We work inside physical limits, not around them.
Turning the heat, air and water around AI compute into verified capacity is a materials, thermodynamics, controls and metrology problem at once. We treat it that way — separating what is commercial, what is validated, what is active R&D, and what is still a concept — so we can be ambitious without making a claim physics won't support.
Water from heat and air is a heat balance.
Recovering water is never free. Pulling it from a facility's own streams, or from the air, costs energy set by the thermodynamics of the working material — the isosteric heat of adsorption for a sorbent, the latent heat of condensation for a dew-point path. The entire Pleione thesis is that the energy a data centre already rejects can pay that cost, at a grade and timing that line up.
So our first output for any site is not a headline number — it is a heat-and-humidity balance that says what is recoverable there, at what energy, with what confidence. Performance varies with temperature, humidity, airflow, water chemistry and configuration. We say so, every time.
Six disciplines, one system.
Sorption materials
Metal-organic frameworks and advanced sorbents — aluminium fumarate, CAU-10-H and related chemistries — show steep water-uptake steps at moderate humidity and release that water at temperatures compatible with data-centre waste heat. Our work is on working capacity across real duty cycles, hydrothermal stability over thousands of adsorption–desorption cycles, and shaping the sorbent into a cartridge that moves heat and mass fast enough to matter.
Sorption thermodynamics
Desorbing water costs energy — the isosteric heat of adsorption sets a floor. The engineering question is whether low-grade heat that a facility already rejects can pay that cost. We characterise the enthalpy of the working step, the desorption temperature window, and the coefficient of performance of the cycle, so a claim rests on a heat balance rather than a hope.
Psychrometrics & condensation
Atmospheric water is bounded by physics: yield is a function of temperature, relative humidity, airflow and the dew point you can reach with the cold you have. We model the humidity-to-yield surface for a site before we ever quote it, and design the condenser and recovery path against that surface — not against a headline.
Water chemistry & reuse
Recovered and reclaimed water has to be fit for its next use. We work on cycles of concentration in cooling loops, membrane and polishing steps, remineralisation, and the quality thresholds that decide whether a stream can be reused inside the fence or must be returned. Reuse is a chemistry problem before it is a plumbing one.
Physics-informed control
Guardian reads the facility as one coupled system and finds the least-resource operating point. The research is in physics-informed and reduced-order models of the twin, model-predictive control under forecast uncertainty, and — critically — a deterministic safety layer that bounds what any model is allowed to do. No learned model actuates a valve directly.
Verification science
A number nobody can check is worth nothing. We measure on the pipe rather than derive from an invoice, hash-chain each reading, and sign the totals so a third party can recompute them — with the internet switched off. The research is in metrology at the point of measurement, tamper-evidence, and provenance that survives audit.
Where each piece actually is.
Mapped to technology-readiness level, not marketing. High is proven; low is honestly early.
Measurement, verification and orchestration — live on simulated and pilot feeds.
Integrated recovery module — first end-to-end demonstration platform, in build.
Sorbent cartridge + heat integration — component validation on the test rig.
Multi-site storage and orchestration — formulated, not yet demonstrated.
We measure. We don't model-and-hope.
Claims graduate on evidence. Sorbents are characterised for uptake and stability before they enter a cartridge; cartridges are cycled on a rig instrumented for temperature, humidity, flow and mass; the integrated Node is validated end-to-end against metered inputs and outputs. Every stage produces a signed record — the same verification we sell is the verification we hold ourselves to.
What we have not solved yet.
A serious research programme names its hard problems. These are ours.
Working capacity vs. durability
The sorbents with the steepest uptake are not always the most stable across thousands of humid cycles. Finding the pairing that survives a decade of duty is unsolved.
Heat integration in practice
Waste heat is real but messy — variable grade, timing and availability. Matching a desorption cycle to a facility's actual thermal profile is a systems problem we are still characterising.
Yield honesty at the edge
Low-humidity, high-heat sites are exactly where water is scarce and hardest to recover. We are explicit about where atmospheric recovery does and does not pay.
Verifiable at scale
Signing one litre is easy; keeping a tamper-evident, offline-checkable record across a network of Nodes without a trusted central party is an active design question.
Dated, and revised as we learn.
- Guardian orchestration
- Node architecture
- Cartridge test rig
- Digital twin v1
- Water Passport
- Node 01 on site
- Customer pilots
- Thermal Water Engine trials
- Cycling & durability data
- Commercial Nodes
- Facility integrations
- Manufacturing
- Guardian autonomy L3→L4
- Multi-site orchestration
- Regional resource networks
- Water Battery
- Resource markets