SAN FRANCISCO – Nvidia earned its US$2.2 trillion (S$3 trillion) market capitalization by making artificial intelligence chips that have become the lifeblood powering a new era of generative AI developers from start-ups to Microsoft, OpenAI and of Google’s parent Alphabet.
Almost as important to its hardware is nearly 20 years worth of the company’s computer code, which makes competing with the company nearly impossible. More than four million global developers rely on Nvidia’s Cuda software platform to build AI and other applications.
Now a coalition of tech companies that includes Qualcomm, Google and Intel plans to loosen Nvidia’s stranglehold by going after the chip giant’s secret weapon: the software that keeps developers tied to Nvidia’s chips.
They are part of a growing group of financiers and companies hacking Nvidia’s dominance in artificial intelligence.
“We’re actually showing developers how to migrate from Nvidia’s platform,” said Dr. Vinesh Sukumar, Qualcomm’s head of AI and machine learning, in an interview with Reuters.
Starting with a piece of technology developed by Intel called OneAPI, the UXL Foundation, a consortium of technology companies, plans to build a suite of software and tools that can run multiple types of chips to accelerate artificial intelligence, executives involved in the group told Reuters.
The open source project aims to enable computer code to run on any machine, regardless of the chip and hardware running it.
“It’s specifically about — in the context of the machine learning framework — how to create an open ecosystem and promote productivity and choice in hardware,” Google director and chief technologist for high-performance computing Bill Magro said in an interview with Reuters.
Google is one of the founders of UXL and is helping to determine the technical direction of the project, Dr. Magro said.
In addition to the initial companies involved, UXL will court cloud computing companies such as Amazon.com and Microsoft’s Azure, as well as additional chip makers.
Since its launch in September, UXL has already begun receiving technical contributions from third parties that include foundation members and outsiders eager to use the open-source technology, executives involved said.
Intel’s OneAPI is already usable, and the second step is to create a standard programming model of computing designed for AI.
UXL plans to focus its resources on solving the most pressing computing problems dominated by a few chipmakers, such as the latest AI applications and high-performance computing applications.
Those early plans impact the organization’s longer-term goal of gaining a critical mass of developers for its platform.
UXL ultimately aims to support Nvidia hardware and code, long term.
Asked about open source software and venture funding efforts to break Nvidia’s AI dominance, Nvidia CEO Ian Buck said in a statement: “The world is accelerating. New ideas in accelerated computing are coming from across the ecosystem, and this will help advance artificial intelligence and the scope of what accelerated computing can achieve.”
The UXL Foundation’s plans are one of many attempts to chip away at Nvidia’s influence on AI-powered software.
Venture capital financiers and corporate dollars poured more than US$4 billion into 93 separate projects, according to adjusted data compiled by PitchBook at the request of Reuters.
Interest in ousting Nvidia over potential software weakness surged last year, with start-ups aiming to plug holes in the company’s leadership gobbling up just over $2 billion in 2023, compared with $580 million a year ago, according to data from PitchBook.
Succeeding in the shadow of Nvidia’s AI data crunching group is an achievement few start-ups will be able to achieve.
Nvidia’s Cuda is a compelling piece of software on paper, as it has all the features and is constantly growing thanks to contributions from Nvidia and the developer community.
“But that’s not what really matters,” said Mr. Jay Goldberg, CEO of D2D Advisory, a finance and strategy consulting firm. “What’s important is the fact that people have been using Cuda for 15 years, they’ve built code around it.” Reuters