AMD unveiled its upcoming Instinct MI400 series AI chips at a launch event in San Jose, California, on June 12, 2025.

The chips are scheduled to ship next year and will include a rack-scale system called Helios, which allows thousands of chips to function as a unified system.

The MI400 chips aim to compete with Nvidia’s AI offerings, including its Blackwell chips.

They are designed to provide alternatives for cloud providers and companies developing large AI models.

OpenAI CEO Sam Altman confirmed at the event that OpenAI plans to use AMD’s chips.

He also provided feedback to help refine the MI400 roadmap.

🔗 Source: CNBC

🧠 Food for thought

1️⃣ AMD’s pricing strategy targets Nvidia’s premium margins

AMD’s approach of offering “aggressive” prices reflects a consistent strategic pattern in the competitive semiconductor landscape.

The pricing differential is substantial. AMD’s MI300 AI accelerators cost between $10,000-$15,000, while Nvidia’s comparable H100 GPUs often exceed $40,000, representing a 3-6x price difference 1.

This pricing strategy targets Nvidia’s extraordinary 78% gross margins, which far exceed AMD’s 47% and Intel’s 41%, highlighting the premium pricing power Nvidia has established in the AI market 2.

AMD’s deliberate undercutting mirrors classic semiconductor competition tactics, where challengers typically compete on price while market leaders maintain premium positioning based on performance advantages and ecosystem lock-in.

The emphasis on “meaningful cost of acquisition delta” and “significant double-digit percentage savings” from AMD executive Andrew Dieckmann directly addresses the economic pressures faced by AI infrastructure buyers who are collectively spending hundreds of billions on AI hardware.

2️⃣ The battle extends beyond hardware to software ecosystems

Lisa Su’s statement that AMD chips can outperform Nvidia “despite Nvidia using its proprietary CUDA software” highlights the dual-front competition in AI chips.

Recent Advanced Micro Devices developments