AMD vs. Nvidia: The Battle for the Future of Artificial Intelligence with OpenAI's Support

On Thursday, one of the most important moments in the history of Advanced Micro Devices (AMD) took place in San Jose, California. CEO Lisa Su unveiled new details about the upcoming generation of Instinct MI400 AI chips, which will ship next year. Even more significant, however, was the presence of Sam Altman, CEO of OpenAI, who announced on stage that his company would use AMD chips. "When you first started telling me about the specifications, I thought, that's impossible, that sounds completely crazy," Altman said. "It's going to be an amazing thing."

This moment represents a major breakthrough for AMD in artificial intelligence, where Nvidia has so far dominated with a market share of more than 90%. OpenAI, which is known as a major Nvidia customer, is now providing AMD with feedback on its plans for the MI400 chips. This collaboration signals that AMD is finally gaining the confidence of leading AI players and could mark the beginning of genuine competition for Nvidia.

Revolutionary Helios Architecture

The key innovation introduced by AMD is the Helios system—a complete server rack that can connect thousands of MI400 chips so that they operate as a single unified system. "For the first time, we designed every part of the rack as one unified system," Lisa Su explained. This approach is critical for artificial intelligence customers, such as cloud service providers and companies developing large language models, which need "hyperscale" clusters of AI computers capable of spanning entire data centers.

Su described Helios as "a rack that operates as a single, massive computing engine" and compared it to Nvidia's Vera Rubin racks, which are scheduled for release next year. This "rack-scale" technology allows AMD's latest chips to compete with Nvidia's Blackwell chips, which are already available in configurations with 72 graphics processors linked together. Nvidia remains AMD's primary and only true rival in large data center GPUs for developing and deploying AI applications.

Competitive Advantage in Price and Power Consumption

Andrew Dieckmann, AMD's general manager for data center GPUs, confirmed to reporters on Wednesday that AMD chips would be cheaper both to operate and to acquire. "Across the board, there is a significant difference in acquisition costs, on top of which we layer our competitive performance advantage, resulting in significant double-digit percentage savings," Dieckmann said. AMD plans to compete with rival Nvidia through aggressive pricing and lower power consumption.

The company claims that its MI355X chips can deliver 40% more tokens—a measure of AI output—per dollar than Nvidia chips because they consume less power than their competitor's chips. Data center GPUs can cost tens of thousands of dollars per chip, and cloud companies typically purchase them in large quantities. Lisa Su also said that AMD's MI355X can outperform Nvidia's Blackwell chips, despite Nvidia using its proprietary CUDA software.

Current Offerings and Future Plans

Currently, the most advanced AMD AI chip installed at cloud service providers is the Instinct MI355X, which the company began shipping in production volumes last month. AMD announced that it would be available to rent from cloud service providers at the beginning of the third quarter. The MI355X has seven times the computing power of its predecessor and can compete with Nvidia's B100 and B200 chips, which have been shipping since late last year.

Companies building large data center clusters for AI want alternatives to Nvidia not only to keep costs low and provide flexibility, but also to meet the growing need for "inference"—the computing power required to actually deploy a chatbot or generative AI application. "What has really changed is that demand for inference has increased significantly," Su said. AMD believes its new chips are better for inference than Nvidia's chips because they are equipped with more high-speed memory, allowing larger AI models to run on a single GPU.

Major Customers and Market Prospects

AMD said that its Instinct chips have been adopted by seven of the ten largest AI customers, including OpenAI, Tesla, xAI, and Cohere. Oracle plans to offer its customers clusters with more than 131,000 MI355X chips. Meta representatives said on Thursday that the company uses clusters of AMD CPUs and GPUs to run inference for its Llama model and that it plans to purchase next-generation AMD servers. A Microsoft representative said that the company uses AMD chips to power its Copilot AI features.

Over the next several years, large cloud companies and entire countries are preparing to spend hundreds of billions of dollars building new data center clusters around GPUs to accelerate the development of state-of-the-art AI models. This includes $300 billion in planned capital expenditures this year alone by mega-cap technology companies. AMD expects the total AI chip market to exceed $500 billion by 2028, although it did not say what share of that market it could capture.

Investment and Future Vision

AMD has acquired or invested in 25 AI companies over the past year, Su said, including the acquisition of ZT Systems earlier this year—a server manufacturer that developed technology AMD needed to build its rack-sized systems. "These AI systems are becoming very complex, and full-stack solutions are truly critical," Su said. The Santa Clara-based company is combining its GPUs with CPUs and networking chips from its 2022 acquisition of Pensando to build Helios racks.

Both AMD and Nvidia have committed to releasing new AI chips annually rather than every two years, underscoring how fierce the competition has become and how important state-of-the-art AI chip technology is for companies such as Microsoft, Oracle, and Amazon. Despite these advances, AMD's AI chip business remains much smaller than Nvidia's. AMD said it generated $5 billion in AI revenue in its 2024 fiscal year, but JP Morgan analysts expect 60% growth in this category this year.