Shared network
No more private data centers. Connect through one API, lease compute on demand, pay for what you use.
Next-generation compute
Smart AI Factory Network
Compute is following the path electricity took.
Every company builds and runs its own data center, like every factory running its own power plant in the early industrial era.
A Smart AI Factory Network — centrally dispatched, drawn on demand, shared across the network. Like a national grid.
01 / From Generation 1 to next-gen
The same compute, organised differently — and the economics change with it.
Generation 1 Next-generation
Build Build-and-own; capital-heavy, slow Shared network; lease on demand, nothing to build
Operate Single site; one outage halts compute Multi-node redundancy; loads migrate automatically
Utilization Built for peak, run at average Network-wide dispatch; tasks matched to GPUs
Power Bound by local grids and interconnection queues Follows the power; built where low-cost power is
Upgrade Shut down to retrofit PODs refresh by generation; the network stays up
02 / Four capabilities
One call, two routing stages
Task request
01 Stage 1 · Pick the model
Match the task to the best of 250+ models
Model lake 250+ 02 Stage 2 · Place the model
Place it on the best-fit GPUs network-wide
Result
Dual intelligent routing
Every call passes two routing stages: the task is matched to the best of 250+ models in the model lake, then the model is placed on the best-fit GPUs across the network — trading speed, accuracy and cost automatically per SLA.
Follows the power
Nodes are built where low-cost power is: a million-acre solar flagship in Argentina for training and inference, and distributed POD nodes on surplus power in the US and Europe, close to customers.
Always upgrading, never down
POD modular data centers refresh generation by generation. Upgrading one node never interrupts the network. Customers always see the same API and SLA.
03 / Resilience
The network makes compute resilient
- Multi-node redundancy
- Loads migrate automatically when a node fails
- Diversified power risk
- Multiple power sources plus FGS long-duration storage; no single-grid dependence
- Elastic capacity
- Plug-and-play PODs scale capacity up and down across the network
- Service continuity
- Customers see one API and one SLA, not a single facility
04 / How the network is built
From sunlight to intelligence, every node runs the same stack.
- 01
XCITY FGS Long-duration storage
Turns a solar plant into a 7×24 independent microgrid
- 02
XCITY MGrid Microgrid & distribution
Grid-tied or off-grid
- 03
XCITY POD Modular data center
Factory-built, container-delivered, fast to deploy on site
- 04
Routing & dispatch Unified dispatch
Unified dispatch, metering and settlement that link scattered nodes into one network
Sunlight Power Compute Intelligence
Nodes built where the power is
Argentina
Million-acre solar, carrying training and inference
United States
Surplus power, close to customers
Europe
Surplus power, close to customers
05 / Who it's for
AI companies and model teams
Elastic training and inference compute on demand
Enterprises and developers
One API to 250+ models, no hardware to manage
Super individuals
Create freely on competitive energy, compute, models and an agent ecosystem
Power and land owners
Join the network with surplus power or an existing solar plant and become a node partner
Connect to next-generation compute
Compute that is always available and never down — at the cost of sunlight.