Compute · infrastructure

Compute & infrastructure

Inference clusters, customer environments, and resource utilization across regions.

GPUs allocated

14 A100 · 8 H100

Avg utilization

76%

+4pp WoWtarget 70-85%

VRAM in use

1.4 TBof 1.8

78% capacity

Network I/O

42 Gbps

peak 96 Gbps

Reserved capacity

12 GPUs

1-yr commit

Inference clusters

Abstract cluster view across deployment regions — physical providers omitted

Inference Cluster Alphaus-east-1

Detroit HQ · 8× A100 · 80GB VRAM/GPU

Healthy

GPU78%

Memory64%

Hosting: Philotic-1 v2.4.1 · Cali-OEM v3 · Cali-Insurance v2.1

Inference Cluster Betame-dubai-1

Dubai DSO · 4× H100 · 80GB VRAM/GPU

Healthy

GPU52%

Memory48%

Hosting: Philotic-1 v2.4.1 · Llama 3.3 70B

Embedding Clusterus-east-1

Detroit HQ · 2× A100 · 40GB VRAM/GPU

Healthy

GPU85%

Memory72%

Hosting: Embed-Sovereign v1.4

Training Clusterme-dubai-1

Dubai DSO · 8× H100 · 80GB VRAM/GPU

Degraded

GPU92%

Memory88%

Hosting: Banking-Arabic v1.0-rc · training

Resource utilization · 24h

GPU compute and memory pressure across all clusters

Managed customer environments

Multi-cloud presence — Philotic substrate sits on top of customer infrastructure

Azure East US

Azure · eastus

Caliber Collision

Workloads

Cali-OEM
Cali-Insurance
Knowledge ingest

AWS us-east-2

AWS · us-east-2

Engage Enterprise

Workloads

Phil orchestrator
Doc Processor

Private Cloud

Private · me-dubai-1

Dubai DSO sovereign

Workloads

Phil orchestrator
Banking-Arabic training

Scaling policies

Inference Cluster Alpha

Sustained util > 85% for 5m

Burst +2 A100 from reserved pool

active

Inference Cluster Beta

p99 latency > 3000ms for 3m

Burst +2 H100, route 30% to Alpha

active

Embedding Cluster

Queue depth > 5000

Spawn embed worker · max 2

active

Training Cluster

Manual

Reserved · Banking-Arabic until 2026-05-12

paused

Capacity headroom

Burst capacity

+8 GPUs · 2 regions

Reserved utilization

82%

Spot utilization

34%

Forecast headroom

34 days at current growth

Storage capacity

46 TB · 38% used

Vector index size

4.4 GB · sharded ×4