Kaspix
ActiveAI-nativeKaspix helps OEMs ship cheaper, upgradeable edge devices faster—using swappable manufacturable analog-AI modules that cut BOM and test time without leading-edge digital chips.
What it does
KASPIX is the Physical AI Infrastructure compiler — a platform that transforms any trained neural network into a licensable analog circuit design, deployable on PCB, ceramic, or flexible substrates by any manufacturer, with no GPU and no advanced foundry. Unlike digital AI that runs _on_ chips, Kaspix compiles intelligence _into_ matter — the circuit doesn't execute the model, it physically embodies it. Digital chips cannot operate in these environments — not because they are too expensive, but because they are physically incompatible: they require stable power, generate heat, need firmware, and depend on connectivity. These are not engineering problems waiting for a better chip. They are the physics of silicon. This unlocks the 97% of embedded environments that every chip physically cannot reach: sub-1mW sensors, 300°C+ industrial sites, air-gapped facilities, and zero-connectivity field deployments. Our ARM-style licensing model — per-design fees plus royalties — scales across every substrate and vertical without manufacturing capex.
Next to it in AI infra and compute · Analog AI circuit compilation
- Recursive SuperintelligenceSouth Park Commons
AI that improves AI
- Deep CogitoSouth Park Commons
Post-training research lab
- Lincoin Technologies Inc.Plug and Play PnP 2025
Merging energy intelligence into high performance computing
- Cactus ComputeY Combinator S25
Automation Foundation Models For Tiny Devices
- Iron GridY Combinator S25
AI-Native Data Center Engineering
- LilacY Combinator S25
We sell GPUs.
- Jigsawa16z speedrun SR005
Post-training data for RSI
- JuxtaY Combinator S25
Building a GPS alternative 100x more powerful with no hardware needed.