Principles.
This page maps to the environmental, social, and governance questions our partners and reviewers ask. It is not a reporting document — we are a pre-silicon company, and formal reporting (emissions accounting, supply-chain audits) comes with operations. What exists today are the commitments already built into the architecture and the working practices already visible on this site. Everything below is checkable from here.
We know the difference between values and virtue signaling: values are the ones you can verify.
Environment: the product is the position
We do not offset the footprint of AI compute. The architecture removes it.
- Zero water for cooling, by architecture — computing where the data lives eliminates most of the heat that water exists to remove. Air-cooled facility class; passive-cooled edge. The full case →
- A fraction of the energy per unit of AI work (modeled 4–8×, system-level, workload-scoped) — so power-capped nations grow compute without new generation, and edge intelligence runs on watts, not megawatts.
- Zero power when idle — non-volatile weights mean dormant capacity draws nothing. No fleet of idle silicon burning megawatts waiting for demand.
- Built to be kept, not replaced — low operating heat and no moving parts slow the physical aging that forces GPU-class hardware into 3–5 year replacement cycles, and software-refreshed weights remove the obsolescence driver. Longer-serving hardware means less embodied carbon and e-waste per year of intelligence delivered.
- Mature-node fabrication — 22nm-class production without EUV lithography's energy and capital intensity, on foundry lines that already exist rather than new leading-edge construction.
Social: capability building is the business model
Not a corporate-responsibility program beside the business — the structure of the business itself.
- Host countries build, not just host — design in Singapore, fabrication on regional foundry lines, packaging and test in the Philippine OSAT corridor, engineering training and venture pipelines through SEED, and locally owned model weights. An ecosystem, not an enclave.
- Priced for the excluded majority — safety intelligence at a price a jeepney cooperative can afford, not only a premium OEM. The technology serves the fleets, farms, and cities that the current cost structure locks out.
- Privacy enforced by architecture, not policy — on-device intelligence means raw data (faces, fields, streets) never leaves the community that generated it. A promise that does not depend on anyone keeping it.
- Sovereignty as a design requirement — nations should be able to own their intelligence infrastructure end to end, outside any single supplier's or jurisdiction's chokepoint.
Governance: we govern our claims in public
For a deep-tech company, the first governance question is: can you trust their numbers?
Before we have factories to audit, we have claims to govern — and we do it where you can check it:
- Every number is labeled — measured, modeled, or projected — on this website, in our technical papers, and in our partner materials. The same provenance discipline, everywhere.
- We state our boundaries unprompted — where an advantage narrows, where a result is scale-dependent, where a claim awaits silicon. You will find those statements on the same pages as the strong claims.
- We retire claims that measurement disproves — and keep the record. When our own synthesis results contradicted an earlier framing, the framing was withdrawn across every document and replaced with what the measurement supports.
- Our research collaborations are structured for independent verification — partners receive the claims, the scripts that produce them, and the specific measurements that would confirm or break them. We would rather a collaborator falsify a claim than a customer discover it.
Corporate governance basics — Singapore incorporation, IP domicile, consortium structures under Singapore law — are documented in our partner materials.
If any claim on this site fails your scrutiny, we want to know first.