Bernstein analyst calls AI-driven chip boom first true supercycle in 18 years

Bernstein Analyst Calls Current Chip Boom First True Supercycle

Stacy Rasgon, a senior semiconductor analyst at Bernstein with 18 years covering the chip industry, declared on June 21 that the semiconductor sector is experiencing the first genuine “supercycle” of his career, as AI-driven demand pushes capacity constraints across every layer of the supply chain.

Revenue Racing Toward $1.3 Trillion

The numbers underscore the scale of what Rasgon is describing. The global semiconductor industry generated more than $800 billion in revenue in 2025 and is now on pace to reach $1.3 trillion in 2026, according to both Rasgon’s analysis and a Gartner forecast issued in April projecting 64% year-over-year growth — the strongest in two decades. Gartner expects AI semiconductors to account for roughly 30% of total semiconductor revenue this year, with memory making up half of the total.kucoin

“The only consensus we’re hearing is that no one has enough computing power,” Rasgon said. Every segment — from GPU accelerators and high-bandwidth memory to semiconductor equipment, networking, power chips, and CPUs — is experiencing severe supply shortages. The bottlenecks have spread sequentially, with each constraint resolved only to reveal the next one downstream.kucoin

HBM, Intel, and the Expanding Shortage

Rasgon highlighted high-bandwidth memory as a critical pinch point. HBM may account for more than 85% of the silicon area in AI chips, and producing 1GB of HBM requires roughly four times the silicon area of standard DRAM — meaning that even full-capacity fab expansion yields limited gains in actual storage output. The imbalance has been so acute that Intel, whose inventory was previously written down to zero value, has sold out entirely. “We don’t care — just sell it to us,” Rasgon said customers told the company.kucoin

In a May appearance on CNBC, Rasgon framed the dynamic more broadly: AI “has gotten so big, it is now dragging everything along with it… one at a time, all of these different parts of the industry have sort of become the constraint”. He noted that Nvidia is now generating roughly $20 billion in CPU revenue this year — comparable to Intel’s or AMD’s entire CPU businesses — driven by the 36 CPUs embedded in every 72-GPU rack.matterfact

Inference, ASICs, and the Energy Wall

Rasgon emphasized that the industry’s focus is shifting from model training to AI inference, which he called the key to commercial monetization. He pointed to Anthropic’s annualized revenue surging from approximately $9 billion in December to $30 billion by April as evidence that inference-driven applications are scaling rapidly.kucoin

On the question of whether custom ASICs from companies like Broadcom will displace Nvidia’s GPUs, Rasgon argued the two will coexist in an expanding market rather than compete in a zero-sum fight. The ultimate constraint, he concluded, is not silicon but energy — AI infrastructure growth now requires roughly a 5% annual increase in U.S. power grid capacity, pushing the next wave of bottlenecks into power generation, cooling, and nuclear energy.kucoin

“As long as AI demand does not collapse, the full-supply-chain supercycle will continue,” Rasgon said.kucoin