NVIDIA CEO reports strong growth in T4 GPU sales as AI inference workloads accelerate across data centers.

Nvidia CEO Jensen Huang announced that the company sold more T4 units for inference than V100s for training for the first time, both reaching record highs. "This year we started to see the growth of inference to the point where we sold more T4 units for inference than we sold V100s, which we sell for training, and both were record highs," Huang said Thursday on Nvidia's third-quarter earnings call for its 2020 fiscal year. The T4 GPU launched in September of the previous year, and Huang stressed that inference "is really kicking into gear."

The Santa Clara, California-based chipmaker reported third-quarter revenue of $3 billion, a 5 percent decrease from the same period last year but a 17 percent increase from the previous quarter. Net earnings reached $1.78 per share, beating Wall Street expectations by 20 cents. The company's stock price was down more than 1.5 percent in after-hours trading despite beating revenue expectations by $90 million. For the fourth quarter, Nvidia expected revenue to reach around $2.95 billion. The company also said it could close its $6.9 billion acquisition of high-speed interconnect vendor Mellanox by the end of the year, though it is more likely to happen in early 2020.

Nvidia's data center business grew 11 percent sequentially to $726 million, driven by higher demand from hyperscale cloud service providers. However, a year-over-year decline of 8 percent was attributed to lower DGX sales and a different product mix. Huang explained that a major customer deal for the company's DGX deep learning system, built for training AI models, was a major factor in the year-over-year decline. Excluding that deal, he said sales for the V100 and T4 GPUs "are doing great." Driving the need for inference workloads are conversational artificial intelligence, recommendation engines and edge computing. Huang noted that "deep learning inference is understandably going to be one of the largest computer industry opportunities," adding that "the computational intensity is so high, and for the very first time, aside from computer graphics, this mode of software is not really practical without accelerators."

Nvidia's gaming GPU revenue reached $1.66 billion, marking a 26 percent increase from the previous quarter but a 6 percent decrease year-over-year. The sequential growth came from the company's GeForce desktop and notebook GPUs. According to CFO Collete Kress, RTX cards now represent two-thirds of Nvidia's desktop GPU sales, with nearly every major game developer pledging to incorporate ray tracing capabilities in their games. Huang expressed confidence in the technology, stating "I would say that at this point it's fairly clear that ray tracing is the future, and RTX is a home run."