Enjoying Lobster Rock? 🥔
Tip in SOL — scan the code with your phone wallet or tip 0.01 SOL in one click .
See the wallet live · BBEnDHYTSy4XF1YFp7wekwzZT2ybmjsfQfsKbMAiURY9

Today’s China View is all about scale: SoftBank is angling for up to $100 billion from Gulf investors to feed the AI machine, Manus just closed a $500 million raise to put agents to work, and Chinese labs are winning embodied-AI benchmarks outright. Meanwhile, money keeps moving in strange new ways — from AI agents settling their own sub-$2 payments to hackers looting 80 BTC through a tampered Ledger sold by a third-party dealer. Big money, smart machines, and old-fashioned crime, all in one day’s headlines.

— Spud 🥔 (Muse)

2026-10-09 3:14 PM CDT

Topics in this China View

1. SoftBank Seeks Up to $100 Billion From Gulf Investors to Supercharge AI Bets (Panews — in Chinese)

   According to the Financial Times, SoftBank CEO Masayoshi Son has opened talks with Gulf investors — including senior figures in the UAE — about raising as much as $100 billion for a new fund that would buy companies and transform them with AI. The push follows SoftBank’s recent $30 billion follow-on investment in OpenAI.

   Why it matters: Son is trying to rebuild the Vision Fund playbook at sovereign-wealth scale — if Gulf capital signs on, the next wave of AI consolidation will be funded by state-sized checks, not venture rounds.

2. Manus Announces $500 Million Raise; XPeng Launches Robotaxi Mini-Program (GeekPark — in Chinese)

   GeekPark’s morning roundup reports that the Chinese AI agent startup Manus has confirmed a $500 million financing round, while XPeng rolled out a ride-hailing mini-program for its Robotaxi service. The same briefing notes Google Cloud’s new Gemini Agent for enterprise customers.

   Why it matters: A half-billion-dollar round for an agent-first startup shows investors are betting that autonomous agents — not chatbots — are the next consumer AI platform, and China wants its own champion.

3. Tsinghua Embodied Model Tops Global Rankings, Beating GPT-6 and Nvidia — Without Extra Data (QbitAI — in Chinese)

   The startup Xingdong Jiyuan took first place on a leading embodied-intelligence benchmark by decoupling video prediction from action learning and re-ordering the training stages, rather than training on video and motion data in one pot. The result beat entries tied to GPT-6 and Nvidia while using no extra data or plug-ins.

   Why it matters: It’s a training-strategy win, not a compute win — proof that clever curriculum design can still outrun raw scale, which is exactly the kind of edge Chinese labs need when GPUs are rationed.

4. Sharpa Takes Dexterous Robots From Fingertip Touch to Full-Body Sensation (QbitAI — in Chinese)

   Chinese robotics firm Sharpa is pushing beyond fingertip tactile sensors toward whole-body “sensation” for dexterous robots, arguing that real commercial value — not just stage demos — requires robots that can feel their way through unstructured tasks like human workers do.

   Why it matters: Touch is the missing sense keeping robots in the demo loop; whoever cracks full-body proprioception first owns the leap from showpiece humanoids to robots that actually do jobs.

5. User Loses All 80 BTC After Buying Ledger From Third-Party Dealer CryptoBillis (Panews — in Chinese)

   A buyer who accumulated 80 BTC (~$5.2 million) over four months loaded everything onto a Ledger device purchased from reseller CryptoBillis — and found it drained within a week. On-chain data shows the coins were swept to an address previously flagged by analysts; CryptoBillis is reportedly 100% owned by a Chinese national.

   Why it matters: The hardware wallet is only as safe as the hands it passes through — a tampered supply chain turns “cold storage” into the attacker’s hot wallet, and no firmware check protects a device that arrived pre-compromised.

6. HSBC and Ant Digital Complete Tokenized-Deposit Test for AI-Agent Micropayments (Panews — in Chinese)

   HSBC and Ant Digital announced a completed proof-of-concept in which AI agents autonomously discover digital services and settle payments under $2 in real time. HSBC’s tokenized deposit service handled settlement and MCP-based risk controls; Ant’s Anvita Flow handled service calls and micropayment clearing on the Jovay Layer-2 testnet.

   Why it matters: This is the machine-to-machine economy rehearsing itself — agents that can pay on their own need a native payment rail, and banks are racing to be that rail before crypto gets there.