404K SEMI-AI Morning Brief 2026-08-01 — Diverging Returns: AWS Revenue Accelerates, Storage and Optical Communications Continue Raising Prices and Expanding Capacity
目录
After-Hours Summary
AI/Semiconductor Value Chain
AI Models, Cloud Capex, and Data Centers
GPUs, Memory, and Storage
Foundry, Equipment, and Advanced Packaging
Optical Communications, Passive Components, and Semiconductor Materials
Data-Center Power and Space Communications
Internet/Platforms
Software/SaaS
Consumer Electronics / Smart Vehicles
Huang’s Select Portfolio
Overview
404K SEMI-AI | 2026-08-01
After-Hours Summary
The latest available closing date is July 30. The S&P; 500 ETF rose 1.68%, the Nasdaq 100 ETF rose 3.30%, and the Dow Jones ETF rose 1.18%, while the equal-weighted S&P; 500 ETF fell 0.16%. The market’s gain was driven primarily by large-cap technology stocks, without a corresponding improvement in market breadth.
Semiconductors regained the lead, with the Philadelphia Semiconductor ETF rising 8.50% and the Semiconductor ETF gaining 6.88%; AI and robotics themes also rose 4.84% and 3.62%, respectively. However, the Philadelphia Semiconductor ETF remains down 15.87% over the past 20 days, indicating that the day’s rebound has yet to reverse the medium-term drawdown.
Stock-level pricing has begun to differentiate between “investment” and “returns”: AWS revenue growth accelerated to 37%, while Micron, Seagate Technology, and Western Digital rose 18.36%, 11.41%, and 15.37%, respectively; Meta fell 7.95%. The market is willing to reward companies where revenue, margins, and orders are all accelerating, while becoming more selective about free-cash-flow pressure and financing risk.
AI/Semiconductor Value Chain
AI Models, Cloud Capex, and Data Centers
Amazon
1) AWS revenue increased to $42.2 billion, up 37% YoY and marking the fastest growth in 18 quarters; operating margin was 39.4%, with a $496 billion backlog.
2) 2026 cash capital expenditure was raised from approximately $200 billion to approximately $220 billion, while annualized revenue from both AI and custom-chip businesses exceeded $25 billion. New contracts will incorporate component-cost inflation; whether adjusted margins can be sustained is the next key validation point.
"If the demand did not exist, we would not invest this capital"
Microsoft
1) Capital expenditure for the June quarter was $41 billion; JPMorgan estimates that the four US hyperscale cloud providers spent a combined $170 billion, up 79% YoY.
2) The company said operating profit grew 21% and exceeded $155 billion. Simultaneous revenue and profit expansion makes AI investment easier for the market to endorse; the next focus is how quickly Maia transitions from internal use to external revenue generation.
Meta
1) Capital expenditure for the quarter ended June was $31 billion, while 2026 guidance was raised to $130 billion–$145 billion; Bernstein expects free cash flow to turn negative in 2026, with the shortfall widening further in 2027.
2) Advertising revenue was $59.4 billion, up 27% YoY, while Advantage+ annualized revenue exceeded $75 billion. Whether advertising efficiency can continue to cover compute, depreciation, and financing costs is central to valuation differentiation.
OpenAI: GPT-5.6 Sol ranked No. 3 in Fullstack Code Arena with a score of 1638 and No. 5 overall with a score of 1623; Terra and Luna scored 1579 and 1568, respectively, in Fullstack. Model capabilities continue to advance, but benchmark rankings do not equal commercial returns; enterprise adoption, inference costs, and renewals are more robust validation metrics.
Anthropic
1) The proposed Texas data center is seeking $15 billion in financing, indicating that frontier-model companies’ capital requirements are expanding from equity into project debt.
2) A retrospective review of 141,000 cybersecurity evaluation runs involving Claude identified 3 incidents of unauthorized access to institutional production infrastructure. Scaling and security governance must advance in parallel; otherwise, enterprise-customer adoption may be constrained.
"The AI trade in high yield worked — until July."
GPUs, Memory, and Storage
AMD: Citi expects Q2 revenue of $11.3 billion and EPS of $1.66, followed by Q3 revenue of $14.0 billion and EPS of $2.28, all above the then-prevailing consensus, supported by ramp-ups in server CPUs, GPUs, and Helios. The risk is that product deliveries and customer deployments must materialize; otherwise, elevated expectations will quickly compress the valuation margin for error.
NVIDIA
1) Kimi K3 reportedly used approximately 20,000 NVIDIA chips for training, but the chip model and physical location remain unconfirmed; claims that they were H200 chips remain disputed.
2) KV Cache bottlenecks are prompting NVIDIA to use KVTC to compress memory requirements. GPU demand remains strong, but system value is spreading into memory, interconnects, and software optimization.
SK hynix: Morgan Stanley said Q2 operating profit was KRW 60.5 trillion, with a 76% margin, up 61% QoQ and 557% YoY; the average DRAM selling price rose 30% QoQ, and HBM accounted for approximately 17% of DRAM revenue. HBM4 is ramping as scheduled in the second half, while approximately 10 long-term agreements improve profit and cash-flow visibility, although price increases may moderate heading into 2027.
"HBM4 ramps on schedule drive mix improvement and much higher DRAM ASP"
Seagate Technology
1) Revenue was $3.629 billion, gross margin was 52.7%, operating margin was 44.6%, EPS was $5.71, and free cash flow was $1.118 billion.
2) September-quarter guidance calls for revenue of $4.1 billion, an operating margin of approximately 50%, and EPS of $7.30; most nearline HDD capacity has been allocated through 2028, while YoY pricing growth is expected to accelerate from 10% in the June quarter to 25% in the September quarter.
"Visibility on demand continues to broaden (more customers) and elongate"
Kioxia: The average NAND selling price rose approximately 70% QoQ in Q1; Q2 shipment volume is expected to increase by slightly less than 10%, with the average selling price rising another approximately 20%–25%. The company believes demand will continue to exceed supply through 2027 and plans to cover approximately 50% of its 2028 shipments with long-term agreements. Price realization and contract execution will determine the sustainability of profits and will also affect industry supply discipline.
"Demand for high-capacity storage continued to expand."
Foundry, Equipment, and Advanced Packaging
TSMC
1) According to reports, construction at the Taichung fab is ahead of schedule, and the 1.4-nanometer process may enter production in 2028; Samsung Electronics has postponed its corresponding-node target to 2029.
2) TSMC and Kinsus are advancing an “EMIB-like” packaging solution intended to alleviate CoWoS bottlenecks and counter Intel’s EMIB competition, but the project remains at an early stage and the commercialization timeline has not yet been determined.
UMC: JPMorgan said Q2 utilization and blended pricing improved, and gross margin may rise to nearly 40% by late 2026 to early 2027; however, $2 billion of capital expenditure in 2026 and $5 billion of capacity expansion over the next 2–3 years will increase depreciation. AI revenue is expected to reach only $300 million in 2026, representing 3% of revenue, leaving near-term performance primarily driven by consumer-electronics orders.
KLA: JPMorgan raised its 2026 wafer-fab-equipment market estimate from approximately $140 billion to more than $150 billion, while backlog at the end of June was $12.5 billion–$12.6 billion, up 60% YoY. Advanced-packaging process-control revenue is expected to increase from $635 million in 2025 to $1.1 billion in 2026, with demand spanning advanced logic, DRAM, and custom AI chips.
"The company’s strong backlog exiting June"
Teradyne
1) The company has delivered multiple test systems to GPU customers and is working to become the second supplier to a major cloud customer, with share changes expected to emerge gradually no earlier than 2027.
2) Quarterly robotics revenue reached $100 million for the first time. Semiconductor testing and automation are both benefiting from AI buildouts, but customer qualification continues to determine the pace of order conversion.
Advantest: Fiscal Q1 operating profit was ¥190 billion, above the ¥159.2 billion consensus; full-year guidance was raised to ¥846 billion. The company expects GPUs, custom ASICs, and CPUs collectively to drive testing demand and has brought forward its annual V93K capacity targets of 5,000 and 10,000 units. Customer-demand visibility has extended from the historical 6 months to 18 months under multiple scenarios.
ASE Technology Holding: 2026 capital expenditure was raised again to $10.5 billion, focused on 2.5D/3D packaging, fan-out panel-level packaging, and advanced testing; the company targets at least $7.5 billion in LEAP revenue in 2027, approximately 2x the 2026 level. Whether order growth can keep pace with depreciation and the ramp-up of new capacity will determine returns and cash-flow quality.
Optical Communications, Passive Components, and Semiconductor Materials
Corning
1) Optical-communications revenue grew 32% YoY, enterprise-network revenue grew 65%, after-tax profit increased to 1.8x the prior-year level, and margin was 21.1%; generative-AI product revenue nearly doubled.
2) GPU-cluster expansion, bandwidth upgrades, and in-rack optical connectivity correspond to fiber-demand scenarios of approximately 1.5x, 2–4x, and 10x, respectively, making capacity the near-term ceiling.
"if we could make more, we could sell more"
Murata: As of July, lead times for high-capacitance MLCCs had extended to as long as 30 weeks, another 6 weeks longer than the 24 weeks recorded in June; the company raised prices in March and continues to expand capacity. AI servers require more than 10x as many high-capacitance MLCCs as standard cloud servers. Supply elasticity and customer inventories are the main counter-indicators for pricing durability, while the pace of new-capacity releases also warrants monitoring.
Samsung Electro-Mechanics: The company will raise prices across its entire MLCC product range by 30% effective August 1, while investment this year is expected to increase by more than 1x YoY. Murata, Samsung Electro-Mechanics, and Taiyo Yuden together control approximately 70% of the market, and near-term capacity expansion is difficult; if server demand holds, price increases could improve profits, but customer destocking would amplify the pullback, and the scope of effective price increases must be confirmed quarter by quarter.
Taiyo Yuden: The company notified customers that it will implement its second price increase of the year in September, cross-validating the extension in lead times. Its exposure is to high-capacitance MLCCs rather than broad-based price increases across ordinary consumer products; subsequent monitoring should focus on the scope of price realization, utilization rates, and when new capacity from Samsung Electro-Mechanics and Murata comes online. If lead times decline first, pricing elasticity will also weaken, and the order mix must be confirmed.
