Frontier AI Models Will Run Local in 6 Months | Zain Javaid
MTS
October 11, 2026
Ghost CEO Zain Javaid discusses coming out of stealth with $11M in funding and launching Core, a $3,500 dedicated personal AI computer built to process private user context locally.
Zain Javaid
There is no reason at this point that local models, given the capability level and given how fast they're improving, that when I think about where we're going to be in three months, the models are drastically better than where they are now. You know, we have QAN4 coming out soon. I think we'll have the current level frontier running on a laptop or running on a core within the next three to six months.
SPEAKER_2
Hello, everyone, and welcome back to MTS. Today, I am joined with Zain Javaid, who is the founder and CEO of Ghost. And today, Ghost came out of stealth with $11 million in funding led by Andresen Horowitz. And they launched Core. This is a $3,500 computer built specifically for personal AI. And so this idea is that you can have AI that runs on your own hardware. It learns the context of your life over time and then can take actions for you without sending all of that personal context to an AI company. So Zain, welcome to MTS.
Zain Javaid
Thanks for having me.
SPEAKER_2
Thank you for being here. I want to talk a little bit about Core, why you guys launched it, and specifically, what is some of the problems that you've been noticing in right now and how consumer AI is being used?
Zain Javaid
Yeah, so I think on many levels, we're starting to see the dawn of personal AI as a category. I think we're starting to see agents that are capable of doing things for us. But a big question to us is like, what is this perfect final product for personal AI or for consumers look like? And I think for us, it looked like the most capable personal AI will number one, be the one with the most access to your life. So is it able to reason over your entire life? Does it have full context? And the second part of that is, is it able to take action in the kind of like places that you're able to take action in? And this is everything from your browser to even, we've seen stuff like smart home control or being able to read in data from your health wearables or kind of the world around you. And I think on that front, as much as these products have kind of matured, we're still nowhere close to where they could be. And as we got deeper into like what types of problems people are encountering, it just felt like we were more constrained by the platforms we're building on. Meaning, I think a lot of the processing patterns for AI, when I think about, I want agents that are constantly 24-7 looking at my life, ambiently processing it and doing things on my behalf, are able to work for a long time.
I just don't think that the laptop or the phone are the right way to do that. And so as we kind of got lower and lower level on these problems, we started increasingly getting more into hardware, and we're super excited to announce our first product today, which is Core. So number one, we've bundled all this great software, we stream in context from every single one of your apps, files, devices. But the other thing that we feel really strongly about is if you're going to give a system this much access to your life, as in like, I'm going to basically give it unilateral access to the most sensitive parts of my life, I really need that to run locally. The user should own that. That should not be something that I surrender to a big lab and read back by the token. And so every single one of our cores comes with the GPU on board, that's why it's 3,500.
We can run really capable models, like QWEN 3.8 models, we're really good at running.
And I think we've been really happy with the reception so far.
SPEAKER_2
Yeah, I want to talk a little bit more about that. We're, of course, as you alluded to, seeing a big rise in personal AI agents. And I want to understand why they need to run their own computer, specifically when it comes to having all this personal information about you. What are some of the trade-offs, though? You say you run it with your own models. You have QWEN 3.8 that you can actually run, but that's great for today.
Who knows what that will actually look like in a few months from now? And so what are some of the trade-offs that you get when you're running your own computer?
Zain Javaid
Yeah, so we actually made this big bet on Local, I want to say, four months ago. And I think back when we did it, it's really good whenever you can position your company around bets that feel really non-obvious, or they almost feel stupid at the time. So when we started doing Local, the models that were available, like the QEM 3.6 models or the Gemma models, were not good. They were probably sub-HIKU level. And so they worked, but I don't think they were things that you could productize and sell to consumers. But if you look at the trajectory, it was very clear that everything was on an exponential, models were going to keep getting better.
Read the other 18 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:
curl -H "x-api-key: pt_demo" \
https://spoken.md/transcripts/1000651996090Markdown 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:
curl -H "x-api-key: YOUR_KEY" \
https://spoken.md/transcripts/1000795243779