MSX US Stocks Daily Watch: Nasdaq Hits Record High for Two Consecutive Days, AI Trading Enters "Full-Stack Race"! Alibaba Launches In-House Chip, Meta and Qualcomm Vie for the Next Entry Point
- Core View: The Nasdaq index hit a record high for two consecutive days, but market performance is diverging. AI investment is shifting from single-point competition to full-stack competition. The core of the market has entered the revenue realization stage, and investors need to distinguish between three levels: realized revenue, signed contracts, and pre-commercialization roadmaps.
- Key Elements:
- The Nasdaq Composite rose 0.45% on September 22 to close at 27,244.28, setting a record closing high for the second consecutive day; the S&P 500 slipped 0.06 points, while the Dow fell 0.36%, highlighting clear market divergence.
- AMD's Q2 data center revenue reached $6.7 billion, up 107% year-over-year, accounting for about 58% of total revenue; its market cap surpassed $1 trillion for the first time, with cumulative gains of about 185% year-to-date.
- AMD signed a multi-year agreement with Meta, planning to deploy up to 6GW of Instinct GPUs, with the first 1GW supported by custom MI450 architecture GPUs and sixth-generation EPYC processors, but this is a phased target rather than delivered capacity.
- Alibaba launched its in-house Zhenwu V900 chip, with plans for mass production in Q1 2027; latest-quarter AI cloud revenue was about $7.1 billion, up 45% year-over-year, while capital expenditure approached $10 billion, up 75% year-over-year, showing that AI investment is already reflected in financial data.
- Meta's Muse accumulated about 2.8 million installs in the first 12 days after launch, surpassing ChatGPT's performance over the same period, but downloads do not equal revenue and retention, and sustained usage and commercialization capability still need to be verified.
- Qualcomm, Lumentum, and Corning showcased an optical chip-to-chip interconnect solution with a bandwidth density of about 1Tb/s/mm and a long-term target of 4Tb/s/mm. Optical interconnect is extending closer to the chip, but it remains at the technology demonstration stage.
- Analyzing AI companies requires distinguishing three layers: businesses that have already generated revenue (AMD data centers, Alibaba Cloud AI), signed contracts awaiting delivery (the AMD-Meta 6GW agreement), and roadmaps not yet commercialized (Alibaba's 20GW target, Qualcomm's optical interconnect). The three have different levels of certainty and should not receive the same valuation.
[MSX Research Institute · Daily U.S. Stock RWA Observation] is a flagship daily report produced by MSX, a leading RWA trading platform. Leveraging our strong macro research capabilities, we capture the core pulse of global traditional U.S. equities, liquidity changes, and the RWA tokenization market, helping you position ahead of quality assets.
Today's Observation:
The Nasdaq index has once again set a new all-time record.
On September 22 local time, the Nasdaq Composite rose 0.45% to close at 27,244.28, marking its second consecutive trading day of record closing highs. The previous trading day, the Nasdaq had just surged 2.26%, breaking through 27,000 for the first time.
However, unlike the previous day when AI stocks such as Meta, AMD, and Arm rallied collectively, Tuesday's market performance was noticeably more divergent.
The S&P 500 was nearly flat for the day, edging down 0.06 points to 7,764.64; the Dow Jones Industrial Average fell 0.36% to close at 51,863.69. Meanwhile, the Russell 2000 small-cap index rose about 0.5%.
This indicates that not all stocks rose in sync that day; rather, tech stocks, chip stocks, and some small-caps continued to strengthen, offsetting declines in banks, energy, and other traditional sectors.
Within the tech sector, market capital continued to concentrate in semiconductors, chip design tools, and AI infrastructure-related companies. Qualcomm rose 9.3%, Cadence gained 6.4%, Intel climbed 6.3%, Synopsys rose 5.8%, and Lam Research gained 5.7%. Meta continued to rise 5.3% after jumping 11.4% the previous day, while Apple and Microsoft gained about 1.9% and 1.1% respectively.
Meanwhile, the recent signals from Alibaba, AMD, Meta, and Qualcomm each represent different positions along this industrial chain: Alibaba aims to control chips, cloud, models, and applications simultaneously; AMD is transitioning from a single-chip supplier to a full AI system platform; Meta is building compute power while using Muse to compete for the consumer AI entry point; and Qualcomm is attempting to bring low-power computing and connectivity capabilities from smartphones into data centers.
This means AI investment is shifting from point competition to full-stack competition. On September 22, Alibaba unveiled its new AI roadmap at the Yunqi Conference, introducing its self-developed Zhenwu V900 training and inference chip, and announced that Qwen 4 is in training. The planned scale of subsequent Qwen 4.5 and Qwen 5 will reach 5 trillion to 10 trillion parameters, and the company also proposed increasing Alibaba Cloud's globally operated data center capacity to over 20GW by 2032. These numbers are eye-catching enough, but what truly matters is not the scale of model parameters, but that Alibaba is connecting chips, servers, networks, storage, models, and agent platforms together.
This strategy is similar to the logic behind Meta, Google, Amazon, and Microsoft pursuing self-developed chips: as AI computing scale continues to expand, cloud providers cannot rely solely on external GPU supply—they also need self-developed chips to reduce unit computing costs, control product roadmaps, and reduce constraints from single suppliers.
Meanwhile, after AMD's market cap broke through $1 trillion for the first time, the factors driving valuation re-rating have shifted from AI concepts to actual revenue from data center business. AMD's second-quarter revenue was $11.536 billion, up 50% year-over-year; data center revenue reached $6.7 billion, up 107% year-over-year, accounting for approximately 58% of total company revenue.
Therefore, the current AI rally is no longer just investors' imagination about future products. Some companies have already entered the revenue realization stage, while others are still in the product demonstration and capacity building stage. Distinguishing these two types of companies is key to judging whether this rally can continue.
Data in a Minute:
• The Nasdaq Composite rose about 2.3% on September 21, setting a record closing high, with semiconductor and AI-related companies as the main driving force.
• As of breaking through $1 trillion, AMD's cumulative gain in 2026 was approximately 185%, significantly outpacing the Nasdaq's performance over the same period.
• AMD's second-quarter data center revenue was $6.7 billion, up 107% year-over-year, primarily driven by demand for EPYC server CPUs and Instinct GPUs.
• AMD signed a multi-year cooperation agreement with Meta, planning to deploy up to 6GW of AMD Instinct GPUs; the first 1GW is expected to be supported by custom MI450 architecture GPUs and sixth-generation EPYC processors. The agreement represents phased deployment targets and does not mean 6GW has already been delivered.
• Alibaba's Zhenwu V900 features 216GB of memory and 1,200GB/s inter-chip bandwidth; according to company statements, performance is approximately three times that of the previous generation M890, with mass production and commercial release planned for Q1 2027.
• Alibaba stated that more than 650 customers have adopted Zhenwu series chips; the new-generation server architecture can support clusters of up to 500,000 cards, but this is the system design ceiling, not the scale of clusters already deployed.
• Alibaba's latest quarterly AI cloud and computing services revenue was approximately $7.1 billion, up 45% year-over-year; cloud business adjusted EBITA grew 133% to $830 million, with a margin of approximately 12%. AI-related product revenue was approximately $1.8 billion, achieving triple-digit growth for the twelfth consecutive quarter.
• Alibaba's capital expenditure for the quarter approached $10 billion, up 75% year-over-year, indicating that cloud revenue growth still requires high-intensity infrastructure investment.
• Third-party agency Apptopia estimates that Meta's Muse accumulated approximately 2.8 million installs in the first 12 days since launch; in the U.S. and Canada iOS market, downloads reached approximately 1.8 million in the first 12 days, higher than ChatGPT's approximately 1.3 million during the same period after its mobile launch.
• Qualcomm, Lumentum, and Corning showcased optical inter-chip connectivity solutions for AI servers, with prototypes using 32Gb/s NRZ channels, current design bandwidth density of approximately 1Tb/s/mm, and a long-term target of approximately 4Tb/s/mm.
MSX View:
The bottleneck of AI systems is expanding from a single chip to the entire data center. Models need GPUs for training and inference, CPUs for data processing and task scheduling, high-speed networks to transmit data between thousands of chips, and storage, power, cooling, and software platforms to keep these resources highly utilized.
When AI shifts from chatbots to agents capable of continuously executing tasks, this change becomes even more pronounced. Agents need to constantly call models, browsers, databases, and external software, and a single task may trigger multiple rounds of inference. Computing demand depends not only on the number of users but also on how many tokens each user consumes to complete a task.
This is precisely the significance of Meta launching Muse. Muse's early download performance suggests that Meta may leverage the distribution capabilities of Facebook, Instagram, and WhatsApp to rapidly establish a consumer AI entry point. But downloads do not equal revenue, nor do they equal long-term retention. What Muse needs to prove next is whether users will continue to use it, whether they are willing to pay, and whether the computing cost generated per user can be lower than subscription and other commercialization revenue.
If Muse can achieve high-frequency usage, Meta's demand for GPUs, CPUs, and data center capacity will continue to increase. This also explains why Meta is simultaneously partnering with AMD, Qualcomm, and other suppliers. It is not simply "abandoning Nvidia," but rather seeking to build a more price-negotiation-capable supply system across GPUs, server CPUs, self-developed chips, and networking equipment.
AMD is currently the company with a higher degree of commercial realization in this logic. Compared to the partial roadmaps announced by Alibaba and Qualcomm, AMD already has $6.7 billion in quarterly data center revenue and has signed a deployment agreement of up to 6GW with Meta. The simultaneous growth of EPYC server CPUs and Instinct GPUs also allows AMD to generate two types of revenue from a single AI system, rather than relying solely on its traditional CPU business.
But after AMD's market cap broke through $1 trillion, market expectations for it have also risen significantly. Merely proving that products can compete with Nvidia is no longer enough; the company also needs to deliver MI450 and Helios systems on schedule, expand its software ecosystem, and truly convert large deployment agreements into revenue, profit, and cash flow.
Alibaba's investment logic is different. Alibaba's greatest potential advantage is not having a single AI chip, but simultaneously possessing chips, Alibaba Cloud, the Qwen model, enterprise customers, and application scenarios such as Taobao and Tmall. Even if self-developed chips are not sold externally, as long as they can reduce Alibaba Cloud's inference costs and improve supply stability, they may create economic value.
More importantly, Alibaba's AI investment has already begun to reflect in its financial data. The latest quarter's cloud computing revenue grew 45%, and adjusted EBITA grew 133%, indicating that AI demand not only drives revenue but also has not prevented cloud business margin improvement for now.
But this strategy remains highly capital-intensive. Alibaba's single-quarter capital expenditure has approached $10 billion, and the 20GW data center capacity target for 2032 is only a long-term goal, not the scale already in operation. Parameter scale likewise cannot be directly equated with model quality. The 5 trillion to 10 trillion parameters describe a future model roadmap; ultimate capability also depends on model architecture, training data, inference efficiency, and post-training methods. Larger models also mean training and operating costs may be higher.
Qualcomm and Lumentum represent another branch: connectivity between computing chips is extending from electrical interconnects to optical interconnects. As a single AI rack and cluster contains more and more accelerators, whether data can move between chips with low latency and high bandwidth will directly affect GPU utilization. The optical inter-chip connectivity solutions showcased by Qualcomm, Lumentum, and Corning indicate that the application scope of optical communications may continue to expand from between data centers and between racks to inside racks and even near chips.
This could expand the long-term market space for Lumentum, Corning, and other optical component companies, but caution should still be maintained for now. Technology demonstrations prove feasibility, not customer orders; bandwidth targets prove design potential, not revenue already realized. The commercialization timeline for Qualcomm's data center CPUs is also mainly concentrated in the coming years, and it is not yet possible to directly count all new markets beyond smartphone chips into performance.
So the core of this rally is that the value volume of AI infrastructure is expanding. Nvidia still possesses a mature software ecosystem, system capabilities, and customer base. The growth of Alibaba's self-developed chips, AMD's accelerators, and Qualcomm's data center products is more likely to first expand the total size of the AI computing market and help large cloud providers establish multi-supplier systems, rather than immediately forming zero-sum substitution.
To analyze these companies going forward, news needs to be divided into three levels:
The first level is businesses that have already generated revenue, such as AMD's data center revenue and Alibaba Cloud's AI revenue;
The second level is contracts that have been signed but still need to be delivered, such as AMD's deployment agreement of up to 6GW with Meta;
The third level is roadmaps that have not yet been commercialized, such as Alibaba's 20GW capacity target, future 10-trillion-parameter models, and Qualcomm and Lumentum's optical interconnect solutions.
The certainty of these three is completely different, and they should not receive the same valuation. This AI trade is upgrading from "buying GPUs" to "competing for control of the entire AI system." The companies that can truly continue to outperform in the future are not just those with larger models or faster chips, but those that can convert technology roadmaps into product delivery, high utilization, sustained revenue, and free cash flow.
About MSX:
MSX is a leading RWA trading platform dedicated to providing secure, efficient, and transparent access to global financial markets. As one of the world's earliest on-chain U.S. stock trading platforms, MSX has always been at the forefront of the industry, leading market transformation through continuous innovation.
The platform deeply integrates blockchain technology with compliance frameworks, comprehensively offering spot and derivatives trading for nearly 400 tokenized stocks and Pre-IPO assets, perfectly bridging the gap between traditional finance and the digital asset industry.
Centered on its core mission of "enabling quality assets to circulate freely," MSX has now built a diversified digital financial services system covering U.S. stock spot and perpetual contracts, crypto-to-crypto trading, Pre-IPO, and MSX Research Institute, aiming to provide global investors with around-the-clock, high-performance access to quality assets.
Official Website: https://msx.com/
Chinese Telegram:https://t.me/MSXOfficial
Global Telegram:https://t.me/MSXGlobal
Now simultaneously available on the global App Store / Google Play.
Risk Disclaimer: Macroeconomic conditions and U.S. stock markets are subject to significant volatility. This content is provided solely for academic and research observation reference by MSX Research Institute and does not constitute any investment advice.
