Chip manufacturer United Microelectronics (NYSE: UMC) has reported strong quarterly results, with its board approving a capacity expansion plan. This includes adding cleanroom space at its large Singapore fab complex and constructing a new fab building at its flagship Tainan Science Park in Taiwan. The move is aimed at meeting the growing semiconductor demand driven by artificial intelligence computing needs.
The latest results and guidance from United Microelectronics represent a significant fundamental positive for the recently battered AI computing infrastructure chain. They notably weaken the bearish assumption that demand for AI computing infrastructure has peaked, as the strong results show orders are spreading from the most advanced AI computing chips to peripheral chips in mature process nodes, such as the power supply chain. However, some analysts more pessimistic about the semiconductor outlook believe that UMC's results cannot immediately erase market concerns about high leverage, overcrowded positions, high valuations, cyclical financing, returns on capital expenditure, and the rise of Chinese competitors. The Philadelphia Semiconductor Index has fallen about 25% from its June 22 high, entering a technical bear market. On July 28, the SMH (Semiconductor ETF) fell about 3.6%, and South Korea's Samsung Electronics and SK Hynix plunged 13.4% and 14.7%, respectively. AI-related semiconductor stocks in Asian and US markets again suffered a sharp selloff on Wednesday. South Korea's stock market triggered circuit breakers for a second consecutive day. On Wednesday, the benchmark KOSPI index once plunged over 12%, triggering a circuit breaker for the second day in a row, briefly falling below 5300 points, meaning it has dropped over 43% from its recent peak.
In the earnings statement, UMC CEO Jason Wang said: "The cleanroom capacity and capacity expansion plan will be executed in phases, allowing UMC to flexibly deploy capacity to meet customer AI infrastructure needs while maintaining capital expenditure discipline. Therefore, the capital expenditure budget for 2026 will be significantly increased to US$2 billion." Additionally, the UMC board has approved a capital expenditure budget of approximately US$5 billion for 2026 and 2027. Wang stated that the capacity expansion plan reflects the company's expectations for industry growth over the next five years. He noted that growth will be primarily driven by optical high-speed connections and power products for AI data centers, continued automotive electrification and intelligence, and emerging applications like humanoid robots. "These industry growth trends, coupled with our foray into advanced packaging, such as logic and memory stacking and silicon photonics, will accelerate the growth of UMC's addressable market," Wang said during the earnings call. The company expects AI-related revenue to reach about US$300 million this year and exceed the US$1 billion mark within three years.
Unlike the world's largest foundry and chip manufacturing giant Taiwan Semiconductor Manufacturing Co. (NYSE: TSM), which is heavily expanding capacity for leading-edge 3nm processes and investing heavily in advanced 2nm and even 1nm technology to support massive AI computing demand, United Microelectronics primarily focuses on more mature process nodes (from 14nm to 28nm and larger mature technologies). The division of labor between these two major Taiwanese chipmakers is fundamentally different. TSMC directly manufactures the 2nm, 3nm, 5nm, and other leading-edge processes required for AI accelerators (nearly all global AI chips), CPUs, and other core computing chips, and provides high-end advanced packaging capacity like CoWoS. In contrast, UMC's revenue from sub-14nm nodes was still zero in the second quarter, with its core capacity concentrated in 22/28nm and more mature nodes and specialty process technologies. UMC is not the primary manufacturer of NVIDIA's flagship AI GPU dies but focuses more on peripheral components critical for AI server clusters and edge AI systems, such as power management ICs, BCD chips, microcontrollers, sensors, network and RF connectivity chips, HDD and flash/DRAM controller chips, silicon photonics chips, interconnects, and backend processes for 2.5D/3D advanced packaging. From an AI computing infrastructure investment perspective, UMC represents not a "substitute for TSMC in advanced nodes" but a revaluation of the value of mature node capacity as AI computing power spreads from core processors to data center power, control, high-performance storage, and optical communication/interconnects.
AI Demand Spills Over to Mature Nodes: UMC Raises Capital Expenditure, Expands Dual Sites to Capture AI Demand
In terms of the latest quarterly results, driven by strong AI demand, United Microelectronics' second-quarter revenue was approximately NT$68.73 billion (US$2.12 billion), up 17% year-on-year. Net profit surged a record 374.7% to NT$42.26 billion. The continuous explosive growth in AI inference demand has driven a significant expansion in UMC's valuation and fundamental expectations. The company's stock has risen 120% year-to-date, significantly outperforming the broader Taiwan stock market benchmark, the Taiex Index, which gained 38.24%. The stock closed down 9.69% on the Taiwan stock exchange before the earnings release on Wednesday. UMC's US ADR (NYSE: UMC) has also fallen into a correction due to the recent global selloff in the AI computing and semiconductor sectors, dropping over 35% in July, but its year-to-date gain is still 125%.
UMC's core operating quality improved significantly in the second quarter: revenue was NT$68.733 billion, up 17% YoY; gross profit was NT$22.323 billion, up 32.3% YoY; operating profit was NT$14.950 billion, up 38.2% YoY, with the operating margin rising to 21.8% from 18.4% in the same period last year, and the gross margin rising to 32.5% from 28.7%. Wafer shipments reached 1.129 million 12-inch equivalent wafers, an increase of about 16.8% YoY, and capacity utilization rose from 76% to 85%, indicating that growth was primarily driven by actual shipments, higher utilization, and an improved product mix, rather than simple price fluctuations. It is important to distinguish that the net profit surge of 374.7% YoY to NT$42.260 billion included approximately NT$30.045 billion in investment gains, so this profit increase cannot be entirely extrapolated as sustainable operational growth. A more representative indicator is the 38.2% YoY increase in operating profit and the NT$23.97 billion in free cash flow in the second quarter.
UMC's forward guidance carries more industry signal significance than its historical profit figures. The company expects third-quarter wafer shipments to grow by a high single-digit percentage quarter-over-quarter, the average selling price in US dollars to remain firm, gross margin to rise to the mid-30% range, and capacity utilization to exceed 90%, compared to 85% in the second quarter. Capital expenditure for 2026 has been raised to US$2 billion from the original plan, a 25% increase from the actual US$1.6 billion expected in 2025, with 90% allocated to 12-inch capacity. This will fund the phased expansion of the Singapore P4 cleanroom and the new Tainan plant in Taiwan. UMC management also noted a strong recovery in demand for power management chips, sensors, and microcontrollers from 8-inch products. Revenue from 22nm reached a record 17.5% of quarterly revenue, and 12-inch silicon photonics chips have achieved initial volume production and delivery, with the silicon photonics platform set to open to a wider range of customers in 2027. This indicates that AI demand is shifting from advanced GPU wafer manufacturing and foundry to a broader "companion silicon computing chain" encompassing power, control, sensing, storage controllers, connectivity, and optical interconnects.
Fundamental Contradiction in the Chip Stock Crash: TSMC, SK Hynix, Seagate, and UMC Jointly Prove AI Demand Has Not Collapsed
Looking at the recent results and guidance from United Microelectronics, TSMC, SK Hynix, and Seagate together paints a clearer picture: the physical demand associated with AI computing infrastructure has not deteriorated in tandem with the sharp stock price declines and the liquidation of extremely leveraged positions. A recent research report from Wall Street financial giant Citigroup indicates that the AI arms race is shifting from "whose model is the smartest" to "who can produce intelligence at the lowest cost and highest efficiency under physical constraints." Open-weight models like Kimi K3 are rapidly approaching the frontier of closed-source models, suggesting model capabilities are accelerating commoditization. However, the simultaneous expansion of parameter scale, long context windows, and multi-step agent reasoning is moving the bottleneck from simple FLOPs to HBM capacity and bandwidth, high-speed GPU interconnects, cluster scheduling, and power access. Nvidia's own research also indicates that when model size, sequence length, and batch size increase, HBM often becomes the primary scaling constraint. IEA forecasts show that the electricity consumption of AI data centers is growing significantly faster than overall electricity demand, while grid construction cycles are generally longer than data center deployment cycles.
TSMC reported second-quarter revenue up 36% YoY and net profit up 77.4%, with high-performance computing accounting for 66% of revenue and 7nm and more advanced nodes accounting for about 77% of wafer-type revenue. The company guided third-quarter revenue between US$44.6 billion and US$45.8 billion, raised its full-year capital expenditure to between US$60 billion and US$64 billion, and increased its full-year US dollar revenue growth forecast to slightly above 40%. SK Hynix reported revenue up 257% YoY and operating profit up 557% YoY. HBM4 has begun volume shipments, and the company has signed long-term supply agreements with about 10 customers. Seagate Technology, the US-based HDD storage giant, saw its quarterly revenue rise 48.5% from US$2.444 billion to US$3.629 billion, non-GAAP gross margin jump from 37.9% to 52.7%, and free cash flow reach US$1.1 billion. The company provided guidance for the next quarter of US$4.1 billion in revenue and adjusted EPS of US$7.30. The simultaneous strength of advanced AI GPU/TPU/data center CPU logic chips, HBM/server DRAM/NAND, massive HDD storage, and mature and specialty process nodes constitutes a cross-validation of the "breadth of AI computing demand."
Therefore, the latest results and outlook from United Microelectronics represent a significant fundamental positive for the recently battered AI computing infrastructure chain. They significantly weaken the bearish assumption that demand for AI infrastructure has peaked, as the strong results show orders are spreading from the most advanced AI computing chips to peripheral chips in mature process nodes. However, the latest results from UMC and other leaders in the AI computing chain may not be sufficient on their own to reverse the deleveraging trend in global AI stocks. They cannot immediately eliminate market concerns about high valuations, cyclical financing, returns on capital expenditure, and competition from China. The most typical evidence is that even after posting record profits, SK Hynix's stock price still fell sharply due to not meeting extremely high expectations, and the KOSPI index dropped 12.6% at one point. Earlier, after TSMC reported significantly better-than-expected results, the global chip sector continued to face downward pressure. The current market is not trading on "whether there is AI demand," but rather on "what is the outlook for AI investment returns" and "whether the growth rate of AI computing demand can long-term cover AI capital expenditure, credit market financing costs, depreciation, and high valuations."
Comments