AI venture studio · hardware lab · McLean, VA

We build AI companies in our own lab, and incubate them until they can stand on their own.

Joshua8.AI is a founder-led studio. We build the MVP with the founders and carry the overhead of space, compute and the lab until the company can raise outside money and support itself. Then we stay on as investor, advisor and board member.

We also take a small number of select advisory engagements.

The studio model

Build the MVP. Incubate to the first raise. Stay on.

  1. 01

    Build the MVP

    Hands-on engineering alongside the founders, in our own lab: one focused product that solves a single real problem.

  2. 02

    Incubate to the first raise

    We provide the space, the compute, the lab and the overhead until the company can raise outside money and support itself. That may be a seed round.

  3. 03

    Stay on

    We step back from day-to-day operations and remain investor, advisor and board member while others lead later rounds.

Advisory firmCapital-only investorJoshua8.AI studio
Builds the productNoNoYes, the MVP, with the founders
Carries the overheadNoNo, writes a checkSpace, compute and lab, until the raise
Brings a labNoNo4,500 sq ft, GPUs to machine shop
After the raiseEngagement endsStays investedInvestor, advisor, board member

Built in our lab

The companies

Flagship · in Early Access

TeraContext.AI

AI-powered pre-construction for commercial general contractors. It reads the RFP's specs and drawings, classifies every section to a work breakdown structure, and builds trade scope packages, sub bid analysis and the GC proposal. We built the MVP in our own lab.

  • 5,998spec and drawing items classified
  • 246drawing sheets indexed
  • 122trade scope packages generated

Counts from one public municipal bid set, as shown on teracontext.ai.

TeraContext.AI Review and Classify screen: a WBS tree of 5,998 classified items beside a list of classifications with confidence scores
TeraContext.AI, Review & Classify

The hidden advantage

A lab that takes a venture from concept to shipped physical product

Compute 224 GB VRAM on 8 NVIDIA Blackwell GPUs
  • 2 NVIDIA Pascal GPUs (32 GB)
  • ~144 x86 cores, 800 GB RAM
  • Training, fine-tuning, quantization, inference
Infrastructure 25 Gbps fiber LAN
  • 5 Gbps WAN, redundant
  • Redundant power: UPS + generator
  • Classroom for up to 24 people
Machine shop & fabrication 5-ton lift, with a 2-ton hoist
  • Lathe, milling machine, drill press
  • MIG/TIG welding, hydraulic press
  • 3D printing and PCB CNC
Electronics & RF 6 GHz top of the RF transceiver range
  • Xilinx Zynq UltraScale+ and Zynq-7000 FPGAs
  • Software-defined radios from 1 kHz
  • Spectrum and LTE network analyzers

Inside the lab: the full equipment list →

Build with us

Have an AI venture idea, or need a technical co-founder?

We take outside ideas and co-found companies. You bring the problem and the commitment to build it. We bring three things most founders have to go and find separately.

  • EngineeringWe build the MVP with you, hands-on.
  • OverheadSpace, compute access and running costs, carried until the company can raise and support itself.
  • The labGPUs, machine shop and RF bench, from prototype to physical product.

Tell us the problem, how you would like to work, and your timeline. We typically respond within 24 hours.

Jim Smith, Managing Member of Joshua8.AI

Founder-led

Jim Smith, Ph.D., P.E.

Two decades investing in technology companies, with more than 20 investments led across semiconductors, power, RF and software-defined radio, SATCOM and security. Before that: engineering leadership at L-3 Communications, Honeywell and AlliedSignal, co-founder of Transparent Networks, and MEMS development at Sandia National Laboratories.

About the founder → Why "Joshua 8": strategy and a hidden advantage over a frontal assault →