Every personal device contains idle compute. Fabric turns these devices into a global, invisible datacenter.
Based on current observations across Fabric's network. Average home or personal device (laptop, desktop, mini-PC):
| Component | Typical Value |
|---|---|
| Usable compute power | 50 to 150 watts (assume 100 watts average) |
| Sustained compute capacity | ~2 teraflops (CPU, GPU, neural engine combined) |
| Node uptime (usable) | 25 to 60% depending on operations |
| Cost to Fabric | $0 capital expenditure |
| Customer acquisition cost per node | $1 to $10 (growth dependent) |
| Maintenance cost | Near-zero (agent updates only) |
| FLOP / teraflop / petaflop / exaflop | Floating Point Operations Per Second. 1 teraflop = 1 trillion ops/sec; 1 petaflop = 1,000 teraflops; 1 exaflop = 1,000 petaflops |
| Watt / kilowatt / megawatt / gigawatt | Unit of power. 1 KW = 1,000 watts; 1 MW = 1,000 KW; 1 GW = 1,000 MW. Typical home ~1 KW; datacenter 10 to 100+ MW |
| Fabric Today | ~70 nodes; 7 KW; 140 teraflops; 2 team members |
| Category | Fabric | Datacenter |
|---|---|---|
| Hardware Capital Expenditure | $0 | $40 million to $50 billion |
| Build Time | 1 to 3 months per 10x nodes | 2 to 15 years |
| Operating Cost | Software + small team | Power, cooling, physical staff, repairs |
| Scaling Speed | Viral, exponential | Linear, construction-limited |
Across Fabric's network today, the average device delivers ~2 teraflops of sustained compute, combining CPU, GPU, and neural engine throughput from M1, M2, M3-class Macs and multi-core Windows/Linux desktops.
With this real-world capability, 100 nodes deliver ~200 teraflops, 1,000 nodes reach the 2 petaflop range, and 10,000 nodes become a 1-megawatt cluster delivering ~20 petaflops of compute.
At a million nodes, Fabric enters the exaflop era with 2 exaflops of distributed compute, matching the footprint of a regional cloud datacenter. At 10 million nodes, Fabric passes 1 gigawatt of compute and exceeds 20 exaflops, rivaling hyperscale datacenter deployments that would cost $30 to $50 billion and take a decade to build.
For reference: The Stargate Project, announced by OpenAI, Oracle, and SoftBank, plans to spend $500 billion over 4 years to build AI infrastructure. Fabric's distributed model could achieve comparable compute capacity for less than 0.1% of that cost.