Building next-gen computing
A completely new class of computing systems will be needed in the coming decade. We are building its foundation technologies.
AI is fundamentally changing how we use computers
Keyboards → voice and AI assistants
People speak to the machine; AI works continuously, at machine speed, on every task at once.
Programs → models
Inference everywhere, all the time, is the default workload.
Kilobytes of text → floods of data
Enormous volumes flow between memory, compute and devices.
Computers have to evolve to keep up with this changing landscape.
Our thesis
The heart of the computer needs radical innovation.
The processor must become AI-native high-performance compute: massively parallel, fed by very high bandwidth, secure by design. Not a general-purpose CPU with an accelerator bolted on.
Security belongs in the architecture, not the network
Your data, your models, your processing
Computation on personal data happens here and never leaves. No network path reaches it.
Physical boundary
AI services, the cloud, other machines
Treated as untrusted by default. Post-quantum cryptography guards every crossing of the boundary.
In an AI-native, quantum-native world the network is not really secure. So personal data and its processing stay physically separate.
We are building the full stack
Applications & models
AI assistants, model inference and ordinary general-purpose workloads, running on the same machine.
OS, SDK & compiler
Our own toolchain from application to hardware, on a Linux-based operating system.
AI-native processor
Massively parallel compute with very high bandwidth data movement. Built around an open RISC-V host, so nothing in the stack depends on a licensed ISA.
Secure system architecture
Post-quantum secure, with personal data and processing physically separate from the network-facing side.
One design technology, three product domains
Edge AI
Devices that run inference on site, continuously: defence, infrastructure, industrial, automotive.
Personal computing
A computer that runs its own models locally and keeps personal data physically separate from the network.
Server compute
The same massively parallel, high-bandwidth design scaled up for data-centre AI.
Watch it run
Edge intelligence on live video
YOLO on real video, end-to-end on our platform: continuous, on-site, no cloud round trip.
Also on YouTubeSped up. The first ~12 s is prefill, the model reading the prompt.
State-of-the-art small language models
Qwen2.5-Coder generating working C code, live on our platform.
Small language models of 2–4B parameters run today.
Any model of this complexity can be ported to the platform. Model complexity is not the constraint.
Also on YouTubeWhy build in India? Why now?
~100%
of India's AI compute runs on imported silicon
~6%
India's share of global semiconductor consumption, heading to ~9% by 2031
<5
Indian startups working on AI-first chips
Sources: Kotak Mahindra MF semiconductor report (Aug 2026); IESA and industry reporting on India's AI-chip imports. No Indian-designed AI accelerator has shipped in volume yet.
Let's build the computer of the next decade.
Partners, early adopters, investors and engineers: we would like to hear from you.