Ghost Core AI Computer: Would I Actually Pay $3,499 to Run My AI at Home?
A 19-year-old just raised $11 million to sell you a computer with no screen.
That’s the Ghost Core AI computer in one line. Ghost calls it “the personal AI computer.” It’s a $3,499 box that sits on a shelf, runs AI models on its own Nvidia graphics card, and works on your stuff around the clock. You talk to it from your phone.
The first batch sold out in hours, according to SiliconANGLE. It ships October 31. I priced out the alternatives and asked the only question that matters. Would I buy one?
Who’s behind Ghost (and yes, he’s 19)
The founder is Zain Javaid. He told TechCrunch he’d wanted to be a quant most of his life, actually became one, and got bored. Then ChatGPT came out in 2022 and he “felt fundamentally excited in a way that I never had been before.”
After a year of building, Ghost came out of stealth on Monday, October 5, with an $11 million seed round led by Andreessen Horowitz. Abstract, Audacious Ventures, SV Angel and Nova also put money in. His co-founders are Nicholas Chua (19), Gautam Sharda (23) and Yifei Chen (24), and they build the units in San Francisco.
I don’t really care how old a founder is. I care whether the box shows up on time and still works in three years.
So what is the Ghost Core AI computer, exactly?
Javaid calls it a “brain in a box.” There’s no monitor and no keyboard. You reach it through Ghost’s app on your phone or desktop, by text or voice.
The idea is that your files, apps, smart home gear and wearables all feed into it. Then AI agents run on it all day, doing tasks and pinging you when something looks off. (New to agents? Here’s what an AI agent actually is.)
Javaid’s example to TechCrunch: Core spots a higher resting heart rate and a late night of work, knows it’s flu season, and warns you that you might be getting sick.
That’s either really helpful or a little creepy. Probably both.
Here’s what’s inside, from Ghost’s product page:
- GPU: Nvidia RTX Pro 4000 Blackwell SFF Edition with 24GB of GDDR7 memory
- CPU: AMD Ryzen 5 7600, 6 cores
- RAM: 64GB DDR5
- Storage: 1TB NVMe SSD
- Preinstalled models: Qwen 3.8-Next, Qwen 3.8-27B, Gemma 4-31B and Muse-Glimmer-30B
- 1-year warranty, 30-day returns, free shipping for batch one
That last model is the fun one if you read my Meta Muse Gadgets post. It’s not Meta’s Muse assistant. Muse Glimmer is the open 30B model Meta released on August 10, built for always-on local agents. (TechCrunch lists only three models, so Glimmer looks like a recent addition.)
You can add other models from Hugging Face. Ghost also says Core offers an endpoint compatible with OpenAI’s Responses API, so apps like OpenClaw, OpenCode and Codex can use its local models with no per-token fees.
What it doesn’t run is OpenAI’s Dots or Meta’s Muse agent. Those live in the cloud. Core is the opposite bet.
The privacy pitch lands in the right week
On September 29, OpenAI launched Dots, always-on agents with their own cloud computer. OpenAI’s help page says you can’t view or delete individual things a Dot remembers. To wipe that memory, you delete the whole Dot.
Then Wikipedia’s operator said agents it believes OpenAI runs made unapproved edits and flooded its services with millions of requests. I covered OpenAI’s rogue agents hitting Wikipedia in a separate post.
Ghost’s answer is to keep it at home. Javaid says the models and your personal memory run on the device, encrypted, and no third party, Ghost included, can meaningfully get at it. “The user owns all of it,” he told TechCrunch.
He also says Core keeps working if Ghost shuts down, because “all of the source code, logic, and model weights live on the device.” That’s the line I’d want in writing before spending $3,499.
But local isn’t magic:
- Core still needs the internet for web search, email and software updates, per Ghost’s own FAQ.
- An agent with your passwords can still do dumb things. Ghost is building a firewall that checks every outgoing request and blocks anything sketchy. TechCrunch says the team is “building” it, so don’t assume it’s done on day one.
- Model updates come over the air, “similar to how Tesla works,” so you’re still trusting Ghost’s software.
Local means nobody else holds your data. It doesn’t mean your agent can’t screw up. Same lesson as my piece on why Apple got nervous about AI agents and Full Disk Access.
The cost math: Core vs a ChatGPT subscription
Okay, money. ChatGPT Plus is $20 a month. Pro comes in $100, $200 and $500 tiers, and Dots launched on Pro and Business Premium.
Here’s how long it takes for $3,499 to match each one:
| If you’d otherwise pay | $3,499 equals |
|---|---|
| $20 a month (Plus) | about 14.5 years |
| $100 a month (Pro 100) | about 35 months |
| $200 a month (Pro 200) | about 17.5 months |
Add electricity. Ghost doesn’t publish power draw, but the GPU is rated at 70 watts and the CPU at 65. My rough guess: averaging 100 to 150 watts around the clock costs about $13 to $20 a month at July’s US average of 18.31 cents per kWh.
The bigger catch? Subscriptions give you frontier models. Core gives you open models in the 27B to 31B range. They’re good now, but they’re not the best cloud model. You’re trading top-end smarts for privacy and no monthly bill.
On Plus, the math never works. At $200 a month with privacy as your top concern, it starts to.
Ghost Core vs Mac mini, DGX Spark and a DIY PC
Here’s the part that surprised me. $3,499 isn’t overpriced. Not in 2026.
The same GPU alone lists for $2,999.99 at Central Computer (out of stock when I checked) and $3,240.98 at Neutron USA. The cheapest new 64GB DDR5 kit I found was about $930, because RAM prices went wild this year. That’s roughly $4,000 before a CPU, motherboard, SSD or case.
Javaid reportedly said on X that Ghost locked in its GPUs when the card sold for about $1,700, and that batch two will likely cost more. I couldn’t load the post myself, so treat that as reported.
| Option | US price | Memory for AI | The catch |
|---|---|---|---|
| Ghost Core | $3,499 | 24GB on the GPU, plus 64GB system RAM | Batch one sold out, brand new startup |
| Mac mini, M5 Pro, 64GB, 1TB | $2,999 | 64GB shared | You set up the AI yourself |
| Mac Studio, M5 Max 40-core GPU, 64GB, 1TB | $3,799 | 64GB shared | Same, with more GPU |
| Nvidia DGX Spark, 64GB | from $4,999 (launches Oct 23) | 64GB shared | Aimed at developers |
| DIY PC with the same GPU | about $4,000 for GPU and RAM alone | 24GB on the GPU | You build and maintain it |
Javaid told TechCrunch that Core was built for agents with models preinstalled, while the Mac mini is a general-purpose computer you configure for AI. Fair. A Mac mini also edits your videos. Core does one job. (Fuzzy on “home vs cloud”? My cloud computing explainer covers it.)
Who should actually buy one
My honest take:
- Buy it if you already pay $100 to $200 a month for AI, want your data at home, and like tinkering with local models or smart home gear.
- Skip it if you’re happy on a $20 plan or want the smartest model available.
- Wait if you’re curious. Batch one is gone anyway. Let the first owners find the bugs after October 31.
My verdict
Would I pay $3,499 for a Ghost Core AI computer today?
No. Not yet.
And it’s not the price, which is fair for the parts. It’s that nobody outside Ghost has used one, the firewall is still being built, and an always-on agent that knows my heart rate and my inbox has to earn trust slowly.
But Ghost is right about the big thing. If agents are going to know everything about us, plenty of people will want that brain in their house, not in somebody else’s data center. If batch one ships on time and holds up, I’ll be looking hard at batch two.
FAQ
What is the Ghost Core AI computer?
A $3,499 screenless computer from San Francisco startup Ghost, built to run personal AI agents locally. It has an Nvidia RTX Pro 4000 Blackwell GPU with 24GB of memory, 64GB of RAM and a 1TB SSD, and you control it from an app.
When does the Ghost AI computer ship?
Ghost says batch one ships October 31, 2026. SiliconANGLE reports it sold out within hours, and Ghost is taking sign-ups for batch two without a date.
Does Ghost Core need a subscription?
No. Ghost says the AI runs on the device with no subscription. You still need an internet connection for web search, email and software updates.
Can I run AI locally on a Mac mini instead?
Yes. A Mac mini with 64GB of memory can run open models in the same size range. The M5 Pro version with 64GB and 1TB lists at $2,999, but you set up the models and agent software yourself.
Sources: TechCrunch, Ghost (Core product page and FAQ), SiliconANGLE, Meta AI (Muse Glimmer), OpenAI Help Center (ChatGPT Pro tiers), OpenAI Help Center (ChatGPT Plus), Tom’s Hardware (DGX Spark 64GB), AppleInsider (Mac mini prices), AppleInsider (Mac Studio prices), Central Computer, Neutron USA, EIA, CellCog (Javaid’s X post, as quoted).
