Battle of the million GPU computing clusters! Broadcom predicts that three major cloud companies will deploy AI supercomputers with 1 million XPUs in the fiscal year 2027
Broadcom also expects that by 2027, the serviceable market (SAM) for AI XPU and networking will reach USD 60 billion to USD 90 billion, and the company is expected to capture a leading share in this market
The scale of AI supercomputers will reach one million GPUs.
Recently, Musk announced plans to expand xAI's Colossus AI supercomputer from the current 100,000 GPUs to one million, which has undoubtedly attracted widespread attention in the industry. However, xAI is not the only company with such grand plans.
During Broadcom's Q4 fiscal year 2024 earnings call, President and CEO Hock Tan stated:
“We currently have three hyperscale customers who have developed their own multi-generational AI XPU roadmap and plan to deploy at different speeds over the next three years. We believe that by 2027, each of them plans to deploy a million XPU clusters on a single architecture.”
According to reports, Broadcom develops AI and custom data center hardware chips for several tech giants, including Google and Meta. The company collaborates with customers to determine workload requirements, such as AI training, inference, or data processing, and then defines chip specifications and develops key differentiating aspects. Broadcom is responsible for implementing the architecture as silicon and equipping it with platform-specific IP, cache, chip interconnects, and interfaces. These high-performance XPUs are manufactured by TSMC.
Broadcom expects that by 2027, the serviceable available market (SAM) for AI XPUs and networking will reach $60 billion to $90 billion, and the company is poised to capture a leading share of this market. However, analysts point out that the statistical methods for market size are still unclear, and this year NVIDIA's revenue from sales of GPUs, DPUs, and networking hardware to the AI market is expected to reach approximately $100 billion