NVIDIA: The Infrastructure Layer of the AI Supercycle

Business Overview

NVIDIA Corporation designs and sells graphics processing units (GPUs), system-on-chip units, and related software. Founded in 1993 by Jensen Huang, Chris Malachowsky, and Curtis Priem, the company has evolved from a gaming graphics company into the dominant provider of AI computing infrastructure.

The company operates through two segments: Graphics (gaming, professional visualization) and Compute & Networking (data center, automotive). As of FY2025, the Compute & Networking segment accounts for approximately 88% of total revenue — a complete reversal from its historical mix.

How the Company Makes Money

NVIDIA's business model is built around selling high-performance computing hardware (GPUs, networking chips) and the software ecosystem (CUDA, cuDNN, TensorRT) that makes them useful. The company sells directly to hyperscalers (Microsoft, Google, Amazon, Meta) and through OEM partners.

Revenue streams include: (1) Data center GPU sales to hyperscalers and enterprises, (2) Networking products (InfiniBand, Ethernet) via the Mellanox acquisition, (3) Gaming GPUs for consumer and professional markets, (4) Automotive compute platforms (DRIVE), and (5) Software and services (NVIDIA AI Enterprise).

Industry Overview

The AI infrastructure market is in the early stages of a multi-decade buildout. Hyperscalers are committing to $200B+ in annual capex, with a significant portion directed toward AI compute. IDC estimates the AI accelerator market will grow from $45B in 2023 to $300B+ by 2028 — a 46% CAGR.

The key structural driver is the shift from CPU-centric to GPU-centric computing. Training large language models requires massive parallel compute — exactly what GPUs are designed for. Inference (running trained models at scale) is now emerging as an equally large opportunity.

Revenue Breakdown

NVIDIA's revenue has undergone a dramatic transformation over the past three years. The data center segment, which was a secondary business as recently as 2021, now dominates the company's financials.

Competitive Advantages

NVIDIA's competitive position rests on three interlocking advantages that reinforce each other over time:

1. The CUDA Ecosystem: CUDA is a parallel computing platform and programming model that NVIDIA introduced in 2006. Over 18 years, millions of developers have learned CUDA, thousands of libraries have been built on it, and every major AI framework (PyTorch, TensorFlow, JAX) is optimized for it. This is not a technical moat — it's a social and economic moat.

2. Full-Stack Integration: NVIDIA sells not just chips but complete systems — DGX servers, networking (InfiniBand), storage (VAST), and software (NVIDIA AI Enterprise). This full-stack approach makes NVIDIA the default choice for enterprises that want a turnkey AI solution.

3. Architectural Cadence: NVIDIA has maintained a consistent 2-year architecture cycle (Ampere → Hopper → Blackwell → Rubin). Each generation delivers 2-3x performance improvements, making it economically rational for customers to upgrade continuously.

Growth Drivers

The next phase of NVIDIA's growth will be driven by inference demand. Training large models is a one-time cost; inference (running those models for billions of users) is a recurring, scaling cost. As AI applications proliferate, inference compute demand will grow faster than training demand.

Sovereign AI — governments building national AI infrastructure — is an emerging growth vector. Countries from India to Saudi Arabia are investing billions in domestic AI compute, and NVIDIA is the default supplier.

Financial Performance

NVIDIA's financial performance over the past two years has been extraordinary by any historical standard. Revenue grew from $26.9B in FY2023 to $130.5B in FY2025 — a 5x increase in two years. Gross margins expanded from 56.9% to 74.6%, reflecting the pricing power of a near-monopoly supplier.

Management

Jensen Huang, co-founder and CEO, is one of the most visionary technology executives of his generation. He has led NVIDIA for 30+ years and owns approximately 3.5% of the company. His long-term thinking — investing in CUDA when it had no clear commercial application, acquiring Mellanox for networking — has defined NVIDIA's competitive position.

The management team is deep and stable. CFO Colette Kress has been with the company since 2013. The company has a strong culture of technical excellence and long-term thinking.

Risks

The primary risks to our thesis are: (1) Geopolitical — US export controls on advanced chips to China have already cost NVIDIA significant revenue; further restrictions could accelerate Chinese domestic chip development. (2) Competition — AMD's MI300X is gaining traction in inference workloads; Google's TPUs and Amazon's Trainium are reducing hyperscaler dependence on NVIDIA. (3) Valuation — at 35x forward earnings, NVIDIA prices in significant continued growth; any deceleration would be severely punished.

Valuation Discussion

NVIDIA trades at approximately 35x FY2026 earnings estimates of ~$4.00 per share (post-split). This is not cheap in absolute terms, but it must be evaluated in the context of the growth trajectory and the durability of the competitive position.

Our base case assumes 40% revenue growth in FY2026, 30% in FY2027, and 20% in FY2028, with gross margins stabilizing at 73-75%. This yields a 3-year EPS CAGR of approximately 35%, implying a PEG ratio of ~1.0x — reasonable for a company with NVIDIA's competitive position.

Investment Thesis

We believe NVIDIA is the defining infrastructure company of the AI era, analogous to Cisco in the internet buildout of the 1990s — but with a more durable competitive position. The CUDA ecosystem, full-stack integration, and architectural cadence create compounding advantages that will be difficult to replicate.

The key question is not whether AI infrastructure spending will grow, but whether NVIDIA will capture the majority of that spending. We believe the answer is yes, for the next 3-5 years at minimum.

Key Takeaways