semiconductor

Analysis of the AI semiconductor and infrastructure cycle indicates that the demand for AI infrastructure will continue to exceed that of the internet era, as user penetration and per capita token are multiplied and converted into tokens, significantly raising the ceiling. User penetration has surpassed 50%, with growth primarily coming from the still-early per capita token; the median monthly AI spending per employee in U.S. companies is about $12, which could long-term approach around 10% of white-collar salaries, approximately $1,000 per month, leaving nearly two orders of magnitude of space in between. Unlike the flat subscriptions and extremely low marginal hardware consumption of the internet, the high costs of inference make the marginal cost of a single access higher, requiring greater infrastructure intensity.After developers program, the next ARR growth will still mainly come from broad programming: non-programmers use programming infrastructure to complete non-programming tasks across industries, with programming becoming the default execution kernel for agents. Tasks related to broad programming account for about 60% to 70% of ARR. As of June 2026, Codex accounted for 64% of the total output tokens from Codex and ChatGPT among OpenAI's enterprise clients; since February, Codex has seen a much higher weekly growth in verticals such as law, sales recruitment, and marketing compared to engineering. In Anthropic's revenue, narrow development/software accounts for about 40%, while finance and insurance exceed 20%, with law, life sciences, retail, and others also having considerable shares.The demand-side token growth logic remains, with a high overlap between funders and beneficiaries.

SMIC (Hong Kong Stock 00981, A Share 688981) released its semi-annual report for 2026. The reporting period is from January 1, 2026, to June 30, 2026, and is unaudited. The company achieved operating revenue of 38.635 billion RMB, a year-on-year increase of 19.4%; total profit of 7.133 billion RMB, a year-on-year increase of 96.7%; net profit attributable to shareholders of the listed company of 4.467 billion RMB, a year-on-year increase of 94.2%; net profit attributable to the parent company after deducting non-recurring gains and losses of 2.994 billion RMB, a year-on-year increase of 57.2%.During the same period, the net cash flow generated from operating activities was 20.760 billion RMB, a year-on-year increase of 252.0%; earnings before interest, taxes, depreciation, and amortization (EBITDA) was 24.511 billion RMB, a year-on-year increase of 40.7%. The gross profit margin was 23.2%, an increase of 1.3 percentage points compared to the same period last year; the net profit margin was 16.7%, an increase of 6.3 percentage points; basic and diluted earnings per share were both 0.56 RMB, a year-on-year increase of 93.1%. The board of directors approved the submission of this report on August 27, 2026, and the company has no profit distribution or capital reserve transfer plan.

According to eight people familiar with the discussions, despite warnings from tech companies that this move could affect the United States' dominance in the field of artificial intelligence, the Trump administration is still considering imposing a new round of comprehensive tariffs on semiconductors.

According to SemiWiki, WSTS data shows that the global semiconductor market sales reached $368 billion in the second quarter of 2026, a quarter-on-quarter increase of 35% and a year-on-year increase of 104%, setting a new historical high. The growth is mainly driven by AI demand, with significant quarter-on-quarter increases from memory manufacturers, SK Hynix and Sandisk at 51%, Kioxia at 76%; Nvidia's AI processor business grew 18% quarter-on-quarter, and Broadcom is expected to grow 33% quarter-on-quarter.Kioxia and Sandisk rose to eighth and ninth place globally, respectively, with both companies' revenues growing more than four times year-on-year. Excluding major AI suppliers, the other top 20 semiconductor manufacturers only saw a quarter-on-quarter increase of 9%, with guidance indicating a total quarter-on-quarter growth of about 6% for the third quarter of 2026. The growth expectations for memory manufacturers in the third quarter have clearly slowed down.Based on the latest forecasts, the semiconductor market growth rate for the entire year of 2026 is expected to reach or exceed 90%, with Semiconductor Intelligence predicting a growth of 101% and RCD Advisors predicting 112%; in the storage sector, WSTS expects a growth of 249%, while Gartner predicts 280%. Most institutions expect the growth rate to fall back to the mid-20% range in 2027.

According to Nikkei Asia, storage chip manufacturer Kioxia Holdings plans to build new production facilities at its factory in northern Japan, with project investments expected to exceed 1 trillion yen (approximately 6.27 billion USD).Kioxia currently ranks third in the global NAND flash memory market, behind Samsung Electronics and SK Hynix. The company's move aims to follow its South Korean competitors in increasing production capacity to meet market demand.

According to Gate market data, after-hours trading on Wednesday for U.S. stocks saw Nvidia's new quarterly results continue to exceed market expectations, with a forecast of approximately 70% sales growth for the next fiscal year, driving AI concept stocks higher in after-hours trading.The storage sector collectively strengthened, with SK Hynix rising 4.48%, SanDisk up 3.71%, Micron Technology up 3.67%, Seagate Technology up 3.15%, and Western Digital up 3.11%. The Neocloud sector saw widespread gains, with Nebius up 6.27%, CoreWeave up 4.76%, Applied Digital up 3.26%, and both IREN and Core Scientific up 3.16%.Optical communication stocks rose across the board, with Applied Optoelectronics up 4.25%, Marvell Technology up 4.08%, Credo up 3.78%, Coherent up 3.35%, and Astera Labs up 3.34%. Semiconductor stocks also strengthened, with Arm up 4.18%, Marvell Technology and Nvidia both up 4.08%, Lam Research up 2.65%, and KLA up 2.42%.

On August 25, Bank of America data showed that actively managed long-term funds significantly reduced their holdings in global semiconductor stocks last month, selling approximately $44.4 billion, indicating that institutional funds are withdrawing from the most crowded AI trades. The flow of funds has shifted towards telecommunications, energy, materials, and grid modernization, reflecting a more noticeable redistribution within the AI theme.This set of data explains part of the recent market pressure. Ahead of NVIDIA's earnings report, the market still holds high expectations for AI demand, but semiconductor stocks have previously risen too much, and positions have become more concentrated. Once long-term interest rates rise, AI revenue expectations cool, or the return on capital expenditures for cloud vendors is questioned, the semiconductor sector will be the first to bear the pressure of position reduction. Bank of America also pointed out that the themes that funds sold the most in the past year include AI computing and quantum computing, indicating that funds have not completely left AI but are reducing exposure to highly crowded sectors.Bank of America expects that semiconductor stocks will continue to be influenced in the short term by NVIDIA's performance, cloud vendor guidance, and interest rate trends; in the medium term, funds may be more willing to allocate to sub-sectors such as electricity, equipment, networks, and storage that can share in AI infrastructure spending.

SemiAnalysis reported that the Nvidia Rubin Ultra will feature 192GB of HBM4 8-hi memory. This is not only a significant reduction from the initially previewed 1TB HBM but is even lower than the 288GB capacity of the standard Rubin.The reasons for the HBM configuration being cut down to about one-fifth of the original include: the cancellation of the four-die Rubin Ultra scheme, reducing the number of cubes to 8; the stacking height being lowered from 16-hi to 8-hi; and the use of HBM4 with 24Gb dies instead of HBM4E with 32Gb dies. An upgraded version with 8-hi HBM4E may be released later.

According to a report by ZDNet Korea, Samsung Electronics and SK Hynix plan to continue investing in equipment for NAND flash production lines in China until next year to strengthen local production bases. Since the first half of this year, Samsung has been advancing V9 NAND conversion investments at the Xi'an X2 production line, which features an advanced process with approximately 280 layers stacked, and is expected to achieve a monthly production capacity of 40,000 to 50,000 wafers. To stabilize mass production, the company has confirmed a supplementary investment plan, focusing on additional etching process equipment, with related procurement orders expected to be clarified around the end of the year, and the supplementary investment planned to be executed starting in the second half of next year.SK Hynix has been advancing equipment investments related to NAND product upgrades at its Dalian Plant 2 since the third quarter of this year, with a discussed scale of approximately 30,000 wafers per month. Key manufacturing equipment will begin arriving next month, with plans to complete deployment by the first half of next year. The Dalian plant is an asset acquired after SK Hynix took over Intel's NAND division in the second half of 2020. In 2022, a groundbreaking ceremony was held for the expansion of Plant 2, but equipment investment was temporarily halted due to market conditions; this investment is now being restarted.Industry insiders point out that the above arrangements aim to gradually expand advanced NAND capacity within China to meet the rapid growth in demand for high-end storage in fields such as artificial intelligence servers.

Goldman Sachs' research report on August 23 pointed out that the forecast for wafer front-end equipment (WFE) for 2026 to 2028 has been raised to $150 billion, $218 billion, and $281 billion, which is an increase of 6%, 17%, and 35% respectively compared to previous forecasts. The adjusted year-on-year growth rates for the next three years are 36%, 45%, and 29%, while the previous expectations were 28%, 32%, and 12%.Goldman Sachs believes that DRAM and advanced process foundry are the main short-term drivers. In the DRAM sector, capital expenditure forecasts for Samsung and SK Hynix have been raised by 22% and 19% respectively, with HBM4 mass production driving demand for etching, deposition, and other equipment; in the foundry sector, TSMC's annual capital expenditure has been revised up by $8 billion, and the N2 process is entering the mass production ramp-up phase. NAND and logic/others (including Terafab) will take over in the medium term, and the approximately $16.8 billion initial phase equipment investment promised by SpaceX and Tesla has been included in the WFE forecast.Goldman Sachs is bullish on global semiconductor equipment stocks, with a focus on recommending Applied Materials, Lam Research, ASML, Tokyo Electron, ASMI, BESI, Lasertec, and Ebara.

According to the Economic Daily, semiconductor silicon wafers have seen a significant price increase for the first time since the COVID-19 pandemic, covering the full range of specifications including 6-inch, 8-inch, and 12-inch, with an increase starting at 10%. Industry insiders believe this is the first price adjustment in over three years, and major Taiwanese silicon wafer suppliers such as GlobalWafers, TSMC, and Hejian are expected to see improvements in revenue and profitability.GlobalWafers stated that recent demand in some end markets has gradually improved, inventory adjustments in the supply chain are healthier than in the past, and market signals are more positive than last year, but prices still depend on the product, specifications, and customer situations. TSMC mentioned that under cost pressures and demand support, they have begun communicating price adjustments with customers, and operations in the second half of the year are expected to outperform the first half. Hejian indicated that they are actively negotiating price adjustments with customers and continuing to promote advanced packaging and related products.Industry analysis suggests that this round of price increases mainly reflects the rebound in chip demand driven by AI, with increased usage in advanced and mature processes, allowing the upstream silicon wafer industry to gradually feel the recovery. Currently, the negotiation progress varies among factories and customers, with some 12-inch polished wafers already subject to new quotes with double-digit percentage increases, and discussions for 8-inch and 6-inch products are also moving in a similar upward direction.

SK Hynix (SKHY.O) stated in a document that it has not yet decided whether to build semiconductor manufacturing facilities in Japan. The company responded to media reports regarding its potential factory construction in Japan.SK Hynix stated: "To enhance the competitiveness of the memory chip business, the company is evaluating various options, including the establishment of additional production bases, but no decisions have been made as of now." The company said it would further disclose relevant information when specific details are confirmed, or within one month after the release of this report.

Micron Technology's CEO announced the establishment of Micron Research Labs, planning to invest $10 billion, with an investment cycle measured in decades. The lab will bring together customers, suppliers, academia, government, and the broader semiconductor ecosystem, dedicated to breaking through existing technology roadmaps and exploring future possibilities.Micron stated that this move reflects a long-term commitment to early innovation. The company's team has accumulated over 62,000 patents. Micron Research Labs will join Micron's investment layout of over $250 billion in U.S. manufacturing, expected to create more than 90,000 American jobs, covering fabs, engineers, technicians, apprentices, suppliers, and communities.Micron emphasized that current decisions will determine who can lead the future AI economy, and the future of AI in the U.S. will be built on memory manufactured in the U.S., produced by Micron.

According to BeInCrypto, Micron CEO Sanjay Mehrotra stated at a semiconductor manufacturing plant near the company's headquarters in Boise, Idaho, that memory has become a strategic infrastructure in the AI era, and the growth in AI-driven demand will permanently reset the economic logic of this cyclical industry. In terms of financial data, Micron's revenue for the third quarter reached $41.46 billion, a significant increase from $9.3 billion last year, with a gross margin rising to 84.6%, and it is expected that the gross margin for the fourth quarter will further increase to about 86%.On the supply side, the demand from data center customers is about 150% of the supply that Micron can commit to, indicating a significant gap. Mehrotra also listed autonomous vehicles, robotics, and AI consumer devices as sources of incremental demand in the future.

According to a report by Infomax under the Korea Herald, SK Hynix is advancing its plan to build a semiconductor factory in Miyagi Prefecture, Japan, with an investment scale of hundreds of trillions of Korean won. SK Group Chairman Choi Tae-won recently visited the area personally. Miyagi Prefecture is one of the three major semiconductor industry bases supported by the Japanese government.The report states that this factory will be an overseas production base added by SK Hynix outside of its clusters in Yongin, South Korea, and Hunan, China, aimed at meeting the ongoing tight global supply and demand situation for memory. If the plan materializes, SK Hynix will become the third foreign semiconductor company to establish a factory in Japan, following Micron and TSMC. The report points out that due to ongoing pressure from the United States for South Korean companies to increase investments in the U.S., this investment in Japan may further intensify U.S. pressure, while the Japanese government's subsidy policy is also an important consideration. SK Hynix has not officially responded to this news. Previously, on August 20, SK Hynix announced a 40 trillion won share buyback cancellation plan.

According to Reuters, citing informed sources, Samsung Electronics has raised prices for some new advanced foundry service orders by up to 15%, due to a surge in demand for AI chips leading to tight capacity. The company has long lagged behind TSMC in the foundry sector. Demand from Chinese customers is particularly strong, but Samsung must prioritize serving U.S. customers and reserve some capacity to support its own chip production, making it unable to take on all orders. Affected by U.S. export restrictions on advanced chip equipment to China, Chinese companies are increasingly reliant on overseas foundries, with some Chinese customers accepting the maximum price increase.Informed sources stated that Samsung raised the prices for the 4nm (SF4) process in July, with price increases of 10% to 15% for Chinese and U.S. customers, and 5% to 10% for Taiwanese customers; the price for 5nm (SF5) wafers increased by 10% to 15%, nearly 10% higher than the older 8nm process. The SF4 production line at the Pyeongtaek factory has been operating at full capacity since the end of last year, producing logic chips for customers like Qualcomm and providing basic bare chips for Samsung's own HBM. Analysts point out that after TSMC's advanced capacity was fully booked, customers turned to Samsung and Intel, giving Samsung room to raise prices.Samsung's foundry business has been continuously losing money since 2022, but the company expects to quickly turn a profit driven by improved utilization rates, yield improvements, and rising prices. Foundry revenue in the second half of the year is expected to grow by double digits year-on-year, with advanced processes accounting for more than half, and AI and high-performance computing applications accounting for over 30%. Companies like Tesla, Apple, Broadcom, and NVIDIA have already reached relevant cooperation agreements with Samsung, and Google is also in talks to use the SF4 process.

Park Chang-min from Samsung Electronics Semiconductor Research Institute's Foundry Process Development Team stated at the "2026 Next Generation Lithography + Patterning Academic Conference" that the company plans to apply High-NA EUV technology starting from the 1-nanometer (A10) process, aiming for mass production around 2030. Samsung is currently still using the existing 0.33 NA EUV multiple patterning scheme for processes at 2 nanometers and below, with plans to achieve mass production of 1.4 nanometers by 2029, and then advance to 1 nanometer.High-NA EUV will increase the lens numerical aperture (NA) from 0.33 to 0.55, providing higher resolution, which can simplify ultra-fine processes that originally required multiple patterning into single patterning, helping to reduce manufacturing costs, improve production efficiency, and enhance design flexibility. However, this technology is highly challenging, equipment is expensive, and it requires simultaneous upgrades in supporting material technologies. Park Chang-min expects that the 1.4 nanometer and 1 nanometer processes will still primarily use 0.33 NA EUV multiple patterning, after which High-NA patterning will become the core technology of the new generation processes.

According to Bloomberg, former OpenAI researcher Leopold Aschenbrenner's AI fund Situational Awareness has splurged $400 million this week to support a private company, which is the semiconductor manufacturing startup Source Foundry. It is reported that VC giant Sequoia Capital has also provided support for this startup. So far, representatives from Situational Awareness and Source Foundry have not responded to requests for comments, while Sequoia Capital has declined to confirm the hedge fund's investment situation.It is reported that Source Foundry is dedicated to developing a semiconductor manufacturing process that is simpler, cheaper, and faster than existing methods, aiming to bridge the widening gap between the demand for AI computing power and the capacity expansion of chip manufacturing. Analysts believe that Leopold Aschenbrenner's large-scale investment reflects the ongoing strong demand in the market for new chip manufacturing technologies, as AI developers' demand for computing power is increasing day by day. However, this investment also reveals that Situational Awareness has formed a highly concentrated risk exposure before facing significant financial pressure due to severe fluctuations in its publicly held technology positions.