Intel, Google, Arm, Qualcomm, Samsung and other technology companies have formed an alliance called the UXL Foundation in an effort to develop an open source software suite to free AI developers from a long-standing situation of being dependent on Depends on NVIDIA chip. In other words, the UXL alliance wants to find the key that allows developers' programs to operate on any machine and is not "picky" about the chip used.
NVIDIA's Achilles heel?
According to AFP, NVIDIA has become the most valuable chip manufacturing corporation in the world with a total market capitalization of up to 2.200 billion USD thanks to the provision of AI chip lines. These are the types of chips that create the era of generative AI developers, a form of artificial intelligence that focuses on creating new content and data based on existing data.
Chips from American corporations are currently "sought after" all over the world, whether they are startups or long-standing technology giants such as Microsoft, Google, or OpenAI, the father of ChatGPT. . Despite the US embargo, many Chinese companies still seek to buy NVIDIA chips.
Besides hardware, part of NVIDIA's formidable power comes from a computer code called CUDA that has been researched and developed for nearly 20 years. This makes NVIDIA almost in an unbeatable position. More than 4 million developers globally rely on NVIDIA's CUDA software platform to build AI and other applications.
NVIDIA strengthens its dominance with AI super chips
Now, a coalition of technology companies such as Qualcomm, Google and Intel, is planning to loosen NVIDIA's grip on the market. They are part of a group that continues to grow and includes financiers and companies looking to break away from NVIDIA's dominance in AI. The goal is to focus on attacking the chip giant's secret weapon: CUDA software, which forces developers to use NVIDIA chips.
“We have shown developers how they can separate from the NVIDIA platform,” Reuters yesterday quoted Mr. Vinesh Sukumar, Director of AI and machine learning at Qualcomm.
New weapon
Starting with a technology developed by Intel called OneAPI at its founding in September 9.2023, the UXL alliance plans to build a suite of software and computing tools that can go on to power a variety of AI chips. different. According to Reuters, this open source project aims to allow companies' computer code to work on any machine, no matter what chip or computer hardware they are powered by.
“Within the machine learning framework, we want to create an open source ecosystem, and promote productivity and enable choice in hardware,” said Bill Magro, director and chief engineer of the computer performance. Google is one of the founding members of UXL and helps determine the technical direction of the project.
Intel spends 100 billion USD for the throne of the semiconductor industry
UXL's technical steering committee is preparing to list specific technical specifications for the first half of this year. And the alliance's engineering team plans to complete the technical details to reach a state of completion by the end of 2024. In addition to the initial companies, UXL will continue to recruit names in the industry cloud computing such as Amazon, Azure, as well as other chip manufacturers.
Intel's OneAPI is now in use, and the next step is to create a standard computing programming model designed for AI. Not stopping at the goal of confronting NVIDIA, UXL in the long term also wants to support this corporation in both hardware and computer code issues.
NVIDIA competes with nearly 100 startups
UXL's plan is just one of the efforts to compete with NVIDIA's market share in the AI field. Reuters cited data collected by PitchBook Company (USA) showing that many venture capital fund sponsors and other companies have spent more than 4 billion USD on 93 different startups. Their goal is to "overthrow" NVIDIA through exploiting software weaknesses. And they achieved initial success. PitchBook noted that the above startups earned more than $2 billion in profits in 2023 exploiting NVIDIA vulnerabilities compared to $580 million the previous year.