☀️ TRENDING AI NEWS 🚨 Nvidia Fallout: US Treasury chief Scott Bessent is blaming Howard Lutnick's Commerce Department for inaction on Nvidia's alleged Chinese spy link. 🤖 China Models: New research shows China's most advanced AI models were trained on export-banned Nvidia chips. 🏢 LangChain: The agent startup hit a $3B valuation, claiming 7,000 customers for its LangSmith tool. 🛠️ ElevenAI: ElevenLabs launched a meeting notetaker and dictation app aimed at teams and enterprises. |
One question. 160 of them, actually, and he got 159 right. That's how a South Korean tax exam topper made everyone suspicious, and today he admitted why.
Meanwhile, the Nvidia saga keeps widening. Two days after Jensen Huang's exit, the story is less about one executive and more about who knew what in Washington, and whether the chips he sold ended up where they shouldn't have.
| 🛠️ Tool of the day | |
Zapier - connects your apps so an event in one app starts actions in another, with no code. It helps anyone who wants to automate repeat work. It has a free plan.
| ⚡ Quick hits | |
OpenAI and Databricks explain how firms can use ChatGPT, Codex and GPT models on governed Databricks data. Read more →
| 🤓 AI Trivia | |
The 'KV cache' that Google's new TPU Sync library moves around stores what, exactly, during LLM inference?
The answer is hiding near the bottom of today's newsletter... keep scrolling. 👇
| 🚨 Nvidia Scandal Reaches the White House | |

Treasury and Commerce start pointing fingers
The fallout from Jensen Huang's resignation from Nvidia is now a Washington story. Treasury Secretary Scott Bessent has publicly questioned Commerce Secretary Howard Lutnick's decisions, and Lutnick is reportedly facing growing pressure, with questions emerging about what senior Trump officials knew of the alleged Chinese spy's ties to Huang.
The security angle is just as awkward. Nvidia's security chief quietly left in August, and a former intelligence officer says the kind of leak at issue runs through everyday conversations that most corporate security teams are poorly equipped to catch. Separately, a petition urging Nvidia to reinstate Huang had passed 115,000 signatures Saturday, with Elon Musk among the signers.
The bottom line
This stopped being a personal scandal and became a test of how the US polices AI chip exports. Expect the real consequences to land in policy, not in the boardroom.
Attio is the #1 fastest-growing CRM on Brex and Ramp
Built for the most ambitious companies in GTM, Attio is an end to end revenue platform that allows teams to run multiple plays across different revenue motions and lifecycles.
Teams at the frontier are already building on it: Granola, turbopuffer, Modal, and many many more.

| ⚠️ Banned Nvidia Chips Power China's Best Models | |

Export controls meet an awkward paper trail
New research indicates the most advanced AI models from China were trained on Nvidia chips that were banned from export. That lands at the worst possible moment for a company already under scrutiny over alleged chip diversion.
Governments across Asia are already reacting to Huang's exit: some moved to confirm chip supply commitments, while others are monitoring the impact on local projects. For anyone tracking AI regulation, the gap between export rules on paper and chips in practice is the number to watch.
The bottom line
If banned chips trained China's frontier models, tighter enforcement is now politically hard to avoid, and that is a supply risk for everyone building on Nvidia hardware.

| 🏢 Amazon Wants a Vault of Every Product Online | |

Keeping shoppers from leaving for Reddit
According to Reuters, Amazon is building a "World Product Graph," a database covering every product sold anywhere on the internet, including items Amazon doesn't sell. The information would be served through Amazon's own site.
The logic is simple: if shoppers do product research on Reddit or Google, they buy elsewhere. If Amazon answers those research questions itself, they stay and buy there. It's an AI search play wearing a retail hat.
The bottom line
Product research is becoming the battleground for AI search. Whoever owns the comparison step owns the purchase.

| 🛠️ LangChain Hits $3B as Agent Tooling Heats Up | |

The AI agents startup LangChain reached a $3B valuation, claiming 7,000 customers for LangSmith, its model and platform agnostic agent management tool. The pitch is that companies running many agents across many models need one place to monitor and manage them.
The same day, Vercel Labs released an Apache-2.0 beta UI SDK with 50+ UI behaviors that coding agents can compose into accessible components across React, Vue, Svelte and Solid. And if you need a quick site for your own project, our sponsor 60sec.site will build one with AI in about a minute. For more daily AI news, visit dailyinference.com.
The bottom line
The money is flowing to the plumbing around agents, not just the models. Management and tooling layers are where the next wave of enterprise spending is landing.

| ⚠️ Top Scorer Admits Using AI to Cheat on a Tax Exam | |

The top scorer on South Korea's national exam for aspiring tax accountants missed just one of 160 questions. On Sunday he confessed to using generative AI, after months of speculation that the result was too good to be true. He admitted sending photos of questions to an AI tool.
The timing is awkward: there are fears about the national college entrance exam next week, and this case shows how hard it is to stop test takers from feeding questions to a chatbot.
The bottom line
High-stakes testing without strict device control is now effectively open book. Exam designers, not students, are the ones who have to adapt.

| ⚡ Google's TPU Sync Slashes KV Cache Transfer Time | |

Google released TPU Sync, an open-source C++ library that moves KV caches and weights across TPU HBM, host DRAM, storage and network. Google reports over 90% lower KV transfer latency on 235B to 480B parameter mixture-of-experts models.
The bottom line
Faster cache movement means cheaper, snappier serving for huge models. It is unglamorous infrastructure work that directly lowers inference costs.

| 🌎 Trivia Reveal | |
The answer is attention keys and values for tokens already processed! The KV cache saves those so the model doesn't recompute them for every new token, which is why moving it quickly, as TPU Sync does, matters so much for inference speed.
| 💬 Quick Question | |
Would you let an AI agent spend your money on its own if it had a spending cap? Hit reply and tell me where your comfort line is - I read every response!
That's it for today - see you tomorrow!
| 🎁 Share Daily Inference | |
Know someone who wants AI news in five minutes a day? Share your link. One referral unlocks the AI Tools Starter Kit. Five referrals unlock the AI Insider Briefing.


