Nvidia

Founded on April 5, 1993 and headquartered in Santa Clara, California, NVIDIA started with a focus on 3D graphics for PCs and helped define the modern GPU. From gaming and professional visualization, the company expanded into accelerated computing that runs the most demanding workloads in AI and high performance computing.

‍

Today NVIDIA positions itself as a full-stack computing infrastructure company, combining GPUs with data center CPUs, high-speed networking, and a large software layer. CUDA sits at the center of that stack, giving developers a common programming model plus libraries and tools that power AI training and inference, data analytics, simulation, and 3D graphics. Its platforms serve four core markets: Data Center, Gaming, Professional Visualization, and Automotive.

‍

In recent product cycles, NVIDIA launched its Blackwell data center platform and then introduced the Rubin platform at CES 2026 as its next generation roadmap for rack-scale AI systems. Across chips, systems, and software, the company frames its mission as solving the most challenging computational problems.

NVIDIA designs accelerated computing platforms and sells them as chips, boards, modules, systems, and software. Revenue is anchored in Data Center compute and networking, supported by Gaming, with smaller contributions from Professional Visualization, Automotive, and OEM and Other products.

‍

Core activities
‍

1. Data Center compute platforms
   NVIDIA sells GPUs and systems for AI training and inference, data analytics, graphics, and scientific computing. It pairs GPUs with CPUs and DPUs and delivers many deployments as systems, subsystems, or modules. Production shipments of the Blackwell architecture began in the fourth quarter of fiscal 2025.
   ‍
2. Rack-scale AI systems
   Systems like GB200 NVL72 bundle Grace CPUs with Blackwell GPUs into a rack-scale design connected through NVLink, targeting large model training and inference clusters.
   ‍
3. Networking for AI clusters
   NVIDIA sells end-to-end InfiniBand and Ethernet platforms, spanning adapters, DPUs, switches, cables, and software. Spectrum-X is positioned as its Ethernet stack for AI data centers.
   ‍
4. Software licensing and services
   NVIDIA monetizes enterprise software through NVIDIA AI Enterprise and Omniverse Enterprise licensing, offered via subscription, cloud consumption, or perpetual licensing with support.
   ‍
5. Graphics and edge markets
   Gaming revenue is driven by GeForce GPUs and services like GeForce NOW. Professional Visualization centers on RTX workstation GPUs, vGPU software, and Omniverse. OEM and Other includes Jetson for robotics and embedded platforms.
   ‍

Market position

• Full-stack control across processors, interconnects, systems, and software lets NVIDIA optimize performance at the platform level, not only at the chip level.
  ‍
• CUDA and its developer base strengthens adoption and raises switching costs for AI and HPC workloads.
  ‍
• Broad distribution through major server makers and cloud providers increases reach, while a faster annual cadence in Data Center architectures pushes frequent refresh cycles.
  ‍
• Roadmap visibility extends beyond Blackwell, with the Rubin platform announced for partner deployments starting in the second half of 2026.

China remains a meaningful market for NVIDIA across Gaming GPUs, professional RTX workstations, and data center products used by internet platforms, cloud providers, and research institutions. In fiscal year 2025, China (including Hong Kong) represented $17.1 billion of revenue by customer billing location.

‍

Data center performance in China is shaped by U.S. export controls and local policy scrutiny. NVIDIA states that China data center revenue grew in fiscal 2025, yet it stayed below pre October 2023 levels, supported by China-specific products that do not require an export license. The company also reports that Blackwell systems require licenses for shipments to China, and licenses had not been received as of its fiscal 2025 annual filing.

‍

Recent friction increased uncertainty. NVIDIA reported zero H20 sales to China-based customers in fiscal 2026 Q2, and January 2026 reporting described Chinese customs blocking imports of NVIDIA’s H200 AI chips.

NVIDIA’s medium-term growth path is tied to the buildout of AI data centers and the shift from single chips to integrated systems that bundle compute, networking, and software. In Q3 fiscal 2026 (ended October 26, 2025), revenue reached $57.0B, with $51.2B from Data Center, split between $43.0B compute and $8.2B networking. Management guided Q4 fiscal 2026 revenue to $65.0B.

‍

Key growth drivers
‍

1. Rack-scale AI systems and faster platform cadence
   Demand is moving toward factory-style deployments where GPUs, CPUs, and interconnects ship as validated racks. NVIDIA frames this around Blackwell and its follow-on roadmap, with the Rubin platform positioned for partner rollouts in the second half of 2026.
   ‍
2. Networking attach and cluster complexity
   As clusters scale, spending shifts toward fabrics, switches, and DPUs that keep GPUs fed. NVIDIA’s Q3 networking revenue jump was linked to NVLink compute fabric for GB200 and GB300 systems, alongside InfiniBand and Ethernet.
   ‍
3. Inference growth and “agentic” workloads
   NVIDIA highlights demand tied to larger model serving, copilots, and agent workflows. That mix tends to favor broader deployment footprints and recurring refresh cycles in cloud and enterprise environments.
   ‍
4. Software and cloud services as margin support
   The company is expanding software packaging and cloud offerings, reflected in $26.0B of multi-year cloud service agreements reported in Q3 fiscal 2026, aimed at R&D and DGX Cloud.
   ‍
5. Automotive and robotics as longer-cycle options
   Automotive remains smaller than Data Center, yet partnerships around DRIVE and robotaxi programs signal continued platform adoption and a pipeline tied to model-year launches.
   ‍

Challenges ahead
‍

• Export controls and China volatility: Licensing changes already forced an H20-related charge in fiscal 2026 Q1, and policy shifts around which products ship where stay a structural overhang.
  ‍
• Rising competition and custom silicon: AMD is pushing Instinct MI350-series positioning for large-model training and inference, while hyperscalers keep investing in in-house accelerators like Google TPUs and AWS Trainium systems.
  ‍
• ‍Execution risk in system ramps: Bigger systems raise supply-chain and integration complexity, and NVIDIA’s Q3 disclosure of $50.3B in supply-related commitments shows how much capacity planning matters.