Broadcom to Help OpenAI Design AI Chips

Key Concepts:

• A6 (ASIC): Application-Specific Integrated Circuit, custom-designed chips for specific applications.
• Hyperscalers: Large cloud service providers (e.g., AWS, Google, Microsoft).
• Accelerator Landscape: The market for hardware and software designed to speed up specific computing tasks, particularly in AI and machine learning.
• GPU: Graphics Processing Unit, a specialized electronic circuit designed to rapidly manipulate and alter memory to accelerate the creation of images in a frame buffer intended for output to a display device.
• Unit Volumes: The quantity of individual units (e.g., chips) sold.
• Backlog: The total value of orders that have been received but not yet fulfilled.
• Sovereign Customers: Government entities or organizations with specific data sovereignty requirements.
• Neo Clouds: Emerging or smaller cloud providers.

Broadcom's Position in the Accelerator Market

• Dominant Share: Broadcom, alluded to by the speaker as having an "A-6 customer" (likely referring to a major hyperscaler), holds a significant position in the ASIC accelerator landscape. Before a recent announcement, they were the majority share owner, and their share is now expected to increase.
• Number Two: Broadcom is currently the number two ASIC company, trailing only Nvidia.
• Revenue Comparison: Broadcom's revenue in this area is approaching the entire data center revenue of AMD, the number two GPU provider. This highlights Broadcom's growing influence.
• Growth Drivers: Broadcom is experiencing growth by both acquiring new customers and increasing unit volumes with existing customers.

Broadcom vs. Nvidia: Competition or Coexistence?

• Nuanced Relationship: The speaker argues that the relationship between Broadcom and Nvidia is more complex than a simple competition. It's not a zero-sum game where Broadcom ASICs will "kill" Nvidia.
• Limited Use Case for ASICs: The primary use case for Broadcom's ASICs is limited to the top ten global hyperscalers.
• Compute Demand: The compute demand from these top ten hyperscalers is so high that it's not coming at the cost of Nvidia GPUs.
• Future Potential Competition: The speaker acknowledges a potential future scenario where hyperscalers, having improved their custom chips and software, could become sellers of compute based on their own chips. The IWC anthropic deal is cited as a case study.
• Enterprise and Sovereign Market: A significant market remains for enterprises, sovereign customers, and neo clouds that will continue to rely on chips like Nvidia's.

Backlog and Customer Growth

• OpenAI Backlog: A $10 billion future backlog from OpenAI is mentioned as a positive sign for Broadcom.

Conclusion

Broadcom is rapidly increasing its share in the ASIC accelerator market, driven by growth with both new and existing hyperscaler customers. While there is potential for future competition as hyperscalers develop their own chips, a large market will continue to exist for GPU-based solutions, particularly among enterprises, sovereign customers, and neo clouds. The relationship between Broadcom and Nvidia is therefore more of a coexistence than a direct competition, at least in the short to medium term.