Part of The Install Base Is the Company
13 Platform Play

Put the Platform on Every Chip

Huang shipped CUDA on every consumer GeForce card, not just pro hardware: a cost that ate the company's profits for years and built its moat.

I always say that NVIDIA is the house that GeForce built - because it was GeForce that took CUDA out to everybody.

Huang describes the decision on the Lex Fridman Podcast (#494, March 2026, quoted here partly via Wccftech). Putting CUDA on every GeForce raised the cost of the chips, and it “completely consumed all of the company’s gross profit dollars.” He said “we increased our cost by 50%” at a time when “we were a 35% gross margin company,” and that “our market cap went down to like one and a half billion dollars.” He did not give a year for that last point. He called it “the first strategic decision that is as close to an existential threat.”

01 GeForce volume A consumer gaming franchise ships GPUs in huge numbers.
05 Funds the next bet Data-center revenue ($193.7B in FY2026) pays for the next architecture.
Install base
02 CUDA on every chip From 2006, every GeForce can run general-purpose code, at a cost Huang says consumed the company's gross profit.
03 Developers target CUDA Researchers can program the GPU already in their PC; AlexNet trained on two GTX 580s in 2012.
04 New workloads Deep learning and then generative AI arrive on hardware that is already everywhere.

The payoff. In 2012 AlexNet trained on two GTX 580 gaming cards and won ImageNet (15.3% versus 26.2% top-5 error): a university lab’s choice, not Nvidia’s plan.

Doesn’t transfer without a cash cow, or without luck. The bet was affordable only because gaming paid for it, and it paid off because deep learning happened to arrive on GPUs, which nobody, Nvidia included, was targeting in 2006. Skill kept CUDA alive through lean years; timing turned it into a fortune.