TAIPEI -- With U.S. graphical chipmaker Nvidia set to roll out a development platform for self-driving technology in the latter half of next year, CEO Jensen Huang expects fully autonomous vehicles to become a mainstream reality no later than 2021.
"It will take no more than four years to have fully autonomous cars on the road," Huang told reporters in Taipei on Thursday. But he declined to predict when the majority of vehicles would become driverless.
Huang's comments came two weeks after Nvidia introduced its Drive PX Pegasus supercomputer, which it claims will help make possible "level 5," or fully autonomous, vehicles which can carry out all driving tasks on the road whether or not there is a human on board.
Working on self-driving technology with some 250 carmakers worldwide such as Japan's Toyota Motor, Germany's Audi and America's Ford, as well as ride-hailing companies, Huang said his company is well-positioned to have a central role in the vehicles of the future. Automakers are looking to Nvidia -- in which Japanese tech giant SoftBank Group reportedly holds a roughly 5% stake -- to run the data-intensive deep-learning calculations for autonomous driving, but many of these projects are still in the pilot stage.
Several carmakers and tech companies share Huang's view about the large-scale rollout of fully autonomous vehicles. Germany's BMW, which is working with Intel-controlled Israeli startup Mobileye on self-driving technology, has targeted putting fully automated automobiles into production by 2021. Elon Musk, CEO of America's Tesla, embraces an even more aggressive schedule -- he is looking to release self-driving cars under the electric-vehicle maker's brand by 2019. But Toyota is more skeptical. It sees self-driving vehicles on highways by 2020, and expects more advanced "level 4" cars to operate in specific areas within a decade.
Tesla, one of Nvidia's partners in self-driving technology, has reportedly turned to rival Intel for in-car infotainment solutions to reduce its reliance on the chip designer. But Huang said, "our focus for the auto industries is not car infotainment systems ... those are very easy to do and almost everyone can do them. We are very focused on autonomous vehicles. Computation there is very complicated and deep learning is very essential -- and we are very good in these areas."
Despite a growing number of competitors entering the artificial intelligence field, Nvidia has emerged as a tech superstar, with its flagship graphics processors are now widely used to accelerate sophisticated AI applications in data centers, smart cars, infrastructure and industry automation.
Shares of the U.S. chipmaker have nearly tripled over the past 12 months. For the second quarter ended in July, it reported record revenue of $2.23 billion, up 56% year on year. The company's data center sales shot up 175% from the previous year to $416 million, while its automotive business grew 19% on the year to $142 million.
But while leading tech firms such as Apple are adding more AI features such as facial and voice recognition for smartphones, Huang said AI chips for mobile and connected devices are not the markets into which Nvidia will venture.
With regard to Nvidia designing AI chips for mobile phones, Huang said, "I don't really see it. For these simple AI chips for cellphones and [Amazon's voice assistant] Alexa, that's not the area we will focus on." He said peers Qualcomm, NXP, Broadcom, Texas Instruments and others have already made the chip market for connected devices very crowded.
The executive said Nvidia will only concentrate on segments that require heavy computing workloads, foreseeing a rosy future for the company, especially in automated machine applications for smart cities and robotics.
Huang said his goal is to make AI computing more affordable. Years ago, processing 45,000 images per second would require a data center server to have 160 core processor units, or CPUs, according to Huang. Now, it only takes eight graphics processing units and one-sixth of the original costs to handle the same amount of data, he said.
"Artificial intelligence computation is expensive and we are soon to make it very affordable," Huang predicted.
