Bernstein Senior Analyst: The First True Chip Supercycle is Here, with 'Bottlenecks' as Wealth Creators

06/21 15:31

On June 21, Bernstein's star chip analyst Stacy Rasgon stated that this is the first time in his 18-year career that he has truly witnessed a semiconductor supercycle. With a PhD from MIT and an engineering background, Rasgon provided shocking data: the semiconductor industry surpassed $800 billion in total revenue last year and is racing towards $1.3 trillion this year, with all segments from accelerators to memory, semiconductor equipment, optical networking, power chips, and CPUs experiencing overwhelming demand. 'The only consensus we hear now is that no one's computing power is sufficient. For example, in memory, HBM in AI chips may occupy over 85% of the silicon area, and the silicon area required to manufacture 1GB of HBM is about four times that of standard DRAM, meaning that even if fabs ramp up production crazily, the actual increase in storage capacity remains very limited. This supply-demand mismatch has even benefited Intel—its inventory, which had been written down to zero, has been completely sold out, with customers saying they don't care and asking us to sell to them,' Rasgon pointed out. He noted that the industry's core focus is shifting from model training to AI inference, which is key to commercial monetization—training models themselves do not generate revenue; using the models does. Data from Anthropic shows that annualized revenue skyrocketed from about $9 billion in December last year to $30 billion in April this year, nearly a vertical increase. In the chip competition landscape, custom ASICs represented by Broadcom and NVIDIA GPUs are 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 about a dozen percentage points of the AI chip market revenue, which is expected to rise to 25%-30% in the future, but will not completely replace GPUs. Regarding inference chip startups like Groq, recently acquired by NVIDIA, Rasgon quoted 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 the focus from silicon to the physical world—electricity. It is estimated that if NVIDIA's projected annual infrastructure investment of $3 to $4 trillion comes true, the U.S. power grid would need to expand by about 5% each year, a task that analysts in the power industry see as nearly impossible. This means the next wave of bottlenecks will fall on energy generation, cooling, and nuclear power sectors, 'but never underestimate human creativity; engineers always find a way when there is profit to be made.' Regarding Intel, the pragmatic strategy of new CEO Pat Gelsinger with low expectations and the better-than-expected yield of the new 18A process, along with significant investments from the government and NVIDIA, have greatly alleviated previous market concerns about the balance sheet. Rasgon concluded that as long as AI demand does not collapse, the supercycle across the entire industry chain will continue, and the capital market's focus must closely follow the capacity bottlenecks that traverse various segments.

AlcistaBajistaMe gustaCompartir

Descargo de responsabilidad：El contenido anterior no representa las posiciones de HTX.，HTX no ofrece ninguna recomendación de trading.。