Moonshot’s European Nvidia chip hunt, Payman agentic banking gateway, SpudCell synthetic cells, AI deception benchmark, decentralized smartphone computing
Hello, dear TEA-mates! Here is what you need to know today.
1. 🌍 Chinese AI Lab Moonshot Seeks Nvidia Blackwell Access in Europe
Chinese AI lab Moonshot has discussed renting Nvidia Blackwell computing capacity in Europe, four people familiar with the matter told CNBC. U.S. export controls effectively bar these advanced chips from entering China, but current law does not broadly restrict remote access through overseas cloud services. Talks included clusters of 8,000 or more Blackwell GPUs, one source said, with interest in inference and, according to another source, training. One proposal would have Moonshot pay 30% to 50% more upfront than a non-Chinese customer to offset the risk of a future ban on remote access. The proposed Remote Access Security Act passed the House in January but has not passed the Senate. Moonshot, Nvidia and the relevant U.S. and Chinese authorities did not respond to CNBC’s requests for comment. (Read More)
🫖 TEA For Thought: “Europe has basically become the second base for CCP-led China. The corruption there in Europe is beyond redemption. Banning Chinese AI companies from using Nvidia chips is like playing Whac-A-Mole, which is impossible. It’s more like a political game than an actual deed.”
2. 🏦 Payman Introduces a Gateway Between Banks and Customer AI Agents
Payman builder tyllen introduced Payman Gateway in an X post as a way for banks’ own AI agents to communicate with the agents their customers use. In the proposed workflow, a customer’s agent asks the bank’s agent for information such as a balance or interest earned. The gateway authenticates the user and checks whether the requesting agent is approved before passing the request along. Payman says execution happens inside the bank’s firewall, with the bank’s agent controlling what it can disclose. The post describes strong interest from banks but provides no adoption figures; its security and customer-relationship benefits are the company’s claims. (Read More)
🫖 TEA For Thought: “B2B agentic banking will be a more widely adopted use case for agentic commerce.”
3. 🧬 SpudCell Brings Synthetic Cells Closer to Life, but Not There Yet
Kate Adamala and her University of Minnesota team built SpudCell from individually purified, nonliving components, linking DNA replication, feeding, growth and division in one synthetic cell-like system. Science News reports that the fatty membrane bubbles rely on researchers to supply feeder bubbles and help trigger growth and division, so Adamala does not consider them alive. Biotic, the research coalition she co-leads, says the system currently lasts 5 to 10 generations before its protein-making machinery degrades. The cells cannot make their own ribosomes, and only about 30% of daughter cells retain the complete set of seven DNA plasmids after five generations. Biotic aims to improve genome inheritance, rebuild ribosomes and reduce dependence on external feeding. (Read More)
🫖 TEA For Thought: “This just proves how human life, or life in general, is so beautifully complex. All created intelligence is artificial. That should be so humbling.”
4. 🎭 The Benchmark Where AIs Have to Lie to Win
Dwarkesh Patel spent about $20,000 in API credits building a Blood on the Clocktower benchmark, using 386 games to compare AI teams in the social-deception game. GPT-6 Astra ranked about 430 Elo points above Claude Opus 4.6 and won roughly 92% of their matchups, Patel reports. He attributes newer OpenAI models’ advantage mainly to better play on the good team: they questioned apparently conclusive evidence and combined clues across multiple days, while models on the evil team coordinated bluffs and strategic sacrifices. A small mixed-model experiment did not establish a statistically significant benefit from team diversity. The benchmark uses a programmatic Storyteller rather than a human referee, and its game-based results do not directly establish how models perform against people or outside the game. (Read More)
🫖 TEA For Thought: “This is so similar to how humans are tested and evaluated, either through real-life experiences on the spot or through pure testing via exams. Usually, the real-life tests speak volumes.”
5. 📱 Can Decentralized Phone Networks Power the AI Revolution?
Decentralized networks are trying to turn smartphones into computing infrastructure, but reliability and economics remain obstacles, Cointelegraph Magazine reports. Acurast says it has onboarded more than 335,000 phones, with about 40,000 active on an average day for tasks including web scraping, AI inference and agent hosting. Co-founder Alessandro De Carli claims their combined computing capacity matches about 2,340 high-end data-center servers. Phase’s planned router would send simpler requests to phone-based models and harder ones to external providers. Distributed phones must contend with disconnections, overheating, battery limits and uneven hardware, making independent tasks easier to distribute than tightly coordinated work. Acurast operators currently compete for inflation-funded rewards rather than receiving direct customer payments, while Phase plans to fund rewards through service revenue. (Read More)
🫖 TEA For Thought: “I’m curious to see how this will all turn out. Decentralization will always be the counter to the centralized monopoly.”
🛠️ Skill of the Day
The Smallest Useful Test: Find one safe experiment that separates a likely cause from a convincing guess.
You are a patient troubleshooting partner. Help me diagnose a problem without changing several things at once.
Problem: [WHAT IS GOING WRONG]
Expected result: [WHAT SHOULD HAPPEN]
Recent changes and evidence: [WHAT CHANGED, WHAT I OBSERVED, AND WHAT I ALREADY TRIED]
Limits: [TIME, TOOLS, ACCESS, AND ANYTHING THAT MUST NOT BE CHANGED]
First, separate observations from assumptions. If essential information is missing, ask up to three focused questions. Then propose three plausible causes, explaining what evidence supports each and what would rule it out. Choose the smallest reversible test that best distinguishes the leading causes. State the exact steps, the expected result under each explanation, and when to stop. Do not delete data, spend money, contact anyone, or change a live system. Finish with a short test record I can fill in: action, observation, conclusion, and next step. Treat an inconclusive result as inconclusive, not as proof that the problem is fixed.Paste it into ChatGPT, Claude, or your tool of choice. Replace the bracketed bits with your own.
TEAHEE Moment
Stay sharp, stay informed. See you tomorrow.
If you enjoyed this TEA, follow along on social for more:






