AMD and Broadcom Challenge Nvidia in AI Chip Market

The AI semiconductor market is splitting between model training and inference as Advanced Micro Devices (AMD) and Broadcom Inc. scale operations to challenge the dominance of Nvidia. While Nvidia maintains a lead in AI model training through its CUDA software platform, AMD and Broadcom are targeting the AI inference market, which is projected to reach $1.3 trillion by 2032.

AMD has focused on the inference sector by leveraging chiplet designs for high bandwidth memory and acquiring memory optimization firm MEXT and inference chipmaker Taalas. The company has secured multiyear, revenue-generating partnerships with OpenAI, Meta Platforms, Anthropic, and Microsoft, including deals with OpenAI and Meta valued at $100 billion.

Broadcom Inc. is capitalizing on the demand for application-specific integrated circuits (ASICs) from hyperscalers. The company helped Alphabet Inc. develop Tensor Processing Units (TPUs) and has partnered with Anthropic, Meta, and OpenAI to design custom AI chips. Broadcom projects its AI revenue will grow to $115 billion in fiscal 2027 and $230 billion in fiscal 2028.

Despite this competition, Nvidia continues to secure major infrastructure wins. Elon Musk recently announced that Space Exploration Technologies Corp. will exclusively use Nvidia chips moving forward, an agreement expected to generate hundreds of billions of dollars in revenue for the GPU manufacturer.