Thanks to the boom in GenAI, AI has experienced explosive growth in just three years and now accounts for half of global system revenue. So it makes sense that custom ASIC design and AI-related networking businesses will eventually dominate Broadcom's semiconductor business.
AI business leaership
AI-driven business dominance
Thanks to the GenAI boom, AI has experienced explosive growth in just three years and now accounts for half of global system revenue. It makes sense, then, that the custom ASIC design business and AI-related networking business will eventually dominate Broadcom's semiconductor business.
In the second quarter of fiscal 2025, which ended in early May, Broadcom's AI-related semiconductor share exceeded 50% of sales for the second consecutive time, and now it's reasonable to wonder when that share might break 75% given the focus on high-end networking and the growing popularity of Ethernet switching in AI workloads - soon not only for scale-out networks connecting cluster nodes, but also for scale-up networks connecting XPUs.
Specifically, Broadcom's sales in the May quarter were just over $15 billion, up 20.2% year-over-year. Operating profit nearly doubled to $5.83 billion, driven by cost cuts in its VMware software division and adoption of its refocused and simplified VCF product line by more than 8,700 of its 10,000 largest VMware customers. Broadcom's Infrastructure Software Group (which includes product lines such as VMware, CA and Symantec) is smaller than its Semiconductor Solutions chip business, but its operating profit margin was as high as 76% in the past two quarters, while the chip business's operating profit margin was 57% in the past two quarters. The chip business is one-third the size of Broadcom.
Outstanding software contribution
Financial relief for software businesses
Broadcom's legacy software business, which it has built up over the past few decades, has somewhat relieved pressure on its chip business, which has been under tremendous pressure. By partnering with hyperscalers and cloud service providers to make its own host CPUs and AI XPUs, Broadcom can benefit from the risk diversification and cost savings sought by these tech giants, as they can create and install computing engines of their own design without having to compete directly with Nvidia, AMD, and Intel.
It's hard to imagine anyone running the company better than Broadcom CEO Hock Tan. Broadcom's operating mix may seem a little odd from a technical perspective, but it works well from a financial perspective. Net profit, for example, grew 2.3 times to $4.97 billion, or 33.1% of revenue. Net profit as a percentage of sales is now almost the same as it was before Hock Tan (in May 2023) decided to acquire VMware for $61 billion and restructure it to create more profits.
VMware has been included in Broadcom's accounts since the first quarter of fiscal 2024 (ending in February 2024). We estimate that the server virtualization giant has contributed $21.71 billion in revenue to Broadcom and $14.59 billion in interim profits. That's about half of Broadcom's chip business revenue and operating profit. At this rate, VMware has recovered about one-fifth of its $61 billion cost in operating profit in just six quarters. VMware will be able to recoup the cost of the deal within Broadcom in less than three years.
Broadcom's infrastructure software division sales for the May quarter were $6.6 billion, up 24.8% year-on-year but down 1.6% from the previous quarter. Operating profit increased 58.7% year-on-year to $5.03 billion, but fell 1.4% from the previous quarter.
Strong future growth
Focus on growth forecasts and customized trends
Broadcom's Semiconductor Solutions segment is more important to The Next Platform, as Broadcom manufactures and sells networking and storage chips to third parties. Sales in this chip business were $8.41 billion, up 16.7% year-over-year and even slightly up from the first quarter of fiscal 2025.
In the chip business, our model shows that Broadcom's AI chip revenue (including switch ASICs, XPU ASIC design, packaging, chip support for hyperscale computing platforms and cloud service providers, and other chip businesses) grew 46.7% to $4.42 billion, while other chip sales fell 4.8% to $3.99 billion. Broadcom has three custom compute engine customers—Google was the first, but now Meta Platforms and OpenAI are working with Broadcom on chips—and four potential customers—which we believe include Apple, ByteDance, and two others—who are looking to Broadcom as a backer of XPUs.
On the AI sales side, based on Tan & Co's comments on the data on a conference call with Wall Street analysts, we see Broadcom's AI compute sales at $2.65 billion, up 34.5% year-over-year and 7.2% quarter-over-quarter.
AI networking is just starting to take off, with Jericho-3AI and now Tomahawk 6 switch ASICs, which saw sales increase 67.7% to $1.77 billion, up 7.1% from the previous quarter.
As we previously reported, Broadcom's AI chip sales are $3.8 billion in fiscal 2023, and by fiscal 2024, that number increases 3.2 times to $12.2 billion.
Broadcom doesn't reveal much detail on sales for its various divisions, but we think it's as follows:
In our Broadcom model, networking sales increase $33.7 billion to $5.121 billion, while server storage connectivity sales increase 22.6% to $1.01 billion.
Looking ahead to the third fiscal quarter, which will end in early August, Broadcom expects sales to reach $15.8 billion, an increase of about 21%. Chip sales will reach about $9.1 billion, an increase of 25%. If we look closely at the AI and non-AI business forecasts for Broadcom's semiconductor division, the AI computing business is expected to grow by about 42% to about $3.1 billion, and the AI network business will nearly double to about $2.1 billion. This means that Broadcom's total AI revenue in the third fiscal quarter will reach $5.16 billion, and its total AI revenue for the full year of fiscal 2025 is expected to reach about $19 billion, or even higher. Chen Fuyang said that Broadcom is expected to grow this business by about 60% to about $30 billion in fiscal 2026.
Interestingly, technology giants are actively developing their own AI training ASICs and AI inference ASICs - it's not just about training, but also increasingly focusing on inference. Tan explained on a call with Wall Street analysts:
"I think there is no difference between using commercial accelerators and custom accelerators for training and inference. I think the whole premise of moving to custom accelerators still exists, it's not just about cost. As custom accelerators are used and grow on any particular hyperscale computing platform, they will go through a learning curve to learn how to optimize the way large language model algorithms are written and tied to silicon.
This ability has a huge added value in creating algorithms that can drive continuous improvement in LLM performance, far beyond simply separating hardware and software. As they go on this journey, you are actually combining hardware and software end to end. It's a journey. They can't learn this in a year. It takes several cycles of practice to get more and more proficient. That's the value - the fundamental value of creating your own hardware compared to using third-party commercial silicon is that you can optimize the software for the hardware and ultimately achieve performance far beyond what you would otherwise do. We see this happening."
The question is: when will this phenomenon of custom silicon surpassing commercial silicon happen with the switch ASICs that Broadcom is so keen to sell?
Reference link https://www.nextplatform.com/2025/06/09/broadcom-at-the-crossroads-between-merchant-and-custom-silicon/
