Broadcom expects to generate $230 billion in AI semiconductor revenue by fiscal year 2028, doubling from projected 2027 figures, CEO Hock Tan said during the company's third-quarter earnings call Wednesday. The Palo Alto, California-based vendor reported results for the quarter ending August 2, revealing that artificial intelligence-related spending continues to surge as hyperscalers and AI upstarts build out massive compute infrastructure. Tan said the company has secured sufficient supply to meet those revenue targets, driven by custom chips designed for OpenAI, Google, and Anthropic.
The vendor posted $29.6 billion in revenue for the third quarter, an 86 percent jump from the prior year. AI semiconductor sales reached $16.7 billion in the quarter, more than tripling year over year and growing roughly 50 percent from the previous quarter. Operating income hit $20.1 billion on a non-GAAP basis, nearly doubling annually, while free cash flow totaled $13.7 billion, representing 46 percent of revenue. Broadcom's XPU shipments grew more than threefold year on year and accounted for 73 percent of AI revenue during the quarter. The company's AI networking revenue more than doubled compared to the same period last year. For the fourth quarter, Broadcom projects $34.8 billion in revenue, a 93 percent year-over-year increase, with AI revenue expected to hit $21.7 billion—more than triple the prior year. The vendor now anticipates $58 billion in total revenue for fiscal 2026, up from an earlier $56 billion forecast, marking more than double the growth from fiscal 2025.
According to Tan, Broadcom has begun production shipments of Google's next-generation Tensor Processing Unit version 8i, which he said performs comparably to or exceeds Nvidia's Vera Rubin GPU. The CEO also announced shipments of OpenAI's first-generation custom accelerator Jalapeno, stating it "outperforms Grace Blackwell GPU for inference workloads" and can run frontier models at half the cost of a GPU. "The lesson here is when you co-develop a chip that is optimized for your particular LLM workloads, you will outperform any GPU," Tan said. The company plans to deliver 1.3 gigawatts of Jalapeno deployment in 2027, with OpenAI expected to become Broadcom's second-largest XPU customer. Anthropic is on track to become the largest XPU customer in 2027 and sustain that position in 2028, with Broadcom planning to deliver 1 gigawatt of Google's Ironwood seventh-generation TPU in 2026, followed by 5 gigawatts in 2027 and potentially another 10 gigawatts in 2028.
Tan explained that the company's financial outlook reflects constraints around data center site availability and volatile supply conditions, even as demand exceeds current projections. "We can ship significantly more," he told analysts, but deployment timing depends on customers' ability to install chips once delivered and the availability of memory supplies and data center infrastructure. The CEO described this as "a carefully structured outlook that we believe we can achieve," noting the company will adjust upward if supply circumstances improve. Broadcom partnered with Apollo and Blackstone to enable more than 20 gigawatts of compute infrastructure for OpenAI and Anthropic by the end of 2028, closing a $35 billion initial tranche in June for Anthropic's 1-gigawatt deployment already underway. The vendor is expanding production capacity across multiple facilities, including addressing substrate supply bottlenecks through a Singapore facility deployment in fiscal 2027 and increasing capacity for laser technologies over the next two years.
Tan characterized OpenAI and Anthropic as becoming "hyperscalers in their own right," seeking first-party compute capacity and cost-optimized silicon while operating their own data centers rather than relying solely on third-party cloud services. The vendor expects to ship $350 billion of AI semiconductors to major AI customers over the next two years, with the CEO anticipating that compute infrastructure needs will accelerate further in 2027 and 2028. Broadcom has sufficient AI semiconductor supply to double revenue to approximately $115 billion in 2027, Tan said, with actual demand surpassing the company's current guidance. The shift toward custom accelerators optimized for specific workloads positions Broadcom at the center of infrastructure buildout by both established cloud giants and emerging AI-native companies racing to deploy frontier models at scale. For solution providers, the vendor's channel goals include improving partner profitability and expanding professional services delivery through partners, particularly around VMware Cloud Foundation deployments.