Why AI Data Centers Need a New Class of Switch | Jeff Holden (61 chars) artwork

Why AI Data Centers Need a New Class of Switch | Jeff Holden (61 chars)

MTS

October 10, 2026

Jeff Holden discusses Atomic Machines emerging from six years in stealth with a $250M raise and the introduction of the Matter Compiler—a physical manufacturing system that builds micromachines directly from code.

0:00

Jeff Holden

In the world of AI data centers, and this is true in electrification, EV charging and a bunch of other areas, but AI data centers are the extreme.

There's this, as the AI chips advance, they become more power hungry. So you've got racks that are pulling in hundreds of kilowatts growing to a megawatt. You've heard about the megawatt rack. To be able to do that, what we do is we put the PrimeSwitch next to the semiconductor 99.99% of the time when everything's operating normally, all the current is going through PrimeSwitch and you're bleeding off way less heat, way less of the energy as heat.

0:31

Sophia

Welcome back to MTS. Today, we are joined by Jeff Holden, who is the founder and CEO of Atomic Machines. Jeff's an interesting person because he was Amazon's 10th engineer. He helped build Amazon Prime. He later became Uber's first Chief Product Officer. But one of the big things we're announcing today is just yesterday, Atomic Machines emerged from six years in stealth with a $250 million raise with a new manufacturing system called the Matter Compiler that builds working Micro Machines directly from code. Pretty exciting. Jeff, welcome to MTS.

Thank you.

1:07

Jeff Holden

I'm glad to be here.

1:08

Sophia

Glad to have you. What an exciting day. I bet yesterday when you officially launched, you got six years in the making. Congratulations. Tell me a little bit about what you guys announced yesterday.

1:18

Jeff Holden

We announced two things. Well, three things if you include coming out of Stealth. One was that we developed this technology called the Matter Compiler which you described very well. The second one was we've made the first device with the Matter Compiler called PrimeSwitch.

Yes, we talked all about those things.

1:36

Sophia

Amazing. Let's go first into Matter Compiler. If you were to break it down in simple terms, because I used a little bit of jargon when I was explaining it, this new manufacturing system, tell me a little bit about what it actually works and then what wasn't possible before that now is possible.

1:52

Jeff Holden

Yeah, great question. The way to think about this is a software compiler takes high-level software code and turns it into something a machine can run. Imagine if you just kept compiling all the way to a physical object.

What that means is real-time manufacturing. You know, you start from a, you have a digital, you think of the input to the Matter Compiler as being a language, and that language has a grammar, and then you can express the sentences in that grammar. Whatever sentence you give to the Matter Compiler, it will make a machine for you. And those machines can be very different. It could be a micro robot that's used for neurosurgery. It could be PrimeSwitch, which is a very low-resistance, ultra-fast switching device. It could be a micromanipulator for micromanufacturing. Everything has micro in front of it.

But the key thing here is that it's what's called, if you want to get a little geeky, a combinatorial constructor. So that means that it has the ability to make many different kinds of machines. And that technology, there's never been a manufacturing system in history that does that, where you can give it different inputs and see a different machine come out the other side. So I liken it to an auto factory. Imagine you visited a new kind of auto factory and you saw a car rollout and then maybe a truck rollout, and then a laptop rollout. And you're like, didn't expect that. And the refrigerator comes out.

So that's combinatorial constructing. And that's what the Matter Compiler is the first of its kind.

3:17

Sophia

That's interesting. I think in theory, I understand it.

In theory. But when you're actually describing it and you're telling about what this actually does, it sounds very sci-fi. It's almost this AI native manufacturing system. But I get how it works from, okay, you have a code, you're able to code it, but what you're kind of claiming that anything that you can possibly decide with code actually can determine the machine that gets built. And so how does that work? How does it actually happen? If you're a factory, do you have to start from scratch? You have to build up a whole new factory from scratch? Or can you just work within my existing tooling that I have today?

3:54

Jeff Holden

That's a great question.

So it does seem sci-fi when you first think about it. So the attributes you need to have to make it combinatorial in this way. And by the way, we're doing only micro machines. We're not making refrigerators and cars. Makes sense.

Read the other 13 minutes of this episode here.

$15 gets you this episode and 99 more. Credits never expire. After checkout, one link brings you back here to read it.

Not what you expected? Full refund within 14 days.

Thousands of transcripts fetched by people building searchable podcast archives

Fetch transcripts with code

The same credits work from a script or an AI agent. The demo key returns a sample episode in full, no card needed:

request
curl -H "x-api-key: pt_demo" \
  https://spoken.md/transcripts/1000651996090

Markdown with the speakers named, for your notes, your knowledge base, or anything that makes HTTP calls.

From $0.10 per transcript. No subscription. Credits never expire. Prices exclude VAT, added at checkout for EU customers. Not what you expected? Email us within 14 days with 20 or fewer credits used and we refund the pack in full.

Using your own key:

request
curl -H "x-api-key: YOUR_KEY" \
  https://spoken.md/transcripts/1000794599922