How much validated intelligence can you actually deliver per square meter of solar panel? Plug in your workload, your conditions, your hardware. The answer surprises most people.
Modern computing systems doing useful work — answering questions, catching anomalies, controlling industrial processes — need electricity. Most current systems need a lot of it. The substrate Validiti has built does the same kinds of work on far less.
This calculator turns that difference into a number anyone can understand: how much solar panel would you need to power this work, and how much useful output do you get per square meter of that panel. Pick what kind of work you care about. Pick where in the world the panel sits. See the ratio.
A small farm, a solo law practice, a rural clinic, a one-truck operator — each one looks at hyperscale cloud and assumes they would need a parking lot of solar panels to match what Amazon Web Services delivers. The honest answer is the opposite. Substrate-shape compute on commodity hardware fits in a footprint most people would call small. Conservative real-world examples below.
Document storage and search, billing records, client communications, basic records-query and citation work. Roughly the workload a 1–10 person practice runs on Microsoft 365 plus a specialty cloud tool.
The difference between a doormat and a dining-table top. Roughly 10× less.
Tracking animal records, soil sensor logs, equipment maintenance, weather data, crop history. A typical 200-acre operation running cloud agricultural software (Ag Leader, FieldView, etc.) plus general business records.
The difference between a coffee table and a parking space. Roughly 10× less.
Patient records for ~5,000 patients with HIPAA-compliant audit. Conventional setup pays cloud premium for compliance tooling on top of the storage and processing. Signed chain of custody is a separate add-on.
The difference between a doormat and a tool-shed roof. Roughly 25× less. The audit chain costs nothing extra in the substrate version.
Federal hours-of-service compliance, delivery records, fuel + maintenance logs, driver communications. A 1-2 truck owner-operator using fleet-management cloud services.
The difference between a magazine cover and a full-sized solar panel. Roughly 12× less.
Run reports, vehicle and equipment inventory, training records, compliance documentation. The kind of records department of a 20-person volunteer organization that currently pays subscriptions to multiple cloud services or runs everything off paper.
The difference between a doormat and a desk top. Roughly 7× less. And you actually own the records.
Daily inventory turnover, POS transaction history, supplier records, employee scheduling. The cloud-stack a single-location independent grocer or hardware store typically pays for.
The difference between a floor mat and a small desk. Roughly 10× less.
The numerator counts signed-provenance outputs from the architecture — pattern detections, citation-anchored responses, signed decisions, signed records citations. Outputs without provenance do not count. This is the metric’s stance on hallucination: not penalized, just excluded.
The denominator is the square meters of photovoltaic panel needed to sustain the architecture’s effective power draw — including cooling overhead and, optionally, amortized training energy — under the selected solar conditions.
For two architectures running the same workload under the same solar conditions, the ratio collapses to relative effective-power consumption. The framework’s value is in making each component — compute draw, overhead, amortized training — explicit and user-parameterizable.
The metric is in scope for: pattern recognition, signed retrieval, symbolic computation, multimodal sensor fusion, process control, anomaly detection, records-only query, cascade detection, and decision-making under signed-provenance constraints. It is honestly out of scope for: drug discovery molecular simulation, weather and climate model integration, large-scale physics simulations, and open-ended creative generation where the value is in the diversity of output rather than the provenance of any single output. The substrate does not displace those workloads, and the metric does not claim to compare there.
This is the first version of the Compute per Solar Watt benchmark, published by Validiti Institute. The substrate-side numbers behind each preset are honest estimates derived from existing Validiti benchmarks and published architectural mappings. Phase 1 measurement work will replace these with peer-reviewable measured values; nothing here is gated behind that update — the methodology is open in full so anyone can verify, contribute, or override.
Incumbent-side numbers are sourced from published vendor datasheets, MLPerf results, hyperscaler sustainability reports, and the peer-reviewed academic literature on AI training energy. The conservative-comparison principle: where multiple incumbent numbers exist, the calculator uses the most favorable-to-incumbent published value.
Override anything. Disagree with a default? Plug in your own. The tool gives you the framework; the inputs are yours.