The AI infrastructure buildout may be accelerating even faster than industry insiders anticipated. During Broadcom Inc. (AVGO)'s fiscal second-quarter earnings call, CEO Hock Tan revealed that compute requirements from OpenAI and Anthropic are already running ahead of expectations set just six months ago, offering a rare glimpse into the growing appetite for AI capacity among leading frontier-model developers.

According to Tan, demand measured in gigawatts of compute power continues to rise rapidly. "That amount of gigawatts required … is growing very fast," Tan said, referring to the compute capacity needed to support advanced AI models.

The surprise, however, was how much faster that growth is occurring.

"We're talking about capacity as measured by gigawatt power that are way ahead of what we have expected say six months ago," Tan added while discussing OpenAI and Anthropic.

Compute Demand Is Accelerating

The comments suggest that demand for AI infrastructure is not merely growing — it is accelerating.

Broadcom has become one of the key suppliers enabling that expansion through its custom AI accelerators and networking products. The company disclosed that AI semiconductor revenue reached a record $10.8 billion during the quarter and expects AI semiconductor revenue to reach approximately $56 billion for fiscal 2026.

Tan also indicated that the growth trajectory extends well beyond next year.

The Road To 2028 Growth

While discussing future demand, he noted that compute consumption from OpenAI, Anthropic and other customers, including Alphabet Inc. (GOOGL)'s Google, continues to expand and that the outlook for 2028 is becoming increasingly significant.

"If you ask about 27 or 28 that will continue to grow," Tan said. "We expect in fact 28 to be a substantial growth."

The remarks reinforce a theme that has emerged across the AI ecosystem: despite concerns that spending could eventually slow, the largest AI developers continue to signal a need for more compute, more power and more infrastructure.

For Broadcom, that demand appears to be arriving faster than expected. For investors, it raises a bigger question: if today's AI leaders are already exceeding infrastructure forecasts, how much larger could the AI buildout ultimately become?