Pricing
MAP™ Brick Pricing
Pay-As-You-Train Model
Revolutionary metering system that precisely quantifies resource consumption based on actual usage.
Advantages Over Traditional Models
| Feature | Traditional Cloud Billing | MAP Metering System |
|---|---|---|
| Granularity | VM-hours or GPU-hours, regardless of load | Per-core, per-second, precision-adjusted billing |
| Fairness | Users pay for idle time (e.g., unused CPU) | Pay only for active computation (processing-seconds) |
| Energy Transparency | Opaque power costs bundled into flat rates | Net energy consumption with reversible credits |
| Security Premiums | Uniform storage costs, even for encrypted data | Variable rates for encrypted/homomorphic storage |
Core Type Pricing
Tensor Cores
$0.001
per tensor-second (FP32)
Symbolic Cores
$0.005
per symbolic-second (exact arithmetic premium)
Stochastic Cores
$0.0005
per Monte Carlo trial
Metering Components
Processing-Seconds
Definition: Time (in seconds) a specific MAP core is actively executing a task.
Measurement:
- Core-Specific Clocks: Each core has dedicated hardware counters to track active cycles
- Dynamic Precision Scaling: Tasks using lower precision consume fewer "effective processing-seconds"
Example:
A tensor core running a matrix multiplication in FP32 for 10 seconds = 10 FP32 tensor-seconds
The same task in 8-bit POSITs completes in 2 seconds = 2 POSIT-8 tensor-seconds
Memory Usage
Definition: Amount of hypergraph memory allocated during computation.
Measurement:
- Relational Memory Units (RMUs): Memory is billed per gigabyte-hour based on the complexity of stored hypergraphs
- Sparse vs. Dense: Sparse data incurs lower costs due to compression
Example:
Storing a 100GB social network hypergraph for 1 hour = 100 RMU-hr
Compressed sparse tensor storage reduces this to 20 RMU-hr
Storage
Definition: Long-term retention of data in MAP-optimized formats.
Measurement:
- Mathematical Object Storage (MOS): Costs scale with precision and structure
- Exact Arithmetic Storage: Fixed cost for 256-bit cryptographic primes
- Lossless Compression: Symbolic expressions stored as syntax trees
Example:
Storing a 1TB encrypted dataset with homomorphic encryption = 1TB × 2× MOS rate (security premium)
Revenue Model (Pay-As-You-Train)
Instead of one time fees and subscriptions, we partner with companies on a revenue share model and cover the Math Aware Processor device CAPEX moving the cost of computing to OPEX.
Pay-per-use metering system for Math-Aware Processors (MAPs) revolutionizes computational billing by precisely quantifying resource consumption based on exact processing-seconds, memory, and storage used.
This system leverages the unique architectural features of MAPs, like adaptive precision, domain-specific cores, hypergraph memory, to create a granular, transparent, and fair billing model.