접](/ko/n/micron-q3-2026-record-hbm)) track directly to Nvidia's Blackwell production schedule. Nvidia's custom-chip rivals, including the programme described in [오픈AI, 브로드컴과 공동 개발한 첫 자체 칩 '할라피뇨' 공개](/ko/n/openai-jalapeno-broadcom-chip-2026), aim to reduce hyperscaler dependence on Nvidia but have not yet reached comparable training throughput. Inference compute at scale, including the workloads run by [Cerebras for OpenAI](/ko/n/cerebras-openai-750mw-2026), benchmarks against Nvidia performance as the industry baseline.

## Relationships

The US government's export-control regime is Nvidia's most consequential external constraint. Washington banned A100 exports to China in 2022, then barred the H100 and H200 in 2023 under a performance-threshold rule. Nvidia designed the H20 as a China-compliant variant; it earned an estimated US$12-15 billion in China revenue in 2024. In December 2024 the Trump administration approved H200 sales to China, but Chinese buyers have not purchased them as of mid-2026, citing political caution. The Council on Foreign Relations estimated in December 2025 that Huawei, Nvidia's most credible Chinese rival, produces under 3% of Nvidia's AI computing output, a share projected to fall to 1% by 2027. Nvidia fabricates all advanced chips at TSMC in Taiwan; HBM memory is allocated from SK Hynix, Samsung, and Micron.

## What to watch

- Whether Chinese AI buyers resume H200 purchases following the diplomatic standoff, and what export conditions apply to the next-generation Vera Rubin chips
- Vera Rubin supply ramp and whether hyperscaler capital spending adjusts if supply constraints ease
- CUDA moat erosion: AMD ROCm 7.0 and OpenAI's Triton compiler reduce GPU switching costs; enterprise adoption rates will signal how durable the software lock-in is
- US antitrust scrutiny: the EU and US FTC have opened preliminary reviews of Nvidia's data-centre market position
- Geopolitical exposure at TSMC: any Taiwan Strait disruption would immediately constrain Nvidia's production capacity

## Regional takes (batched by bias / lens)

### official record
- **Nvidia Investor Relations** (north-america, en) — Nvidia's official financial reports hub; Q1 FY2027 results show record US$81.6bn revenue and US$75.2bn data-centre revenue, both up roughly 85-92% year-on-year.
  Source: https://investor.nvidia.com/financial-info/financial-reports/default.aspx
- **Nvidia Newsroom** (north-america, en) — Official biography of Jensen Huang, co-founder and CEO since 1993; documents his role establishing Nvidia and the GPU as the foundation of modern AI.
  Source: https://nvidianews.nvidia.com/bios/jensen-huang
- **Nvidia Corporation** (north-america, en) — Nvidia corporate overview confirming the company's self-description as the Engine of AI and its focus areas: accelerated computing, Omniverse, automotive, and healthcare.
  Source: https://www.nvidia.com/en-us/about-nvidia/

### policy analysis
- **Council on Foreign Relations** (north-america, en) — December 2025 CFR analysis estimating Huawei produces under 3% of Nvidia's AI computing output, projected to fall to 1% by 2027, and reviewing the US export-control history.
  Source: https://www.cfr.org/articles/chinas-ai-chip-deficit-why-huawei-cant-catch-nvidia-and-us-export-controls-should-remain

## Across the graph
- Related: [[openai-jalapeno-broadcom-chip-2026]], [[micron-q3-2026-record-hbm]], [[cerebras-openai-750mw-2026]], [[runpod-100m-series-a-2026]], [[together-ai-series-c-800m-2026]]
- Entities: Corporate:nvidia, Jensen Huang, Data Centers, United States, China

---
Canonical: https://rbtfl.xyz/ko/n/nvidia-dossier