AMD CEO Lisa Su says the MI300 AI chips did over $1B in sales in Q2 and that supply of MI300 chips will remain tight through 2025
Sean Hollister / The Verge :
Context & Ripple Effects
MI300’s early sales milestone is an important marker in AMD’s AI-chip pivot: later coverage put AMD’s 2024 AI-chip revenue at $5 billion, versus $100 million in 2023, underscoring how quickly this product line became central to its effort to win AI-market share. AMD’s reported 2024 AI-chip revenue growth provides the longer arc for this quarter’s result.
The supply warning also foreshadows a business shaped by product transitions and market access. In 2025, AMD said its AI-chip revenue was affected by China export curbs and the move to MI350, showing that demand alone does not determine realized sales. The later MI350 transition and China restrictions add context to the earlier capacity constraint.
First-order effects
- AMD has established MI300 as a billion-dollar quarterly AI-chip business, while limited availability caps how quickly it can convert additional customer demand into shipments.
- Customers seeking MI300 capacity face allocation constraints through 2025, making delivery timing a near-term purchasing consideration rather than price or performance alone.
Second-order effects
- Tight supply shifts competition toward securing available accelerator capacity; customers unable to obtain MI300 on schedule may evaluate other hardware options or defer deployments.
- AMD’s AI revenue trajectory becomes more dependent on manufacturing and component availability, while its subsequent MI350 transition adds another variable to product revenue timing.
Third-order effects
- If sustained, accelerator shortages make supply access a durable competitive advantage in AI infrastructure, not merely a temporary sales constraint.
- AMD’s progression from MI300 to later generations suggests AI-chip competition will be judged increasingly on the ability to sustain product transitions while preserving customer supply.
The trend: This is an early data point in the AI-infrastructure supercycle, where accelerator demand and supply execution jointly determine chipmakers’ growth.