Ghost Raises $11 Million for a $3,499 Screenless AI Computer

Ghost, a San Francisco hardware startup run by 19-year-old Chief Executive Zain Javaid, raised an $11 million seed round led by Andreessen Horowitz to sell Core, a $3,499 computer with no screen built to run AI agents locally, TechCrunch reported on Oct. 5. Abstract, Audacious Ventures, SV Angel and Nova also invested. The valuation was not disclosed, which removes the usual way to judge the price.

The round is announced and, in the coverage I read, described as raised. The product is the testable part. Units are due in the last week of October, per TechCrunch and Martin Cid Magazine.

What $11 million buys against $3,499

Divide $11 million by the $3,499 retail price and the seed equals about 3,100 Cores sold at full price. That is not revenue, because no preorder or deposit count appears in any source. It does frame the scale: the company must sell a few thousand units to match its own funding, and Ghost assembles the machines in house in San Francisco, according to Martin Cid Magazine.

The founding team, per TechCrunch, is Javaid, Nicholas Chua (19), Yifei Chen (24) and Gautam Sharda (23). Martin Cid Magazine describes Javaid as a former quantitative trader.

The hardware, line by line

Part	Spec	Reported by
GPU	Nvidia RTX Pro 4000 Blackwell, 24GB	Martin Cid Magazine
CPU	AMD Ryzen 5 7600, six cores	Martin Cid Magazine
Memory and storage	64GB RAM, 1TB SSD	Martin Cid Magazine
Price	$3,499, US dollars only	Martin Cid Magazine
Preinstalled models	Qwen-3.8-Next, Qwen-3.8-27B, Gemma-4-31B	TechCrunch

There is no display. Users drive the machine through a phone app by text or voice, and Martin Cid Magazine lists a built-in firewall for outgoing requests, no subscription and a 30-day return window. Javaid told TechCrunch: "The user owns all of it...everything is stored, run, and processed on the device."

The 24GB of GPU memory is the constraint to check. A 27-billion-parameter model stored at 16-bit precision needs about 54GB for weights alone, so the 27B Qwen and the 31B Gemma only fit on this card in compressed, quantised form, at 4-bit roughly 13.5GB and 15.5GB. That is my arithmetic, ignoring the memory the conversation cache needs, and no source says which quantisation Ghost ships. The same download-the-weights approach underlies open releases we have covered, such as Aleph Alpha's Kolibri-1 under Apache weights and Reflection's 501-billion-parameter Beam.

Where the sources disagree

TechCrunch lists preorders opening Oct. 7. Martin Cid Magazine says Oct. 5, the day of the announcement. I could not reconcile them from the pages I read. The shipping window matches in both.

No source states who built the preinstalled software, how over-the-air model updates are signed, or what the firewall blocks. For a product pitched on privacy, those are the three answers a buyer needs before a card goes in.

The next observable event is the first shipments in the last week of October. The 30-day return window that follows is the earliest public test of whether 24GB of video memory is enough for the models Ghost bundles.