Senior Analyst Bernstein: The first true chip supercycle is arriving, and "bottlenecks" are the wealth-creating machines

BlockBeats News, June 21 — Bernstein star chip analyst Stacy Rasgon recently stated that this is the first time in his 18-year career that he has truly witnessed a semiconductor supercycle. Rasgon, who holds a Ph.D. from MIT and is an engineer by background, presented data that is shocking: the total revenue of the semiconductor industry last year exceeded $800 billion, and this year it is rushing toward $1.3 trillion. All segments—from accelerators to memory, semiconductor equipment, optical communication networks, power chips, and even CPUs—are in full supply shortage. "The only consensus we hear now is—no one's computing power is enough. Take memory as an example, in AI chips, HBM may occupy over 85% of the silicon area, and manufacturing 1GB of HBM requires about four times the silicon area of standard DRAM, meaning that even if wafer fabs expand madly, the actual storage capacity increase remains very limited. This supply-demand mismatch even benefits Intel—its inventory, which has been written down to zero, is snapped up, and customer attitudes are: we don't care, just sell to us."

Rasgon pointed out that the industry's core focus is shifting from model training to AI inference, which is key to commercialization—training models itself doesn't make money; using models generates revenue. Data from Anthropic shows that annual revenue skyrocketed from about $9 billion in December last year to $30 billion in April this year, nearly a vertical leap. Regarding chip competition, custom ASICs represented by Broadcom and NVIDIA GPUs are not zero-sum games. "The correct pain point is whether the opportunity is still growing—if it's big enough, both can thrive." Currently, Broadcom expects AI revenue to reach $100 billion next year. ASICs account for about a dozen percentage points of revenue in the AI chip market, with potential to rise to 25%-30% in the future, but they will not fully replace GPUs. For inference chip startups like Groq, recently acquired by NVIDIA, Rasgon cited Jensen Huang's judgment: 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 most overlooked risk in the industry, Rasgon shifted focus from silicon wafers back to the physical world—electricity. According to estimates, if NVIDIA's predicted annual infrastructure investment of $3 to $4 trillion materializes, the U.S. power grid would need to expand by about 5% annually. However, power industry analysts see a 5% annual growth rate as nearly impossible to achieve. This means the next bottleneck will fall on energy generation, cooling, and nuclear power. "But never underestimate human creativity—when profits are involved, engineers will always find a way." Regarding Intel, Rasgon mentioned that the new CEO Chen Lihwu's pragmatic low-expectation strategy, the better-than-expected yield of the new 18A process, and government and NVIDIA's investments have significantly alleviated previous market concerns about balance sheets. Rasgon concluded that as long as AI demand doesn't collapse, the supercycle across the entire industry chain will continue, and capital markets should focus on capacity bottlenecks that are moving across various segments.