Drug Discovery
Generative chemistry, docking simulations and ADMET prediction pipelines running on dense GPU/TPU clusters — compressing months of wet-lab cycles into days.
/ Deep Tech R&D
AI Container, powered by biology.
A 40-foot state-of-the-art AI data center container engineered for drug discovery, climate modeling, protein-folding and materials research — paired with active bio-electricity R&D inspired by the Levin Lab at Tufts to push compute efficiency past today's silicon ceiling.
One container, four frontier sciences.
Climate Prediction
High-resolution earth-system and regional weather models, downscaling ensembles and renewable-yield forecasting for grid and agriculture operators.
Protein-Folding Research
AlphaFold-class structure prediction and misfolding analysis to study neurodegenerative and prion-related disease pathways at scale.
New Materials Discovery
Graph neural networks and DFT-surrogate models exploring battery chemistries, photovoltaics and structural composites for next-gen energy hardware.
/ Bio-electricity R&D
Inspired by the Levin Lab.
Dr. Michael Levin's group at Tufts (Allen Discovery Center, Center for Regenerative and Developmental Biology, and the Institute for Computationally Designed Organisms) has shown that cells use endogenous bioelectric signals — patterned ion flows across membranes — as a primitive computing and memory substrate that guides form and behavior.
We are translating that insight into hardware. Our R&D explores ion-gradient and membrane-potential inspired modules as ultra-low-power co-processors and sensors alongside silicon — the goal is compute that runs cooler, draws less grid power, and degrades gracefully like living tissue.
Caution — early-stage R&D. Bio-electric co-processing is theoretical and has not yet been proven at production scale. No commercially available hardware currently exists.
/ Compute & cooling architecture
Photons move the data. DC moves the power.
Photonic + wafer-scale silicon
Photonic chips move data between accelerators at a fraction of the energy per bit of copper, while Cerebras wafer-scale parts handle dense AI training and inference on-chip — minimising off-die traffic.
Hybrid cooling stack
Direct-to-chip liquid loops and rear-door heat exchangers do the heavy lifting, augmented by Microsoft-style channelled cooling and trim A/C, with air-cooled fallback for graceful degradation.
High-voltage DC power path
On-site storage delivers HVDC straight to the racks, cutting AC↔DC conversion stages, shedding waste heat, and shrinking the cooling load before a single watt hits a chip.
/ Architecture
Inside the container.
Three subsystems share one 40 ft shell: a photonic interconnect fabric for data, a hybrid liquid loop for heat, and an HVDC bus for power — each color-coded below.
Photonic data path
Cerebras nodes ↔ photonic switch ↔ 400 GbE / sat
Cooling loop
DTC cold plates → manifold → RDHx → dry cooler
HVDC power
DC storage → HVDC bus → racks (no AC↔DC at rack)
From ion flows to power savings.
Bioelectric co-processors
Membrane-potential inspired analog elements that compute pattern-matching and optimization tasks at a fraction of the energy of equivalent GPU ops.
Hybrid silicon + wet substrates
Engineered cellular sheets interfaced with CMOS, drawing on Levin Lab work on bioelectric pattern memory and xenobot-style self-assembly.
Adaptive power orchestration
AI agents shift workloads between silicon and bio-electric modules in real time to minimise total kWh per token, per simulation, per molecule.
Independent R&D program. Not affiliated with or endorsed by Tufts University or the Levin Lab — their published work is a scientific inspiration for our engineering roadmap.
/ Container specsForm factor 40 ft non-ISO composite-shell container, trailer-mounted and mobile Compute 700 kWh continuous IT load — photonic interconnect fabric + Cerebras wafer-scale accelerators for high-bandwidth AI training and inference Cooling Hybrid thermal stack: direct-to-chip liquid + rear-door heat exchangers + Microsoft-style channelled cooling, A/C augmented, air-cooled fallback. PUE < 1.15 Power On-site DC energy storage feeding high-voltage DC bus direct to racks — eliminates AC↔DC conversion losses and reduces heat. Hydrogen fuel-cell prime power on the roadmap from early 2027 Network Dual 400 GbE uplinks, optional satellite backhaul Deployment Crane-and-go on its trailer: site-ready in under 72 hours
Built for the field, not the cloud.
/ FAQ
Common questions.
Research partners & pilot customers wanted.
We are looking for biotech, climate-science and materials labs ready to co-develop the next generation of energy-aware AI infrastructure.
Talk to our R&D team