Semis sentiment cools, memory leads declines; META unveils Muse Wonder, racing ahead in AI apps | Daily News Recap

1013 | Dolphin Focus

🐬 Macro/Industry

1、After the Korea holiday catch-down, the early termination of the largest-ever buybacks by Samsung and SK hynix (~$40bn), and CXMT's breakthrough on a 4F² cell architecture, memory names slid in tandem with SanDisk and Micron sharply lower. The scrapped buyback weakens the perceived floor for the space, while domestic memory advances add supply-side noise and cloud vendors' in-house accelerators stoke substitution risk. Dual headwinds are compressing semiconductor risk appetite, leaving near-term sentiment under pressure.

2、The Ministry of Finance will cut the VAT export rebate on NEVs from 13% to 9% effective Apr 1 next year. This directly erodes price competitiveness for China-made NEV exports and pressures margins at BYD, Nio, XPeng and others with rising export mix. That said, the reduction is moderate with a transition window, and it mostly signals a shift from 'subsidy tailwinds' to a 'product strength' battle.

🐬 Single Stocks

1、$Meta Platforms(META.US)

Meta launched its new model Muse Wonder, which Mark Zuckerberg says is at the AI frontier, priced at $20–$100/month, fueling debate on a 'second growth curve' and whether AI FOMO can ignite earnings season. It is a major upgrade vs. the first post-reorg model, aimed at personal/pro intelligent agents and opening non-ads revenue streams. Shares led megacap tech higher as investors reprice 'AI application monetization', with earnings the key validation.

2、$Marvell Tech(MRVL.US)

Marvell rolled out its 2nd-gen CXL memory expander, supporting up to 9TB of DDR4 per device, targeting the 'memory wall' in AI LLM clusters and large-scale compute fabrics. With model training and inference driving ever-higher memory capacity needs, CXL remains the core path for memory scaling. The new product raises the per-card attach limit to fit AI cluster demand, helping Marvell consolidate share in CXL. Watch server OEM adoption, ramp cadence, and order flow as the key to earnings realization.

3、$MINIMAX-W(00100.HK)

MiniMax opened its first official flagship store on Tmall ahead of Double 11, listing Go/Explore/Build M Plan tiers with monthly and annual billing to sell AI services directly to consumers. This marks a push into consumer subscriptions, leveraging Tmall traffic and moving beyond a pure API-to-B monetization path. Competition in consumer LLMs is intense, and user retention, pay conversion and rival pricing present pressure. This is a diversified monetization attempt, and earnings contribution needs ongoing tracking.

4、$AMD(AMD.US)

AMD introduced the Helion AI agent rack design, supporting up to 32,768 cores per configuration, with endorsements from eight ecosystem partners including Cisco and Dell, aimed at AI agent-era cluster deployments. AMD is building a full-system cluster solution and partnering across hardware to divert share from Nvidia's AI server stack. AI agents require tighter compute interconnects and orchestration, and a standardized rack lowers deployment barriers. However, ecosystem penetration and customer validation take time, limiting near-term earnings impact.

5、$Intel(INTC.US)

Intel is accelerating its open infrastructure push for the AI agent era, shipping samples of its GPU D series to customers to begin downstream testing. Sample shipments signal a shift from R&D to commercial validation, marking Intel's formal entry into the AI compute race. Given the entrenched GPU landscape, Intel must lean on open ecosystems and differentiated solutions. Feedback from testing and the timetable for mass production are the core markers to watch.

6、$BABA-W(09988.HK)

Multiple media reports say Alibaba's Qwen Personal Agent product 'Qwen Go!' is set for public release, a personal agent deployment that extends Qwen's real-world use cases. After APIs and industry models, Alibaba is pushing into personal agents to evolve LLMs from tools into everyday assistants and fill its agent product map. China's personal agent field is crowded, and product experience, scenario stickiness and monetization models are the key tests. Differentiated UX will determine odds of a breakout.

7、$TENCENT(00700.HK)

Foreign media, citing sources, report Tencent plans to issue around $5bn in USD bonds, with proceeds mainly for AI-related capex and compute infrastructure. The sizable USD funding underscores Tencent's continued scale-up in AI compute, where investment is a long-term necessity. Financing helps smooth capex cash flow and supports server and cluster procurement, though sustained heavy capex will cap near-term profit. Track the deployment pace and AI monetization progress.

🐬 Top Gainers by Sector

A-share: Consumer finance, data processing & outsourcing, oil & gas storage and transportation.

HK: Oil & gas refining & marketing, apparel retailers, oil & gas storage and transportation.

US: Home appliances, breweries, agricultural products.

🐬 Watchlist for Tomorrow

1) ASML reports earnings, and Dolphin Research will monitor and publish a read-through.

2) China Sep trade balance (RMB), China Sep trade balance (USD), China Sep CPI YoY.

3) US Sep CPI MoM (sa), US Sep core CPI MoM (sa), US Sep CPI YoY (nsa), US Sep core CPI YoY (nsa).

Risk disclosure and disclaimer: Dolphin Research Disclaimer and General Disclosures