Advanced Micro Devices Corporation (AMD) has established close ties with a group of artificial intelligence start-ups as one of its measures to strengthen its software capabilities and create better chip designs. As AI companies seek alternatives to Nvidia chips, AMD has begun expanding its hardware product line planning, seeking to build a competitive hardware camp by acquiring companies such as server maker ZT Systems.

But to build a successful chip product line, it also needs to be equipped with powerful software to efficiently run the programs written by AI developers. In recent weeks, AMD has acquired a number of small software companies to expand its talent pool and continue to strengthen its software architecture, commonly known as ROCm.

“This is going to be a long-term, thoughtful, step-by-step journey for us,” said Vamsi Boppana, senior vice president of artificial intelligence at AMD.

AMD has committed to optimizing ROCm and other software, which is a boon for AI enterprise startups like Cohere because it means the software can be quickly iterated on and features added.

Cohere focuses on custom AI models for large enterprises, rather than the basic AI models that companies like OpenAI are targeting.

Cohere CEO Aidan Gomez told Reuters in an interview that AMD has made important progress on software improvements. He noted that migrating Cohere software to AMD chips used to take weeks, but now takes "a matter of days."

Gomez declined to say how much of Cohere's software relies on AMD chips, but said it was an "important component" of its global computing infrastructure.

OpenAI’s impact

AMD executive vice president Forrest Norrod revealed that OpenAI has had a significant impact on the design of the upcoming MI450 series of AI chips.

AMD's MI400 series of chips will form the basis of "Helios," a new server the company plans to launch next year. Nvidia has also previously designed overall server solutions, in part because AI computing requires hundreds or even thousands of chips to work together.

OpenAI CEO Sam Altman attended AMD's event in San Jose on Thursday and outlined the two companies' partnership.

Norrod said OpenAI's needs profoundly influenced AMD's design of the MI450 series memory architecture and how the hardware scales to thousands of chips to support building and running AI applications.

The ChatGPT developer also made suggestions for the types of math operations the chip should optimize.

“(OpenAI) gave us a lot of feedback, and I think that feedback greatly influenced our design,” Norrod said.