---

I. Large AI Models

Moonshot AI Responds to Rumors of Dual Listing in HK and Shanghai; Insiders Say Reports Are False

On September 10, media reported that leading domestic large model company Moonshot AI was considering listing in both Hong Kong and Shanghai to secure more funding. Insiders responded bluntly to The STAR Market Daily: the reports are false. South China Morning Post had previously cited two sources reporting the same rumor.

The rumor itself indicates that the market is waiting for a valuation anchor for a large model IPO. Investors are looking for answers on whether the cash-burning-for-computing-power model of top companies can be sustained long-term, and if going public for capital injection is the next exit route. The denial means Moonshot AI hasn't yet reached the point where it must raise funds publicly; primary market money isn't tight enough for that yet; it also serves as a reminder to be extra cautious about listing rumors regarding domestic companies.

This has a calibrating effect on the valuations of the domestic large model tier. If there are actual IPO moves later, it will directly set a comparable price tag for companies in the same tier; now that the rumor has been denied, this anchor cannot be established for the moment, and each company must continue finding their own price in the primary market.

• Strong Chinese Presence at ECCV 2026 — The top computer vision conference is held from September 8 to 12 in Malmö, Sweden, with Google, Meta, Apple, and Amazon as major sponsors. Chinese companies lead multimodal reasoning workshops; 64 teams globally formed groups to solve problems; the first session on 3D reconstruction featured 14 Spotlight papers, with Gaussian Splatting becoming a hot topic, many backed by Chinese teams from Zhejiang University, ByteDance, etc.

---

II. AI Software

Ant Digital Technologies Reveals Transformation Report Card for First Time; AI To B Business Growth Hits Triple Digits

At the Bund Summit media exchange meeting on September 10, Ant Digital Technologies CEO Zhao Wenbiao systematically disclosed the company's business performance since operating independently two years ago: overall business annual average growth near 50%, with enterprise-facing AI business growth reaching triple digits, i.e., year-over-year growth exceeding 100%.

This is a sample for observing the substance of AI implementation in China. While large model companies are generally still searching for monetization paths, Ant Digital attributes growth to enterprise-level AI services, indicating that AI procurement is starting to scale up with real money, rather than just staying at the demo stage. Industry clients with budgets in finance, mobility, and retail are moving AI from pilot projects into formal procurement lists.

This has reference significance for the valuation of enterprise AI service providers. The market has always been hesitant in pricing the domestic AI application layer, lacking quantifiable delivery metrics. Once numbers like triple-digit growth appear, players with strong revenue delivery capabilities will pull away from those who just tell stories.

• Google Gemini Lands on Windows — The native desktop app is finally here, with one-key invocation via keyboard shortcuts, aligning the experience with the Mac version. Google has filled the last high-frequency entry point beyond search, email, and docs, turning the desktop AI assistant from a browser plugin into a system-level feature. For corporate IT departments, Microsoft Copilot faces a true head-to-head competitor, and the battle for AI positioning on the Windows desktop entry point begins.
• India's Pocket FM Annualized Revenue Doubles to $500 Million — This audio story platform doubled its annualized run rate over the past year, with AI-generated content accounting for 93% of its audio content. Traditional radio dramas take weeks to dub one episode; AI dubbing plus multilingual synthesis cuts costs and cycles simultaneously, allowing one drama to be deployed across ten language markets at once. Emerging market content platforms are the first to reap this round of benefits, and also the first to face issues of diluted content quality.
• Amazon Turns Watching Videos into Shopping Trips — Prime Video launched a series of new shopping features, allowing viewers to place orders while watching shows, with products appearing on screen directly linked. This is another attempt by Amazon to weld the traffic end and the shelf end together, further blurring the boundary between content and e-commerce, and giving advertisers new display slots.
• A School Without Teachers — Alpha School in Austin, USA, targets 4th and 5th graders with an annual fee of $40,000. AI customizes course pacing and practice intensity for each child, while human mentors only provide emotional companionship and project guidance. The core proposition is leaving academic parts to machines and keeping human parts for humans. Supporters cite academic progress data, while skeptics worry about social skill development and class filtering. The debate over the boundaries of AI teaching has been reignited by this school, with equal halves impressed and skeptical.

---

III. Humanoid Robots

Mech-Mind Founder Names Galaxy General; Robot IPOs Enter the Calculation Phase

Mech-Mind founder Shao Tianlan recently posted two messages on WeChat Moments, publicly naming competitor Galaxy General. The background is a batch of robot companies lining up to sprint towards listing. Investors' tactics have changed; they no longer pay premiums for backflips in demo videos, but start comparing order quantities, delivery acceptance cycles, and gross margins company by company.

The era of telling humanoid robot stories is winding down. Primary market money is still hot, but the pricing logic has shifted from "who can make it" to "who can sell it, and how much profit per unit." In the coming months, robot companies submitting materials will have their financial statements examined under a magnifying glass; only those who can clearly account for their finances will get the next valuation. Embodied intelligence implementation scenarios are also required for the first time to provide payback periods; demo videos no longer count.

• Maven Robotics Grabs Deployment Business — This startup, founded only in 2024, started with just a cartoon image of a robot and a team. Now specializing in post-factory deployment debugging and O&M for robots, it directly cuts into the delivery segment of body manufacturers. Factories don't buy robots; they buy the result of production lines running stably 7×24. The value of the deployment middle layer is beginning to be priced separately by capital, while body manufacturers face the risk of being pipelined.
• Agents Step Out of Chat Boxes — GeekPark observed that agents are seeking their own "bodies." Hardware companies are embedding Agents into speakers, robots, and car systems. Product forms that only handled files, web pages, and software tasks over the past year are extending into the physical world. The standard for measuring an Agent's capability has shifted from whether it can deliver a document to whether it can complete a physical task from start to finish. "Agents need a body" has become the new industry consensus.

---

IV. Autonomous Driving

• SAIC Roewe Jiayue 07 Pre-sale Set for September 21 — An extended-range SUV equipped with the next-generation Momenta R7 world model advanced intelligent driving solution, with a combined range of 1,480 km. World model solutions are trickling down from the 200k+ RMB vehicle segment to home-use price ranges. The positioning war among smart driving suppliers has already reached volume models, showing the intensity of competition regarding the depth of binding between solution providers and automakers.

---

V. Physical AI & Computing Infrastructure

Google Invests at Least €13 Billion in Finnish AI Infrastructure, Operational Starting 2027

Google confirmed it will invest at least €13 billion (approx. $15 billion) in Finland to build AI and energy infrastructure. The project is expected to go operational in 2027-2028, supporting approximately 7,000 jobs annually upon completion. This is one of Google's largest AI investments in Europe to date.

The bottleneck for AI computing power is shifting from chips to electricity and land. Finland has cheap clean energy and low-temperature cooling conditions. Google betting big here indicates that energy is now the primary factor in hyperscale data center site selection. Data centers in North America queuing for grid connection often delay by more than two years; Finland's shortcut has a demonstration effect for all hyperscalers and is a signal of Europe absorbing the spillover of North American AI capital.

Direct positive impact on power equipment, cooling, and European data center supply chains. For Google itself, this capex will pressure short-term cash flow, but the market is currently willing to pay a premium for certain computing power layouts. Alphabet stock rose 0.59% against the trend on the day of the announcement.

• Anthropic Investor Leads Off-Grid AI Power Startup — Financing models providing independent power supplies for data centers (not connecting to the grid, self-building generation facilities) have received backing from top-tier capital. Grid access queues for new US data centers generally extend beyond 2028. AI companies that can't wait are bypassing utility systems to generate their own power. The AI power business is transitioning from corporate procurement behavior to an independent track in the primary market.
• d-Matrix Joins NVIDIA AI Inference Camp — Chip startups focused on inference continue to expand. Even challengers in niche segments are starting to align with NVIDIA's ecosystem standards, indirectly proving that software stack stickiness is harder to shake than hardware specs. NVIDIA's monopoly in the training market remains unchanged; the inference market competes on cost per token, with everyone grabbing a slice of this incremental pie.
• Transport Drones Used in Nepal Flood Disaster — Mountain roads were extensively blocked after floods, preventing convoys from entering. Drones undertook tasks of delivering food to trapped villages and transporting victims' bodies. Previously, unmanned systems in disaster response were basically limited to reconnaissance and photography. This is the first time they have undertaken actual transport tasks in an organized manner, validating the emergency scenario value of low-altitude logistics in real disasters.

---

VI. Macro & Market Data

Oracle Beats Earnings Expectations, Cloud Infrastructure Revenue Doubles, Stock Jumps 7% After Hours

After the US Eastern close on September 10, Oracle released its Q1 FY2027 earnings report: Total revenue $19.345 billion, net income attributable to shareholders $4.679 billion (up 60% YoY). Cloud infrastructure revenue more than doubled YoY. The stock jumped as much as 7% after hours.

Oracle holds a large number of cloud service orders from AI companies; its cloud revenue has always been a barometer for verifying the authenticity of AI infrastructure demand. Over the past two quarters, the market repeatedly traded on fears of "capex peaking." This earnings report pushed such concerns right back: enterprises' willingness to spend on computing power is still accelerating.

This is a shot in the arm for the entire computing power trading chain. Order expectations for chips, optical modules, and data center operators need to be repriced. Before the Asia-Pacific open this morning, it was a key variable for whether A-share computing power stocks could stabilize after consecutive corrections. RBC analysts' comment after the earnings was just one sentence: Better than feared.

• Nasdaq Injects $100 Million into Kraken Parent Company — Nasdaq Ventures injected $100 million into Payward. Both parties plan to launch tokenized stock trading in 2027, moving listed stocks onto blockchain for 24/7 settlement. With top exchanges personally entering the fray, traditional finance's attitude toward crypto infrastructure is shifting from wait-and-see to equity participation, faster than most people expected.
• Saudi Arabia Considers Merging EA and Savvy Games — Saudi sovereign capital is weighing merging Electronic Arts with its subsidiary Savvy Games Group to create a gaming giant with global-leading revenue scale. EA stock strengthened after the news broke. Saudi Public Investment Fund has bet hundreds of billions of dollars on the gaming industry for years. If realized, this would be another super-consolidation led by state-level capital in the gaming industry after Tencent.
• Defense Tech Mach Industries Valuation Doubles in 3 Months — Recently completed a $600 million Series C extension, raising its valuation to $3.7 billion, just 3 months after the previous round. Targets combining defense and AI continue to be snapped up; hot money is concentrating in the military tech sector.
• Google Signs Largest Rice Methane Carbon Credit Order — Will purchase 1 million carbon credits from Indian climate tech startup Mitti Labs by 2030. This is described by both parties as the largest rice paddy methane reduction deal. Changing irrigation water management rhythms can reduce methane emissions; farmers get compensation, Google gets offset quotas. The carbon market is pricing specific operations in rice paddies for the first time.
• Market Snapshot — A-share AI chain closed almost entirely lower on 9/10: Cambricon -1.05% (1043.84 RMB), Innolight -2.07% (890.10 RMB), Xinyisheng -0.94%, Foxconn Industrial Internet -2.04%, Hygon Information -1.45%, Kingsoft Office -1.71%. There is no bad news regarding domestic computing power prosperity; it is mostly due to steep prior gains, with funds waiting for overseas verification signals.
• US Stocks Polarized — 9/10 Close: Google +0.59%, Microsoft +0.16% closed higher; NVIDIA -2.26%, Meta -1.42%, Tesla -1.16%, Palantir -2.16%. Those spending on infrastructure rose; those pressured by debt expansion concerns fell. Oracle's after-hours earnings gave bulls new ammunition; tonight, the accounts for computing power stocks need to be recalculated.

---