Portfolio

Resolving the bottlenecks that keep compute from reaching demand.

Each venture targets one of three constraints (power, place, or permission) and is de-risked through Constellate's investment strategy.

Power
Energized Capacity Is the Binding Limit

U.S. data-center demand is projected from 23 GW to 134 GW by 2030. Stranded and curtailed power is the fastest unlock.

Fornax

Stranded & Curtailable Power Compute
Fund I · In Formation
Modular compute deployed at curtailed renewables and retired-coal interconnects. Contracted power offtake makes each site project-financeable from day one. The grid can't build fast enough. Stranded power is the fastest path to energized capacity.
Buyer
Hyperscalers, inference providers
Moat
Contracted power offtake agreements
Comparable
Crusoe $1.4B Series E, $10B+ val
Track
Infra Track

Pyxis

Commercial Building Compute
Fund I · In Formation
Modular compute nodes installed in commercial real estate, monetizing spare behind-the-meter power. Revenue shared with property owners. No new grid connection required. Capacity is already energized and permitted.
Buyer
Enterprise AI, edge inference
Moat
Physical-asset access + energy coupling
Comparable
SPAN $420M raised, $600M+ val
Track
Infra Track
Place
Compute Is Being Pushed to New Ground

Land, water, and grid access push compute behind-the-meter, offshore, and into locations traditional datacenters can't reach.

Aquarius

Underwater Compute
Fund I · In Formation
Productize for Western waters what's proven commercially abroad. Sealed pressure vessels with seawater cooling, deployed in a host nation's Exclusive Economic Zone under sovereign concession. No land, no freshwater footprint. Microsoft's Project Natick proved 8x reliability versus matched land deployment.
Buyer
Hyperscalers (wholesale colo)
Moat
Host-government concession
Comparable
Panthalassa $210M raised, $1B val
Track
Frontier

Phoenix

Orbital Compute
Fund I · In Formation
Small incubation position. GPUs running in orbit with near-continuous solar and zero land use. Economics hinge on launch costs; parity expected late 2020s. Scale only when the unit economics arrive.
Buyer
Sovereign and latency-insensitive workloads
Moat
Launch partnerships + orbital slot access
Comparable
Starcloud $200M raised, $1.1B val
Track
Frontier
Permission
Regulation Keeps Some Compute Local

Sovereign data laws and on-site inference requirements mean finance, healthcare, and defense need in-jurisdiction capacity. EU AI Act enforcement begins August 2026.

Corona

Sovereign Regulated Compute
Fund I · In Formation
In-jurisdiction, compliance-aware compute capacity for finance, healthcare, and defense. Data stays on-soil. Premium recurring revenue from buyers who can't use hyperscaler regions due to regulatory or classification constraints.
Buyer
Regulated enterprises, sovereign AI programs
Moat
Regulatory compliance + jurisdiction lock-in
Comparable
Nscale $3.7B raised, $14.6B val
Track
Infra Track

Centauri

Embodied / Edge Inference
Fund I · In Formation
Self-contained inference modules for homes, buildings, and industrial plants. Zero round-trip latency. Physics-bound demand. These workloads can't go to the cloud. Grows even in capex slowdowns because the compute must be on-site.
Buyer
Smart buildings, industrial OEMs, consumer hardware
Moat
Hardware TEE + on-device model optimization
Comparable
SiMa.ai $355M raised, $960M val
Track
Frontier