🔒 Google launches Private AI Cloud

Good morning, AI enthusiast. Google just unveiled its own version of Apple’s Private Cloud Compute, and it could redefine how privacy works in the AI era.

In today’s AI newsletter:

• Google launches “Private AI Compute”

• Yann LeCun to exit meta and launch AI startup

• SoftBank cashes out of Nvidia to fund OpenAI

• How to generate a complete product blueprint for your AI projects

• AI tools & more

AI NEWS

☁️ Google Launches “Private AI Compute”

Google’s new Private AI Compute platform lets devices tap into cloud-based AI power while keeping user data private. The system processes sensitive data in what Google calls a “secure, fortified space”, where not even Google itself can access it.

• Bridges the gap between on-device AI and cloud power for heavy reasoning tasks

• Data used in Private AI Compute stays encrypted and inaccessible to Google

• Powers new features like smarter Magic Cue suggestions and expanded Recorder transcription

• Debuts on Pixel 10 devices, with more integrations across Android and Chromebook planned

• Built to rival Apple’s Private Cloud Compute, blending performance with privacy

As AI models demand more compute, tech giants are racing to balance power with privacy. Google’s Private AI Compute could set a new standard, letting users benefit from advanced cloud AI without sacrificing control over their personal data.

AI TALENT

🧠 Yann LeCun to Exit Meta and Launch AI Startup

Yann LeCun, Turing Award winner and one of the pioneers of deep learning, is planning to leave Meta after more than a decade to launch a new venture focused on “world models”, a next-generation AI architecture designed to reason and understand the physical world like humans do.

• LeCun has led Meta’s Fundamental AI Research Lab (FAIR) since 2013

• His departure follows Meta’s pivot away from long-term research toward faster AI product rollouts

• Mark Zuckerberg recently hired Scale AI founder Alexandr Wang to lead a new “superintelligence” team with a $14.3B deal and 49% stake in Scale

• LeCun’s “world model” research contrasts with Meta’s LLM-focused approach, which he has called “useful, but not truly intelligent”

• Meta’s recent Llama 4 model underperformed, and its chatbot adoption has lagged behind rivals like OpenAI and Anthropic

LeCun’s exit underscores a philosophical rift within Meta, between short-term AI deployment and the pursuit of true artificial reasoning. His next startup could reignite the race toward human-level intelligence built on AI that learns from the real world, not just words.

AI PARTNERSHIPS

💰 SoftBank Cashes Out of Nvidia to Fund OpenAI

SoftBank sold 32.1 million Nvidia shares in October, raising billions to support its $22.5 billion investment in OpenAI and other AI ventures, including robotics and data center projects.

• Nvidia stake sold for $5.83B, plus $9.17B from T-Mobile shares

• Funds to fuel SoftBank’s “all-in” AI strategy, not a retreat from Nvidia

• $22.5B going directly into OpenAI, raising SoftBank’s ownership from 4% to 11%

• Additional $6.5B set aside for Ampere and ABB’s robotics unit acquisition

• Vision Fund posted a $19B gain, doubling SoftBank’s quarterly profit

This move cements SoftBank’s transformation into an AI-first investment powerhouse. While exiting Nvidia may seem surprising, the company is rebalancing, turning chip profits into fuel for the next AI frontier led by OpenAI, robotics, and massive data infrastructure.

HOW TO AI

💻 How to Turn Any Photo and Voice Clip into a Realistic Talking Video

In this tutorial, you’ll learn how to create ultra-realistic talking videos from just a photo and audio using Humo, a free and open-source AI that combines text, image, and voice to produce lifelike facial motion, lip-sync, and emotion, all locally on your computer.

🧰 Who is This For

• Creators who want to turn photos into talking videos

• YouTubers and educators making explainer content

• Developers testing open-source AI video tools

• Marketers creating product or promo videos

• Anyone who wants to make realistic AI avatars

STEP 1: Install ComfyUI

Go to ComfyUI’s GitHub page and download the latest release for your device.
Unzip the downloaded folder.

Next, install an extension called ComfyUI Manager for easier node management.

Open CMD, navigate to the custom_nodes folder, and clone the extension repository.
Once done, run ComfyUI to make sure everything is set up properly.

STEP 2: Add the Humo Models

Download the required Humo models here, in the model preparation section.

After downloading, move all the model files into the appropriate model folders inside ComfyUI.

Restart ComfyUI once done, your setup is now complete

STEP 3: Load the Humo Workflow

Drag and drop the Humo workflow file into ComfyUI.

Inside the node editor, reselect your downloaded Humo model files so they connect correctly.

Now upload your photo and audio.

You can also fine-tune settings like frame rate, resolution, and output length directly within the workflow.

STEP 4: Generate Your Talking Video

Once everything’s set, hit Run.
Humo will instantly combine your image and voice, creating a realistic talking video with natural expressions and accurate lip sync.

Try different photos or voices, and even type custom scene descriptions, Humo will handle lighting, emotion, and motion automatically.

Google unveils new features for Pixel phones, including AI-powered notification summaries, notification prioritization, and AI photo edits in Messages.

Threads adds new features for podcasters, as it tries to become the de facto place for people to talk about shows and engage with creators.

Gamma hit $100M ARR and raised $68M at a $2.1B valuation for its AI-powered presentation and content platform.

Former physician Jenny Shao left her Harvard residency to launch Robyn, an empathetic AI companion designed to provide emotional support, not replace friends or therapists.

🔎 Moondream: Do real-time video analysis

🧠 Kimi K2 Thinking: Moonshot AI’s new open-source advanced reasoning model

🤖 Grok-4 Fast: xAI’s new lightweight model with a massive 2M token context window

📽️ Hedra: Generate 8 videos or images at once

THAT’S IT FOR TODAY

Thanks for making it to the end! I put my heart into every email I send, I hope you are enjoying it. Let me know your thoughts so I can make the next one even better!

See you tomorrow :)

- Dr. Alvaro Cintas