Jensen Huang personally recounts: After ten years of high-stakes CUDA gambling, NVIDIA nearly went bankrupt.

Ask AI · How Jensen Huang Withstood Pressure to Persist with CUDA Development?

IT Home, April 1 — Technology media Wccftech published a blog post yesterday (March 31), reporting that during the Lex Fridman podcast on March 29, NVIDIA CEO Jensen Huang revealed that pushing the CUDA architecture on GeForce graphics cards was a risky gamble that nearly bankrupted the company.

Huang said that in 2006, to change the outside world’s stereotypical view of NVIDIA as a “pure graphics company” and to transform into a comprehensive computing platform provider, he pushed forward with the underlying CUDA technology against all opposition. Today, it has become NVIDIA’s strongest moat leading the global AI race.

According to the blog post cited by IT Home, NVIDIA introduced programmable pixel shaders to break through the limitations of 3D graphics processing, opening the door to general-purpose computing.

The R&D team later solved the challenge of supporting FP32 single-precision floating-point calculations with programmable shaders, attracting many researchers and experts. They began experimenting with using GPUs for high-intensity computational tasks, thoroughly expanding the application boundaries of graphics cards.

Huang stated that pushing CUDA was a gamble that nearly bankrupted the company, causing devastating financial losses in the early promotion phase. At that time, the consumer market only cared about game rendering and didn’t care whether graphics cards could be programmed.

Forcibly embedding CUDA functionality into consumer-grade graphics cards caused NVIDIA’s production costs to surge by 50%, and overall gross profit margin plummeted to 35%. Due to the lack of immediate commercial returns, the capital market疯狂抛售, causing the company’s total market value to once drop to just $1.5 billion.

This daring gamble took 10 years to turn around. Huang and his development team endured immense pressure from Wall Street, holding on to maintain the massive software stack updates.

They firmly believed that this technology would eventually penetrate workstations and supercomputers, generating higher profits. Huang still feels grateful to the GeForce gaming community, explicitly stating that it was the large base of gamers that truly propelled CUDA technology worldwide.