鄭哥，90年代大學畢業，26年橫跨電子，家電、服裝、珠寶、教育、AI算力多個行業。做過中學教師、翻譯、外贸，現在搞七善門科技和星辰算力。口頭禪：「點解唔得？」Hany, university grad from the 90s, 26 years across electronics, appliances, fashion, jewelry, education, and AI compute. Former teacher, translator, trader. Now running Qishanmen Tech and Star Compute. Catchphrase: "Why not?"
After 26 years in export, I've seen too many "some celebrate, some grieve" situations.
But this time, what's interesting is —
The same Muse made chip manufacturers laugh all the way to the bank; e-commerce giants immediately slammed the door saying "you're not welcome"; and Chinese LLM giants collectively shut down agent features — three outcomes, three perspectives, pointing to the same question:
Is an AI agent a tool, or a threat?
Act 1: Muse Made Chip Makers Soar — But This Money Isn't for Everyone
Meta's AI agent Muse, launched September 8, hit 2.8 million downloads by the 12th, topping both iOS and Google Play free charts in the US. Meta stock surged 11.43% in a single day, adding approximately $207 billion in market cap — roughly equal to one Netflix.
Guotai Junan Securities estimates: if Muse reaches 100 million users, hardware investment needs ~$30–50 billion, approximately 1.58 million AMD EPYC server chips, corresponding to ~1.6 gigawatts of power consumption.
Then the supply chain chain reaction:
CPU-to-GPU ratio shifting from 1:4-1:8 to 1:1-1:2 — CPU demand exploding
Intel, AMD products raising prices
NVIDIA, Arm entering server CPU赛道
FCBGA ABFIC substrate prices and volumes both rising
Chip makers are laughing all the way to the bank. But I ask: who gets to enjoy this money?
— Those with the capacity to scale production first. Small and medium factories? Orders booked 18 months out, can't keep up.
Act 2: E-Commerce Closes Its Doors — "You're Not Welcome Here"
On the same day Muse's downloads exploded — Amazon blocked it.
September 20: Amazon officially prevented Muse from purchasing on their platform on behalf of users. Reason: unauthorized, failure to proactively disclose AI identity, account credential scraping, and data security concerns.
Over the past year, Amazon has systematically blocked external AI agents: sued Perplexity's Comet browser; blocked Google and OpenAI shopping agents; Anthropic updated its terms of service in April 2026, comprehensively blocking third-party AI tools from accessing its models via OAuth.
場景：亞馬遜封殺AI智能體Scene: Amazon blocks AI agents — the battle for platform control Image AI-assisted (Z-Image-Turbo)
Platforms are building their own walls. But the deeper reason isn't just security — it's话语权.
When a third-party AI can make purchases on behalf of users on your platform, the platform loses control over the user's decision path. Ad targeting logic, search ranking logic, user behavior data — all can be bypassed. This is the last thing e-commerce platforms want to see.
Act 3: Tencent Shuts Down Agents — Forced by Regulation
On the Chinese side, the plot is completely different.
June 30, 2026: Tencent Yuanbao comprehensively removed its self-built agent entry point. Doubao and Qianwen announced agent features would be fully taken offline on July 15.
場景：騰訊關停AI智能體Scene: Tencent shuts down AI agents — forced by regulation Image AI-assisted (Z-Image-Turbo)
Broker analysis: "The underlying architecture of general LLM main products struggles to adapt to new regulations' massive UGC scenarios. Restructuring the entire risk control system to retain features would require long development cycles and enormous capital investment. Compared to the cost of continuously exposing compliance risks, shutting down the main station entry and splitting to independent carriers is the solution with lowest comprehensive cost and controllable risk."
Tencent shut down agents not because Muse threatened it, but because the regulatory environment forced it to switch tracks.
Act 4: The Pain Point — How Hard Is AI Agent Commercialization Really?
This is the core of what I call the "pain point."
Muse made chip makers soar, but:
If e-commerce platforms collectively say "you're not welcome," Muse's monetization path gets cut off at the waist.
Amazon's ban is just the first shot. Imagine: if Google, Meta, Taobao, JD.com all say "no unauthorized agents allowed," Muse's proud "shop on your behalf" feature becomes a castle in the air.
Meta's attempted solution: building a "partner merchant pool" — Walmart, Best Buy, Gap, Sephora, Wayfair, Expedia, Instacart and others announced integration with Muse. But this solution has one prerequisite: Amazon is not included.
Looking at it from Tencent's side — regulation forced them toward enterprise task-oriented agents: industrial quality inspection, financial data analysis, government document processing, enterprise private knowledge bases. This is another path: abandon C-end emotional agents, focus on enterprise compliance tools.
One-Line Summary
AI agents face this reality: compute factories are celebrating, e-commerce platforms are building walls, regulators are tightening the net. Under three-way pressure, agents' monetization path is still being explored — but the opportunity belongs to whoever can coordinate these three parties.
關於作者About the Author
鄭哥，90年代大學畢業，26年橫跨電子，家電、服裝、珠寶、教育、AI算力多個行業。做過中學教師、翻譯、外贸，現在搞七善門科技和星辰算力。口頭禪：「點解唔得？」Hany, university grad from the 90s, 26 years across electronics, appliances, fashion, jewelry, education, and AI compute. Former teacher, translator, trader. Now running Qishanmen Tech and Star Compute. Catchphrase: "Why not?"
