Nvidia’s Biggest AI Bet Yet, Meta Doubles Down on Agents, and a New Way to Train Coding AI

FEATURED INSIGHT💡

Nvidia’s $20B Groq deal: the biggest signal yet that inference is the battleground

Nvidia has reportedly agreed to buy Groq’s assets for about $20B in cash, marking Nvidia’s largest deal ever. Groq is known for low-latency inference hardware and was founded by engineers tied to Google’s TPU work—so this isn’t just a “startup exit,” it’s a direct move in the AI compute arms race. Nvidia’s Jensen Huang told employees the plan is to integrate Groq’s low-latency processors into Nvidia’s “AI factory” architecture to serve more real-time inference workloads.

What makes this interesting is the structure: Groq framed it as a non-exclusive licensing agreement for inference technology, while also confirming that Groq’s founder/CEO Jonathan Ross and other senior leaders will join Nvidia. Groq says it will remain an independent company and that GroqCloud will continue operating. Translation: Nvidia gets technology + talent without “acquiring Groq as a company,” which mirrors a recent pattern of big players using licensing + hiring deals as the fastest route to capability.

Your competitors are already automating. Here's the data.

Retail and ecommerce teams using AI for customer service are resolving 40-60% more tickets without more staff, cutting cost-per-ticket by 30%+, and handling seasonal spikes 3x faster.

But here's what separates winners from everyone else: they started with the data, not the hype.

Gladly handles the predictable volume, FAQs, routing, returns, order status, while your team focuses on customers who need a human touch. The result? Better experiences. Lower costs. Real competitive advantage. Ready to see what's possible for your business?

See it in action

ON THE HORIZON 🌅

Manus joins Meta: agents are becoming the execution layer for everything

Manus announced it’s joining Meta, calling the move validation of its work building “General AI Agents” that handle research, automation, and complex tasks. Manus claims its agent has already processed 147 trillion tokens and powered 80 million virtual computers, positioning itself as an “execution layer” that can reliably carry out end-to-end work.

Manus says the product won’t be disrupted: subscriptions will continue through its app and website, the company will keep operating from Singapore, and Meta plans to keep Manus running independently while integrating its agents into Meta’s platforms. The statement also notes Manus will sever ties with China, likely to reduce regulatory friction as the product moves deeper into major consumer and business ecosystems.

LATEST IMPORTANT NEWS 📰

Lemon Slice raises $10.5M for interactive video avatars

Lemon Slice raised $10.5M (YC, Matrix, and others) to build diffusion-model avatar tech that can generate a digital avatar from a single image and run it on top of a knowledge base for roles like customer support, tutoring, or training—aiming to finally beat the “uncanny valley” problem that has held avatars back.

New York signs the RAISE Act

New York governor Kathy Hochul signed the RAISE Act, requiring large AI developers to publish information about safety protocols and report safety incidents to the state within 72 hours, with fines up to $1M (and $3M for subsequent violations). It also creates a new monitoring office inside the Department of Financial Services.

OpenAI updates teen safety rules for ChatGPT

OpenAI updated its Model Spec for under-18 users, including stricter rules around romantic/sexual roleplay, added caution around body image and disordered eating, and guidance that models should prioritize safety when harm is involved—paired with upcoming age-prediction to automatically apply teen safeguards.

Why AI Isn’t Replacing Affiliate Marketing After All

“AI will make affiliate marketing irrelevant.”

Our research shows the opposite.

Shoppers use AI to explore options, but they trust creators, communities, and reviews before buying. With less than 10 percent clicking AI links, affiliate content now shapes both conversions and AI recommendations.

Download the full report

FOR THE TECHNICALLY INCLINED 🛠️

Self-play SWE-RL: a coding agent that learns by breaking code, then fixing it

Meta introduced Self-play SWE-RL, a reinforcement learning approach where one model plays both sides of software work: it injects bugs into real repositories, then switches roles and fixes them using tests as the only source of truth—no natural language issues, no human labels, just executable feedback. The paper reports improvements of +10.4 on SWE-Bench Verified and +7.8 on SWE-Bench Pro, outperforming a baseline trained on human issue data across training.

All the stories worth knowing—all in one place.

Business. Tech. Finance. Culture. If it’s worth knowing, it’s in the Brew.

Morning Brew’s free daily newsletter keeps 4+ million readers in the loop with stories that are smart, quick, and actually fun to read. You’ll learn something new every morning — and maybe even flex your brain with one of our crosswords or quizzes while you’re at it.

Get the news that makes you think, laugh, and maybe even brag about how informed you are.

Try it for free

AI TOOL OF THE DAY 🚀

Sketch To is a simple AI tool that converts images to sketches (and sketches back to images) with a standard model and a higher-detail “Pro” option.