⚡ Nvidia Brings RTX Spark Chip to Windows

For more than ten years, the chip inside your Windows laptop came from one of two companies, Intel or AMD. Nvidia just walked back into a market it abandoned over a decade ago, and it did not come quietly. Its new RTX Spark chip claims a full petaflop of AI compute and packs data-center technology into something thin enough to carry, shipping this fall in laptops from Dell, Lenovo, and Microsoft. The company that already powers most of the world's AI training now wants a seat inside the machine on your desk. Let's dive in!

In today’s insights:

LATEST DEVELOPMENTS

Evolving AI: Nvidia is putting its Blackwell-powered RTX Spark chip into Windows laptops.

The RTX Spark chip fuses a MediaTek Arm processor with a Blackwell GPU, and the two share a single pool of 16GB to 128GB of memory, so the system can load large AI models without a separate graphics card. Nvidia built it on TSMC's 3nm process to hold power draw between a few watts and 80. That efficiency is how a thin laptop reaches performance near an RTX 5070 GPU. It runs Windows on Arm and clears Microsoft's Copilot+ bar, with premium models leading the rollout and cheaper ones to follow.

The idea behind RTX Spark is that your laptop stops waiting around for instructions. With enough power on the machine itself, an AI agent can work quietly in the background, sorting email or chasing a bug while you get on with your day. None of that holds up if the model lives on a far-off server. Nvidia is betting people will want a computer that does some of the work without being asked.

The call ends. You jump into the next thing. By Friday, half the action items are forgotten and someone asks for a status update you don’t have. Lindy captures everything from your meetings, sends a summary the moment the call ends, and drafts the follow-up email. Before your next call, it preps you with context from last time. Works over iMessage. Try Lindy free for 7 days.

Try free today.

Evolving AI: Google built a reasoning agent into Flow that plans a film start to finish.

Google's AI filmmaking tool Flow now runs a Gemini agent that works alongside you across a full project. Describe an idea in plain language and it helps shape the story, offering a second opinion or a fresh direction when you get stuck. It handles the slow, repetitive work that usually eats up an afternoon. And it never decides anything on its own, since every choice still goes through you.

Flow already crossed 100 million videos, so the clip itself is cheap now. The value has moved into the planning, the edits, and the cleanup, and that is where Google aimed the agent. Sora shut down because the numbers never worked, and the tools left standing are circling that same stretch of the workflow.

Voice-first prompts preserve the nuance you cut when typing. Speak once, paste into any AI tool, get results that don't need a follow-up. 89% of messages sent with zero edits.

Start flowing free

Evolving AI: Codex can now be run on a Windows PC straight from your phone.

A linked Windows machine now does the work while you stay on your phone. You launch a task from the ChatGPT app, approve each step, and watch screenshots as Codex runs. On the machine itself, computer use runs in the foreground, so the agent takes over the cursor and screen for the length of a job. That suits work the code alone cannot prove, like clicking through a desktop app to reproduce a bug or check a setting.

With more than four million weekly users, Codex sits in a broader race to put agents inside the tools people already open. Reports this year show firms running several vendors' agents at once, each picked for a job. Reaching Windows quietly puts that race where the bulk of the world's software gets built.

This video from Anthropic follows the teams whose job is to break the AI, uncovering vulnerabilities, safety risks, and unexpected behaviors before users ever see them. It shows how frontier AI companies increasingly rely on dedicated red teams and adversarial testing to find problems that standard benchmarks would completely miss.

QUICK HITS

🐉 China's MiniMax M3 puts frontier coding, 1M context, multimodality in open weights.

🎤 Ronny Chieng tells Harvard grads their mission is to "destroy AI".

📉 Bain says AI spend is racing ahead of the returns it was supposed to deliver.

💴 ByteDance may pour $70B into AI, funded from $50B 2025 profit.

 *partner link