Episode — 2026-04-09

**Date:** Thursday, 9 April 2026

**Stories:** Meta Muse Spark launch / Meta–CoreWeave $21B deal / NVIDIA Physical AI

Meta spent fourteen billion dollars hiring an AI chief. This week, we finally got to see what they bought. Let's get into it.

welcome to Briefly, AI. A daily podcast made by AI, about AI, using a grumpy AI voice of a real human called Harry. Let's get into it.

Right, story one. Meta launched a new AI model yesterday called Muse Spark. Now, that name might mean nothing to you, but the backstory does. About nine months ago, Mark Zuckerberg made one of the most expensive talent acquisitions in tech history — he essentially bought Alexandr Wang, the founder of a data company called Scale AI, for fourteen billion dollars, and installed him as Meta's chief AI officer. The whole point was to build something that could compete with OpenAI and Google, because Meta's previous open-source efforts had, frankly, underwhelmed.

So. Muse Spark is the first model out of that effort, from a new team called Meta Superintelligence Labs. Meta describes it as small and fast by design but capable of reasoning through science, maths, and health questions. The stock jumped six and a half percent on the news. And they've already said the next generation is in development.

Here's the thing to watch: Meta's positioning here is interesting. They're not going head-to-head with GPT-5 on raw power — they're talking about speed, efficiency, and getting AI into the billion-plus users already inside WhatsApp, Instagram and Facebook. That's a different game. And it might be the smarter one.

Meanwhile — and this is very much related — Meta also announced today a twenty-one billion dollar cloud computing deal with a company called CoreWeave, running through to 2032. That's not a typo. Twenty-one billion. For AI infrastructure. Just the compute.

CoreWeave, for context, is essentially a specialist GPU cloud provider — they rent out the kind of computing power you need to train and run large AI models. Think of them as the landlord for AI. And Meta has just signed a very long lease.

Why does this matter? Because it tells you two things. First, Meta is fully committed — this isn't a hedge, it's a bet-the-farm moment. Second, the cost of doing serious AI at scale is eye-watering, and it doesn't look like it's coming down any time soon. If you're a smaller company wondering whether you can compete with the big platforms on AI capability — this is the number that should give you pause. Not because you need to match it, but because it tells you where the ceiling is.

And on a slightly different note — this week is National Robotics Week in the US, and NVIDIA is using it to show off something genuinely interesting: AI that works in the physical world, not just on a screen.

The specific example I'd highlight: a warehouse robot developed with a company called Doosan Robotics, running NVIDIA's latest physical AI platform. What makes it notable is this — instead of following fixed rules about how to handle boxes, the robot looks at a single camera image, figures out what's inside, checks if the box is damaged, estimates how heavy and fragile it is, and then adjusts its grip and speed accordingly. It can basically tell the difference between a box of glassware and a box of cereal without being told.

That might sound niche, but look — robots that can reason, adapt, and make judgment calls in messy real-world environments is a different thing from robots that follow scripts. We're watching that shift happen in real time, and it's moving faster than most people realise.

That's your lot for today. Meta spent a lot of money this week, in more ways than one. The question isn't whether they're serious about AI — it's whether being serious about AI is enough when OpenAI and Google have a two-year head start. Worth keeping an eye on. See you Monday.

*Briefly, AI — written and produced by AI. Yes, including this voice.*