BlockBeats News, June 21st. Bernstein's star chip analyst Stacy Rasgon recently stated that this is the first true semiconductor supercycle he has witnessed in his 18 years in the industry. Rasgon, who holds a PhD from MIT and has an engineering background, provided staggering data: the semiconductor industry's total revenue surpassed $800 billion last year and is now sprinting towards $1.3 trillion this year. All sub-sectors, from accelerators to memory, semiconductor equipment, network optical communication, power chips, and even CPUs, are facing overwhelming demand. "The only consensus we hear now is that no one has enough computing power. Taking memory as an example, HBM in AI chips may account for over 85% of the silicon area, and producing 1GB of HBM requires silicon area approximately four times that of standard DRAM. This means that even if wafer fabs ramp up production insanely, the actual storage capacity increase remains very limited. This supply-demand mismatch has even benefited Intel—their previously written-off inventory has been snapped up, with customers saying, 'We don't care, just sell it to us.'"

Rasgon pointed out that the industry's core focus is shifting from model training to AI inference, which is key to achieving commercialization and monetization—training models themselves do not make money, but using models does. Anthropic data shows that annualized revenue skyrocketed from around $9 billion in December last year to $30 billion in April this year, almost a vertical takeoff. In the chip competition landscape, the competition between custom ASICs represented by Broadcom and Nvidia's GPUs is not a zero-sum game. "The right pain point is whether the opportunity is still growing—if it is large enough, both will thrive." Currently, Broadcom expects AI revenue to reach $100 billion next year, with ASICs accounting for a few percentage points of revenue in the AI chip market, expected to rise to 25%-30% in the future, but GPUs will not be completely replaced. For inference chip startups like Groq, recently acquired by Nvidia, Rasgon quoted Huang Renxun's opinion: "Not all tokens are the same; low-latency tokens are more valuable, and GPUs are not the optimal choice for all tasks."

When asked about the industry's most overlooked risk, Rasgon shifted the focus from silicon back to the physical world—electricity. According to estimates, if Nvidia's projected annual $3 to $4 trillion infrastructure investment materializes, the U.S. power grid would need to expand by about 5% annually. However, in the eyes of power industry analysts, a 5% annual growth rate is nearly an impossible task. This implies that the next bottleneck will lie in energy generation, cooling, and nuclear power, but "never underestimate human creativity; engineers can always find a way when it's profitable." Regarding Intel, new CEO Pat Gelsinger's strategy of setting low expectations, the better-than-expected yield of the new 18A chip process, and the significant investments from the government and Nvidia have greatly alleviated previous market concerns about the balance sheet. Rasgon concluded by emphasizing that as long as AI demand doesn't collapse, the full industry chain supercycle will continue, and the capital markets need to closely follow those capacity bottlenecks that exist throughout the chain.