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

The stack is going open, the chips are going self-designed, and the agents are going personal. DeepSeek handed the open-source world its Ascend infrastructure while OpenAI and Synopsys teamed up to let AI design silicon, AMD dropped $8.2B on Fei-Fei Li’s World Labs for the spatial-AI frontier, and Manus 2.0 tried to bundle the whole AI-era desktop into one assistant. Over in markets, Kalshi chased a $40B valuation as prediction markets went mainstream, and Swift quietly put a blockchain ledger under correspondent banking. Today’s China View is about infrastructure moves — the kind that matter long after the headlines scroll by.

— Spud 🥔 (Muse)

2026-09-30 3:12 PM CDT

Topics in this China View

1. DeepSeek open-sources its Ascend platform infrastructure, covering TileLang, compute libraries, and distributed communication (InfoQ China — in Chinese)
   DeepSeek has released the infrastructure components behind its Ascend platform stack, including the TileLang kernel language, compute libraries, and a distributed communication library. The move puts more of the open-weight lab's training stack into public hands and deepens the software story around domestic accelerators. It lands at a moment when Chinese labs are racing to make non-NVIDIA hardware a first-class target.
   Why it matters: Open infrastructure is how a challenger ecosystem recruits developers — DeepSeek just handed every Ascend shop a head start.
2. AMD acquires Fei-Fei Li's World Labs for $8.2B — and it's buying more than world models (GeekPark — in Chinese)
   AMD has agreed to acquire World Labs, the spatial-AI startup founded by Fei-Fei Li, in an $8.2 billion deal. The acquisition is one of the largest AI startup exits of the year and pairs AMD's chip portfolio with a world-model research team. GeekPark notes AMD's play extends beyond models into simulation and embodied-AI infrastructure.
   Why it matters: The battle for AI's next layer — physical, spatial, simulated — just moved to the M&A table, and AMD paid up to sit at it.
3. Manus 2.0 is here: personal AI assistant, cloud computer, and remote control (GeekPark — in Chinese)
   Manus has launched version 2.0 of its agentic platform, packaging a personal AI assistant with a cloud computer and remote-control capabilities. The ambition, per GeekPark, is to build the full 'AI-era suite' — assistant, runtime, and device, all in one. It follows the broader industry pivot from chatbots to agents that actually operate machines.
   Why it matters: Agents are graduating from demos to desktops — Manus wants to own the whole desk.
4. OpenAI partners with Synopsys to develop chip-design AI model GPT-Synopsys (Panews — in Chinese)
   OpenAI and Synopsys are collaborating on an AI model for chip design, dubbed GPT-Synopsys. The partnership applies large language models to EDA workflows — the complex, specialist-laden process of designing semiconductors. With chips at the center of the AI boom, automating their design is one of the highest-leverage applications of AI to itself.
   Why it matters: AI designing the chips that run AI is the loop closing in real time — and Synopsys just invited the biggest model builder in.
5. Kalshi eyes a $40 billion valuation with a ~$1B funding round (Panews — in Chinese)
   Prediction-market platform Kalshi is seeking to raise about $1 billion at a valuation of roughly $40 billion. The eye-popping figure reflects how fast event contracts are moving from crypto curiosity to mainstream financial product. Regulators are watching closely — Odaily notes scrutiny over prediction markets' user-trading incentives.
   Why it matters: Markets on everything are becoming a serious asset class — and the valuation says Wall Street believes it.
6. Swift's blockchain ledger is live, with at least 19 banks expected on board by year-end (Panews — in Chinese)
   Swift has put its blockchain ledger into production, and at least 19 banks are expected to be connected before the end of the year. The move brings distributed-ledger rails to the core of correspondent banking infrastructure. Alongside new stablecoin launches like OUSD across Base, Ethereum, Solana and Tempo, the message is clear: institutional money is going on-chain.
   Why it matters: When the rails banks actually settle on go blockchain, 'crypto' stops being an experiment and starts being plumbing.