NVIDIA Corporation

NVDAleader

NVIDIA is the dominant designer of graphics processing units and AI accelerator chips. Its data-center GPU monopoly powers the generative-AI revolution, with the H100 and successor Blackwell GPUs serving as the de-facto standard for training and running large language models.

⇄ Compare

Продукты и выручка

Доля выручки продукта

Структура выручки ($130.5B)

Статические данные (загрузка финансов…)

Data Center (78%)

Gaming (12%)

Professional Visualization (4%)

Automotive & Robotics (4%)

OEM & Other (2%)

Структура сегментов и ключевые клиенты

Подробности продукта

H100 / H200 GPUs52%

Data-center AI training and inference accelerators based on the Hopper architecture

B100 / B200 GPUs18%

Next-generation Blackwell-architecture data-center GPUs for large-language-model workloads

GeForce RTX Series14%

Consumer and professional gaming GPUs

Networking (Mellanox)10%

InfiniBand and Ethernet switches, NICs, and DPUs for data-center interconnects

Automotive / Embedded6%

DRIVE Orin and Thor SoCs for autonomous vehicles and robotics

Конвейер НИОКР

РазработкаBlackwell Ultra (B300)

H2 2025

Next-gen GPU with 288GB HBM3e and 1.5x the Blackwell B200 performance. Targets 2025 data-center deployments for hyperscalers.

ИсследованиеRubin Architecture (R100)

2026

Post-Blackwell architecture with NVLink 6, next-gen HBM4 memory, and enhanced transformer-engine for future LLM training workloads.

КоммерческийNVIDIA Blackwell NVL72 System

2025 (available)

Full rack-scale AI supercomputer combining 72 Blackwell GPUs with NVLink switches. Available now to hyperscalers; backlog extends into 2026.

РазработкаNVIDIA Cosmos (World Foundation Models)

2025-2026

Physical AI and robotics simulation platform using video-generative world foundation models for training autonomous vehicles and industrial robots.

ВалидацияThor Automotive SoC

2025

2000 TOPS automotive compute platform for level 4 autonomous driving. Replacing Orin in next-gen vehicle platforms at Mercedes, BYD.

Связи цепочки поставок

Ключевые поставщики

Taiwan Semiconductor Manufacturing Company

Wafer fabrication for all NVIDIA GPUs

Влияние: Critical — sole advanced-node foundry

Samsung Electronics

HBM3e memory for H100/H200

Влияние: High — secondary HBM supplier

SK Hynix

HBM3e memory modules

Влияние: High — primary HBM supplier

Micron Technology

HBM3e, GDDR6X memory

Влияние: Medium — third HBM supplier

Ключевые клиенты

Microsoft Corporation

Azure AI GPU clusters

Влияние: $15B+ annually

Alphabet Inc. (Google)

Google Cloud AI infrastructure

Влияние: $8B+ annually

Amazon.com (AWS)

AWS GPU instances

Влияние: $10B+ annually

Meta Platforms

LLM training clusters

Влияние: $12B+ annually

Oracle Corporation

Oracle Cloud GPU infrastructure

Влияние: $4B+ annually

Tesla, Inc.

Training GPUs for FSD

Влияние: $2B+ annually

Макро- и рыночный контекст

반도체Новости сектора

WFE(반도체 장비) 업사이클 + AI HBM 슈퍼사이클 진입

TSMC 가동률

92%+

HBM 수요 성장

+180% YoY

WFE 시장 규모

$105B (2026E)

리드타임 (장비)

18~24개월

Макросводка

NVIDIA is the primary beneficiary of the AI infrastructure supercycle. Hyperscaler capex commitments of $320B+ in 2026 drive GPU demand that exceeds supply through at least H1 2026. Interest rate cuts reduce cost of capital for AI infrastructure investment, further accelerating spend.

Попутный ветер для NVDA

▲Hyperscaler AI capex at record highs: Microsoft $80B, Meta $65B, Google $75B, Amazon $105B in FY2026
▲Inference demand growing faster than training — Blackwell NVL72 rack-scale systems address this directly
▲Sovereign AI: governments building national AI infrastructure (UAE, Saudi, Japan, France) — new demand source
▲Interest rate cuts reduce cost of capital for long-duration AI infrastructure projects
▲Export restrictions on China accelerate US/Europe AI buildout to compensate

Факторы риска для NVDA

▼US export controls on H20/H100 to China (~$5-8B annual revenue at risk)
▼TSMC advanced node capacity constraints limit upside (CoWoS packaging bottleneck)
▼Rising competition from custom silicon (Google TPU v5, Amazon Trainium, Microsoft Maia) in hyperscaler inference
▼AMD MI300X gaining traction in inference — price competition risk on commoditized inference workloads
▼Valuation premium (30x+ forward PE) leaves little room for execution miss

Предстоящие катализаторы

◆TSMC Q2 실적 (2026.04.17)◆NVIDIA GB300 양산 착수 발표◆AMAT/LRCX/KLAC 실적 시즌 (2026.05)

Проблемы цепочки поставок

⚠ 공급 차질GB300 CoWoS 패키징 병목 지속

2026-04

TSMC CoWoS-L 캐파 확충에도 GB300 수요가 공급을 3~4배 초과. Blackwell 출하 대기 기간 14~18주로 연장.

↑ 생산 확대멕시코 조립라인 추가 — 미국 수출 관세 회피

2026-03

Foxconn·Flextronics와 협력, 멕시코 과달라하라에 AI 서버 최종조립 라인 확충. USMCA 적용으로 관세 25% 회피.

⚡ 리스크中 수출통제 강화 — H20 추가 제한 우려

2026-02

미 상무부, H20 칩 대중 수출 추가 제한 검토 중. NVDA 중국 매출 비중 약 13%로 하향 리스크.

AI-анализ

Нажмите "Получить AI-анализ" для анализа цепочки поставок NVIDIA Corporation.

Связанные макро-темы

Все темы

Macro Information Asymmetry

Beyond insider trading — knowing the direction of policy, resource flows, and macro turning points before the market prices them in. Flowvium's news gap score quantifies this edge.

News gap13F signalInstitutional front-running

Sovereign Wealth & State Capitalism

Norway's $1.7T GPFG, Saudi Arabia's PIF, and China's CIC don't just invest — they shape industrial policy across continents. Understanding their flows reveals where governments think the next decade is heading.

Sovereign wealth fundState capitalismIndustrial policy

Информация о компании

Штаб-квартира

Santa Clara, California, USA

Основана

1993

Сотрудники

32,000+

Веб-сайт

nvidia.com

Позиция в каскаде

Роль в каскаде

leader

Типичная задержка

0 (trigger)

NVIDIA's data-center guidance is the single most important demand signal for AI chips, HBM memory, and advanced packaging.

Полный каскад

Обзор сектора — 반도체

Новости сектора