☀️ TRENDING AI NEWS 🏢 Nvidia Fallout: VCs and founders say the chipmaker's tens of billions in startup investments under Huang may shrink to smaller checks and lower valuations. 🛠️ ChatGPT Wallet: OpenAI's Dots agents can now buy things for US Pro subscribers using American Express, Chase, Mastercard and Visa cards. 🏢 LangChain: The agent startup hit a $3B valuation and now claims 7,000 customers for its LangSmith tool. 🤖 Gemini 4 Hydrogen: Google's new model, 2.5x faster than Gemini 3.8 Flash, carries a 1M-token window and $1.50/$7.50 pricing. |
Fifteen years of AI startup funding rested on one assumption: Nvidia would keep writing checks. This weekend, the people holding those checks' receiving end are doing math on what happens when it doesn't.
| 🛠️ Tool of the day | |
Otter.ai - an AI meeting assistant that records and transcribes your calls in real time, then writes summaries and action items. It helps anyone who spends the day in Zoom, Meet or Teams. It has a free plan.
| ⚡ Quick hits | |
ElevenLabs launched ElevenAI, a preview app that combines meeting notes, dictation in over 90 languages and agent integrations, with early-access sign-up. Read more →
New research says the weights of top Chinese open models carry fingerprints of Nvidia Blackwell training, though it does not prove export rules were broken. Read more →
| 🤓 AI Trivia | |
What does the 'KV' in KV cache, the thing Google's new TPU Sync library moves around, stand for?
The answer is hiding near the bottom of today's newsletter... keep scrolling. 👇
| 🏢 Startups Brace for a Funding Hole After Huang | |

Smaller checks and lower valuations ahead
Under Jensen Huang, Nvidia poured tens of billions of dollars into AI startups. Now, TechCrunch reports, founders and VCs are bracing for smaller checks and lower valuations without him at the helm.
The worry is not abstract. Nvidia has been the most reliable strategic investor in the sector, and a lot of seed and growth rounds were priced with its participation in mind. Israel's startup scene is already treating the turmoil as an opening for challengers, while the Hugging Face CEO has pledged to keep fighting for open source AI even as questions swirl over Nvidia's $12.9 billion acquisition.
Meanwhile, a leaked detail from a separate report says Huang described his interim successor, Colette Kress, as a 'deer in the headlights' in a summer meeting, which does not exactly calm nerves.
The bottom line
If your startup's story leans on Nvidia as a backer or a customer, assume the next round is harder and price your plans accordingly.
Sales Owns Retention, Not Just Closed Deals
Pebl’s Chief Business Officer, Dutta Satadip, joins The GTM Garage to show how Pebl merged Operations and GTM under one leader, and how it aligned goals and incentives across the customer lifecycle.
You’ll also see how Pebl uses AI across that lifecycle and the stack behind prospecting, pipeline, and handoff to onboarding.
15 min conversation. 15 min open Q&A. For Seed to Series A teams.

| 🛠️ ChatGPT Wallet Gives Agents a Credit Card | |

Dots agents can now check out
After launching its Dots agents platform earlier this year, OpenAI is rolling out ChatGPT Wallet to ChatGPT Pro subscribers in the US who are 18 or older, have a US billing address, and use an eligible US-issued American Express, Chase, Mastercard or Visa card.
Add a card and your Dots agents can buy things in response to your requests and check out on your behalf. It is a limited rollout, but it is the clearest sign yet that the big labs want agents handling money, not just drafting emails.
The timing is awkward for one party: United Airlines recently barred AI agents from booking flights, citing cybersecurity risks. Expect more of that tension between agent makers and the sites they want to shop on. For more on where agents are heading, see our AI agents coverage.
The bottom line
Agentic shopping just moved from demo to a real payment rail, so decide now how much spending authority you are comfortable handing a bot.

| 🏢 Amazon Wants to Index Every Product Online | |

The World Product Graph
According to Reuters, Amazon plans to build a vault of data on every product sold anywhere on the internet, including products it does not sell, and surface it through its own site. The project is called the World Product Graph.
The goal is simple: make shoppers less likely to leave Amazon to research products on Reddit or Google, and more likely to end up buying on Amazon. It is a data moat play aimed squarely at AI-driven search, where whoever holds the cleanest product knowledge answers the question.
The bottom line
If Amazon's graph works, product research stops being a place you go and becomes an answer Amazon gives you.

| 🤖 Gemini 4 Hydrogen Targets Speed and Agents | |

Faster, cheaper, and agent-ready
Google released Gemini 4 Hydrogen, a fast agentic model Google says is 2.5x faster than Gemini 3.8 Flash. It offers a 1M-token window, four thinking levels and $1.50/$7.50 pricing per million tokens, with a reported 51 Artificial Analysis score and 57% on Terminal-Bench 4.1.
Want to see what those prices mean for your own workload? Try our Token Calculator. The model is the engine behind Google's Search upgrade for multi-part questions, hard-to-parse typos and highly specific requests.
The bottom line
A 1M-token window at these prices makes Hydrogen a serious default for agent workloads where latency and cost matter more than peak reasoning.

| ⚡ Google Open-Sources TPU Sync for KV Caches | |

Moving memory across TPU tiers
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 MoE models.
This is the unglamorous plumbing that decides how many users a model can serve at once. Cutting transfer latency that sharply lets serving stacks park and recall cached context without stalling, which matters most for long-context agent sessions.
The bottom line
Unless you run TPU clusters this is not for you directly, but it lowers the cost floor for everyone who serves long-context models on Google hardware.

| 🏢 LangChain Hits a $3B Valuation | |

LangChain now claims 7,000 customers for LangSmith, its model and platform agnostic agent management tool, and has reached a $3 billion valuation. Being agnostic is the pitch: as companies mix models from several labs, the tooling that sits above them becomes the sticky layer.
The bottom line
Investors are betting the agent management layer outlives any single model, and a $3B price tag says that bet is crowded.

| 🛠️ Vercel Gives Coding Agents a UI Toolkit | |

Vercel Labs released the Vercel UI SDK, an Apache-2.0 beta library of 50+ UI behaviors that coding agents compose into accessible components for React, Vue, Svelte, Solid and Web Components, with a bridge for direct UI control. The idea is to stop agents from reinventing dropdowns and dialogs badly every time.
If you are building quick landing pages rather than full apps, our sponsor 60sec.site is an AI website builder that gets you live in minutes. And for more daily AI news like this, visit dailyinference.com.
The bottom line
Shipping accessible behaviors as building blocks is a smart way to make agent-written front ends less fragile.

| 🌎 Trivia Reveal | |
The answer is Key-Value! During generation, a transformer stores the key and value vectors for tokens it has already processed so it does not recompute them. That stored memory is the KV cache, and shuffling it efficiently is exactly what Google's TPU Sync is built to do.
| 💬 Quick Question | |
Would you let an AI agent spend your money with a card in a wallet like ChatGPT's? Hit reply and tell me where your comfort limit is - I read every response!
That's all for today - see you tomorrow with more.
| 🎁 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.


