The spread of generative AI is redefining the global semiconductor supercycle. A cycle that had been a series of "short, sharp rebounds" has now shifted to a "large and deep upturn." That is because demand for the chips and memory semiconductors (hereafter memory) that support computation and inference is surging exponentially as artificial intelligence becomes ubiquitous. Whereas past semiconductor cycles were a byproduct of uncertainty stemming from mismatches between supply and demand, the current cycle is the result of an industry reshaping driven by the "rise of generative AI." As data center build-outs outpace memory's capacity to support AI inference, a worsening chip shortage is triggering price spikes.
Record first-quarter 2026 results by Samsung Electronics (hereafter Samsung), the No. 1 global memory maker, Taiwan Semiconductor Manufacturing Co. (TSMC), U.S. Micron, China's ChangXin Memory Technologies (CXMT), Taiwan's Nanya Technology, and other semiconductor corporations show that an AI-led industrial shift is accelerating. As the real economy undergoes AX (AI transformation), the semiconductor industry is shedding its dependence on the business cycle and evolving into one that accumulates structural profits.
"57 trillion by price" vs. "66% margin by structure"
Samsung's first-quarter 2026 revenue of 133 trillion won and operating profit of 57.2 trillion won are unprecedented in the history of Korean corporations. Its operating margin of 43% rivals that of global Big Tech such as Apple and Nvidia. As generative AI spreads, demand for graphics processing units (GPUs) and High Bandwidth Memory (HBM) has exploded, generating 54 trillion won—about 90% of operating profit—from the memory business alone. As demand for AI Semiconductor increased, shifting DRAM production to HBM sharply curtailed supply of commodity DRAM, and the resulting shortage led to massive excess profits. In fact, the fixed transaction price of DRAM, which was $1.35 in Jan. 2025, rose to $13 by the end of Mar., climbing more than tenfold in a year, and the price of DDR5 16Gb, the latest server DRAM, also jumped more than sixfold in half a year. On the back of soaring prices, Samsung's memory business operating margin surged above 70%. By contrast, the System LSI and foundry institutional sector, which concentrate design and custom manufacturing technologies, remain in the red and are struggling to secure profitability.
TSMC, with a 70% share of the global foundry market, posted net income up 58% year over year to 572.4 billion Taiwan dollars (about 24 trillion won), with a gross margin of 66.2%. A manufacturing company has effectively achieved a profit structure akin to a software company. Unlike Samsung, TSMC also grew 10%–30% quarter over quarter in high-performance computing (HPC), Internet of Things (IoT), and consumer electronics. In particular, the HPC institutional sector absorbed demand for AI servers and GPUs to become a core growth pillar. Most notably, it booked 74% of sales from advanced process nodes—3nm (25% of revenue), 5nm (36%), and 7nm (13%). TSMC reinvests about 8% of annual revenue in research and development to build a technological barrier to entry.
Semiconductor cycle shifts from technology to capital investment cycle
While Samsung's profits largely reflect excess earnings from price increases caused by memory bottlenecks, TSMC's profits are the product of structural competitiveness that combines advanced process nodes and packaging. If Samsung is enjoying a "price rent," TSMC has built a platform that creates profits by linking design, production, and packaging.
The market expects this semiconductor supercycle to last at least until 2027. Unlike the past "boom-bust" pattern that repeated every two years with the memory product development cycle, today the semiconductor business cycle is structurally lengthening in step with Big Tech's AI infrastructure investment. This year, AI investment by major Big Tech companies—Alphabet (Google), Amazon, Meta, and Microsoft (MS)—is expected to reach $700 billion (about 1,036 trillion won), up 69% from a year earlier. Accordingly, DRAM prices in the second quarter are projected to rise around 60% from the prior quarter. Supply bottlenecks are not expected to ease until around 2027, when capacity expansions by Samsung and SK hynix begin in earnest.
TSMC moves to preempt future demand vs. Samsung wrangles over bonuses
Responses diverge amid the historic boom. TSMC set this year's capital expenditures at up to $56 billion (about 82.9 trillion won). It is deploying more than half of its cumulative capex over the last three years ($100 billion) to preempt future demand. Expanding its cutting-edge 3nm lines to Arizona in the United States and Kumamoto in Japan, and ramping up next-generation packaging investments such as CoPoS (Chip-on-Panel-on-Substrate) following CoWoS (Chip on Wafer on Substrate), are strategies to strengthen its platform dominance that spans design, manufacturing, and packaging.
Samsung, by contrast, faces concerns that internal conflict could limit its ability to compete in the global AI power race. The Samsung Group supracorporate labor union has warned of a strike, demanding that 15% of 2026 operating profit (about 45 trillion won) be paid as performance bonuses. At a moment when profits from the memory boom need to be converted into structural competitiveness, a "bonus fight" risks sapping energy. Major foreign media, including Bloomberg and Reuters, said that could open opportunities for rivals. With the market shifting toward custom memory and Chinese memory makers in pursuit, time to execute a structural pivot is slipping away fast. AI is turning semiconductors from a "cyclical industry" into a "structural industry."