AMD Attempts to Break NVIDIA's CUDA Dominance

Undoubtedly, NVIDIA dominates both the software and hardware aspects of the generative AI industry. Its competitor, AMD, has been trying various methods to catch up. In terms of hardware, AMD has been striving to maintain its leading position through MCM design, and NVIDIA has recently decided to adopt this design as well.

However, in terms of software, AMD has recently made a significant leap forward through a major investment in the open-source AI software company, Nod.ai. Vamsi Boppana, Senior Vice President of AMD's AI division, said, "The acquisition of Nod.ai is expected to greatly enhance our ability to provide AI customers with open software that can easily deploy high-performance AI models tailored to AMD hardware."

For a long time, there has been a pressing need for an open-source GPU architecture. All hardware companies have been striving to build alternatives to narrow the gap with NVIDIA, including AMD and Intel.

For AMD, this deal may also be a more accessible and cost-effective way forward. Nod.ai, based in Silicon Valley, was founded in 2013 by Anush Elangovan and Harsh Menon and raised $20 million in its last funding round. It is reported that AMD may acquire this AI software company, which is currently valued at $36.5 million.

Betting on Open Source

Nod.ai has been dedicated to developing a range of tools and systems to drive AI applications on AMD hardware. The Nod.ai team has agreed to develop leading software for the Instinct data center accelerator, consumer-grade Ryzen AI processors, and Radeon GPUs.

Its biggest advantage is its open-source approach. According to reports, AMD is also committed to developing open-source solutions that lower the barrier of entry for customers, such as "developing tools, libraries, and models," as stated in its AI strategy manifesto. This is in stark contrast to the GPU-specific software developed by NVIDIA.

For a long time, NVIDIA's Compute Unified Device Architecture (CUDA) has been its biggest moat. The only problem is that it is closed-source and only applicable to NVIDIA's GPU workloads. Although people have been looking for several solutions to bypass this limitation, CUDA remains the best choice on NVIDIA GPUs, and the moat has become even larger as everyone is using its GPUs.

It can be said that software is the biggest moat for hardware companies. Bryan Catanzaro, Vice President of Applied Deep Learning Research at NVIDIA, said in an interview that although NVIDIA is known as a hardware company, "what a lot of people don't know is that NVIDIA has more software engineers than hardware engineers." Therefore, despite being closed-source, CUDA is still free to use.

However, AMD is persistent in open-source, which is also agreed upon by one of Nod.ai's founders, Elangovan. He said, "Our journey as a company will solidify our role as a major maintainer and contributor to some of the world's most important AI codebases, including SHARK, Torch-MLIR, and OpenXLA/IREE code generation technologies."

In addition, AMD's Radeon Open Compute (ROC) also increases the possibility of this partnership becoming an alternative to NVIDIA's moat.

A Hundred Billion-Dollar Competitor

AMD has been betting on software for some time now. In August of this year, the company also announced the acquisition of the French AI startup, Mipsology, which has also been a long-term partner of AMD, developing AI software for chip manufacturers, similar to Nod.ai.

Boppana wrote in August, "The team will help develop our complete AI software stack, expand our open software tools, libraries, and model ecosystems, and pave the way for simplified deployment of AI models running on AMD hardware."

The acquisitions of Nod.ai and Mipsology both highlight AMD's determination to challenge NVIDIA's monopoly in the AI market. Members of the Nod.ai team will also join Mipsology and the AMD AI group, which currently has around 1,500 employees.

Ultimately, AMD seems poised to break through NVIDIA's moat in the AI market.