Session 19 · Nvidia: Managing Technology & Innovation

One company controls ~98% of a market
everyone agrees is about to get crowded.

Nvidia's data-center GPUs power almost every serious AI model built since 2023. Google, Amazon and Microsoft are each racing to build their own chips. So: is Nvidia's dominance a durable technology advantage, or a temporary lead about to be commoditized?

The Case

Yoffie, D., & von Bargen, N., "Nvidia, Inc. in 2024 and the Future of AI," Harvard Business School Case 9-725-360. FY2024: revenue of $60.9B, data-center revenue of $47.5B, gross margins near 72.7% — extraordinary numbers for a hardware company, which is exactly the puzzle. Hardware businesses are supposed to be capital-intensive and thin-margin. Nvidia's numbers look like a software monopoly's.

First Principle Question

If a rival built a chip with identical raw performance to Nvidia's tomorrow, would that alone be enough to take Nvidia's customers? What would still be missing?

The Real Product Isn't the Chip

CUDA: The Moat Nobody Can See in a Spec Sheet

First Principle Question

Every serious rival chip is now "good enough" on raw compute. Why hasn't that ended Nvidia's pricing power?

CUDA's developer ecosystem

Over 5 million developers have been trained on Nvidia's CUDA software platform since its 2006 launch. Every AI framework (PyTorch, TensorFlow) is optimized for CUDA first. A rival chip with better silicon still forces every engineering team to rewrite and re-validate its entire software stack — a switching cost that has nothing to do with the chip itself.

The DeepSeek shock, January 2025

Chinese lab DeepSeek reportedly trained a GPT-4-class model for under $6M, versus $100M+ for Western labs — briefly wiping out roughly $600B of Nvidia's market cap in a single day on fears that AI would need far less compute than assumed. The moat didn't disappear; the market re-priced the *size* of the opportunity, not Nvidia's share of it. Worth separating those two questions whenever a "disruption" headline hits a dominant firm's stock.

First Principle Question

Is Nvidia's Position Sustainable?

Weigh the two forces working against each other: Nvidia's ecosystem lock-in versus its biggest customers' incentive to build their own silicon.

Google (TPU), Amazon (Trainium/Inferentia) and Microsoft (Maia) are all Nvidia's biggest customers and its most credible long-term threats — the textbook definition of a buyer with enough scale to threaten backward integration. The question for class discussion: does CUDA's switching cost outlast their patience?

First Principle Question

Check Your Understanding