Breaking
AI & MLDeveloping Story

Ghost's $3,499 AI box sells out first run

A teen-founded startup says its on-device AI computer sold out its first batch, betting privacy beats cloud convenience.

··2 hours ago·6 min read
A computer monitor, keyboard, and game controllers on a white desk in a workspace
Photo by Sebastian Bednarek on Unsplash

A startup called Ghost has built a desktop PC designed to run AI agents entirely on the machine itself, and according to TechCrunch, the first batch sold out on the day pre-orders opened. The device, called Core, is pitched as a way to keep AI workloads local rather than routing them through cloud services — a distinction the company frames as the product's central appeal.

What Core actually is

Ghost describes Core as a personal AI computer that runs models on-device. On the product's site, the company explains: "Core is a personal AI computer that runs models entirely on device. It connects to your apps, files, and devices, builds context on your life over time, and works continuously to help you, not just when you ask."

The machine is a black box built around an Nvidia RTX Pro 4000 (Blackwell) graphics card with 24GB of video RAM, paired with 64GB of DDR5 system RAM. Ghost rates its AI performance at 770 TOPS. Three models come pre-installed: Qwen-3.8-Next, Qwen-3.8-27B, and Gemma-4-31B.

Founder and CEO Zain Javaid has described the device as a "brain in a box" — a compact system meant to sit on a desk and handle AI tasks locally rather than in a data center.

The setup problem it targets

The pitch is aimed at people who want to run local AI but don't want to configure it themselves. Buying a Mac mini or Mac Studio with enough memory to run larger models, then setting up the software stack, is the alternative Ghost is positioning against.

Javaid argues that running models on existing hardware is technically possible but painful. "You can run [AI] models on existing hardware, but it's a terrible experience on many levels," he said.

Ghost says Core ships with software built specifically for running the pre-installed models, removing the configuration work. The system updates its models automatically as new versions arrive, unless the user opts out — an approach Javaid compared to how Tesla handles over-the-air updates.

Privacy as the sales argument

Ghost is targeting users who see AI's potential for work or daily life but are uneasy about sending personal data to cloud services where they don't know how queries or uploads are handled. Core runs AI agents continuously while keeping data on the device, and Ghost says that data doesn't leave the machine.

According to the company, personal data on the PC is encrypted, and even Ghost itself cannot access it. Core is also designed to keep running if Ghost were to shut down, and there is no subscription fee attached to the device.

Javaid framed that independence as deliberate. "All of the source code, logic, and model weights live on the device, not the services," he said. "We build this device to not be dependent on having an outside entity maintaining the service."

The credential firewall

Because Core connects to apps and services, Ghost has built a firewall that limits how any agent can use credentials such as passwords. Users can also set how much autonomy Core's agents are allowed, restricting the AI if they prefer.

Javaid said the firewall monitors all outgoing requests. "If it sees the agent performing a sanctioned action or exfiltrating data it shouldn't, it rejects the request and flags it for user review," he told TechRadar.

That design reflects the fact that an agent with access to accounts and files is a different risk category from a chatbot that only answers questions. The firewall is Ghost's attempt to put a boundary between the agent and the credentials it can reach.

Price and availability

Core is priced at $3,499 in the US, which works out to around £2,600 or AU$5,000. Pre-orders for the first batch sold out on the first day they were offered — the day before the TechRadar report was published. Those devices ship at the end of the month, and Ghost is now taking registrations of interest for a second batch coming off the production line.

  • $3,499 — US price of Core
  • 24GB — video RAM on the Nvidia RTX Pro 4000 (Blackwell) GPU
  • 64GB — DDR5 system RAM
  • 770 TOPS — rated AI performance
  • $3,199 — US price to configure a top-of-the-line Mac mini for AI, at the time of writing

The Mac mini comparison

The price invites an obvious comparison. A top-of-the-line Mac mini configured for AI runs $3,199 in the US at the time of writing, and it doesn't include software built specifically for running local AI, nor the privacy and security features Ghost emphasizes.

That said, Apple's hardware is familiar to many buyers, while Ghost is offering a turnkey local AI box with a firewall and encryption. The trade-off is between a known ecosystem and a more specialized, purpose-built machine from a new company.

The comparison also highlights who Ghost is targeting: the AI enthusiast who doesn't want to buy a Mac mini or Mac Studio with maxed-out memory and then spend time setting everything up.

Why the next batch may cost more

Ghost bought its stock of Nvidia GPUs when they were much cheaper, which the company says makes the current PC a bargain relative to the GPU's present price. That advantage won't extend to later production runs — the next batch will be pricier.

The broader memory market is a factor. The RAM crisis remains in full force, and the latest signs are described as far from positive. For a device that leans on 64GB of DDR5 system RAM and a 24GB GPU, component costs are not incidental.

A startup promising a fixed price of $3,499 on a machine built from components whose costs are moving against it is making a bet on volume and supply timing. Whether that price holds beyond the first batch is an open question, and Ghost has already signaled that it won't.

What buyers are actually betting on

Core is designed to hook into existing devices and services, smart-home gadgets, fitness watches, and personal health data, learning over time about how a user lives and offering advice on that basis. The more it connects to, the more sensitive the data it holds — which is why the encryption and firewall claims carry weight.

Exactly how useful the system will be is less clear, and it may depend on how much connected tech a user already owns. There could be scope beyond life monitoring for Core's AI agents, with the flexibility to get them doing whatever a user wants.

Because this is currently a single-source report, details about real-world performance, the robustness of the firewall, and the actual privacy guarantees still need independent verification. Ghost's claims about encryption and no outside dependency are company statements, not yet audited by third parties.

The question of worth

Is $3,499 a lot to pay for a dedicated AI PC? It's undeniably a hefty sum, but there are heavyweight components under the hood, and the current price of the Nvidia GPU makes the machine a bargain as it stands — a window that closes with the next batch.

For buyers weighing a local AI box against a configured Mac mini, the decision comes down to whether turnkey setup, on-device processing, and the credential firewall are worth the premium and the uncertainty of a young company's roadmap.

Core's real test isn't the sold-out first batch — it's whether the second batch, at a higher price, finds the same demand once independent testing can confirm what the device actually delivers.

#ghost#local ai#on-device ai#nvidia#ai hardware

Sources

Iliyas

Founder & Editor, Xploitwire

This article was written and reviewed against the sources listed above before publication, under editorial policies set by Iliyas. Read our Editorial Policy →

← Back to all stories