HARCH · INTELLIGENCECasablanca · Dakhla · Dakar
Harch IntelligenceSovereign AI Compute
on African soil · powered by African sun
1,798
Carbon-Optimized GPUs (Planned)
5
Hubs Across Africa
99.98%
Target Uptime SLA
// 01Overview
Africa's largest AI hyperscale
Africa imports 95% of its AI compute from US and EU clouds. Harch Intelligence changes this. We plan 1,798 GPUs across 5 hubs in Morocco, Senegal, and Côte d'Ivoire — sovereign AI infrastructure designed to run on African renewable energy. Train your models on African soil, under African data sovereignty law. No GDPR conflicts, no US CLOUD Act exposure.
Casablanca HQDakhla FlagshipTanger CableDakar HubAbidjan Hub
FLAGSHIP FACILITY
Dakhla · Hub #001
512 GPU · PUE 1.15 · 100% renewable
// 02Sovereignty
African data, under African law
Harch Intelligence operates exclusively under Moroccan Law 09-08 and the African Union Convention on Cybersecurity. Your training data, model weights, and inference traffic never leave African soil. No US CLOUD Act exposure. No GDPR cross-border conflicts. Sovereign AI by design.
Law 09-08·Sovereign by design
Governing framework
MA-09/08
Moroccan Law 09-08
Personal data protection — CNDP registration, on-soil processing, mandatory breach notification.
AU-CONV
AU Convention on Cybersecurity
Pan-African data sovereignty framework ratified by 14 member states including Morocco, Senegal, Côte d'Ivoire.
US-CLOUD
US CLOUD ActPASS
NOT applicable — Harch Intelligence has zero US entity, zero US-issued warrants honored, zero data replication to US regions.
GDPR-ART49
GDPR Article 49 derogationsPASS
NOT triggered — data never leaves Africa, so no adequacy decision or derogation is required for EU customers.
// 03Live Dashboard
One console for compute, storage, and inference
HarchOS unifies GPU scheduling, object storage, and inference endpoints under a single control plane. Switch views below to inspect the same fleet from three angles — every metric live, every workload auditable, every byte sovereign.
https://harchos.africa/dashboard
live
GPU CLUSTER · 1,798 H100 SXM5 (PLANNED)
GPU Utilization
87.4%
Active Jobs
142
PUE
1.15
Carbon Intensity
0.02 gCO2/kWh
GPU rack · live
8/8 nodes readygpu-00
89%
47°C
gpu-01
85%
46°C
gpu-02
91%
48°C
gpu-03
73%
44°C
gpu-04
78%
45°C
gpu-05
87%
47°C
gpu-06
82%
46°C
gpu-07
90%
48°C
harchos-cli · provision
$ harchos login --region dakhla
→ authenticated · tenant harch-corp-prod
$ harchos gpu provision --type H100 --count 8
→ reserving 8x H100 SXM5 · Dakhla Hub #001
→ InfiniBand fabric configured · NCCL verified
→ liquid cooling loop nominal · 24.1°C delta
→ provisioning ok · elapsed 87.4s
scale event · 1 → 1024 GPUs87.4s
kubectl get pods -n harchos-compute● 8/8 readylive·Dakhla · Hub #001·8/8 nodes ready·100% renewable
// 04Hardware
Hyperscale datacenter infrastructure
Tier-III designed datacenters with N+1 cooling, 2N power, biometric access. NVIDIA H100 and H200 GPU clusters. Liquid cooling for density. PUE 1.15 (industry average 1.5). Powered by 100% renewable solar + wind. Submarine cable connectivity via ACE, MainOne, Africa Coast to Europe.
PUE 1.15
Power usage effectiveness, 30% below industry
100%
Renewable energy solar + wind
Tier-III
Datacenter design certification
// 05Hardware
NVIDIA H100 and H200 GPU clusters
1,798 GPUs planned for deployment — H100 SXM5 80GB, H200 141GB, A100 80GB, L40S for inference. NVLink fabric, InfiniBand 400G, liquid cooling. Carbon-aware scheduler routes workloads to the greenest hub in real-time.
| GPU | Memory | Bandwidth | TDP | Status |
|---|---|---|---|---|
| H100 SXM5 | 80 GB HBM3 | 900 GB/s NVLink | 700 W | In production |
| H200 SXM5 | 141 GB HBM3e | 900 GB/s NVLink | 700 W | Q1 2026 |
| A100 80GB | 80 GB HBM2e | 600 GB/s NVLink | 400 W | In production |
| L40S | 48 GB GDDR6 | 864 GB/s PCIe5 | 350 W | Inference fleet |
| Hyperscale fabric | InfiniBand NDR 400G | 16:1 non-blocking | — | Live |
harchos-cli · provision
$ harchos login --region dakhla
→ authenticated · tenant harch-corp-prod
$ harchos gpu provision --type H100 --count 8
→ reserving 8x H100 SXM5 · Dakhla Hub #001
→ InfiniBand fabric configured · NCCL verified
→ liquid cooling loop nominal · 24.1°C delta
→ provisioning ok · elapsed 87.4s
ubuntu@harch-gpu-001:~$ nvidia-smi --query-gpu=name,memory.total --format=csv
name, memory.total [MiB]
NVIDIA H100 80GB HBM3, 81920 MiB
NVIDIA H100 80GB HBM3, 81920 MiB
... 8/8 GPUs ready for workload
// 06Models
African-language LLMs
47 African languages by 2027. Harch Intelligence trains and hosts open-weight LLMs for Arabic, Wolof, Swahili, Hausa, Amazigh, and more — on African compute, with African data, for African builders.
ArabicWolofSwahiliHausaAmazighDarijaFulfuldeYorubaIgboShonaZuluXhosaBambaraSoninkéMalagasy+32 more → 47 by 2027
13B
Darija-13B
Moroccan Darija + MSA
Fine-tuned on 4.2B Moroccan tokens. Open weights. Hugging Face: harch/darija-13b.
7B
Sahel-7B
Wolof, Fulfulde, Bambara
Sahel region multi-lingual. Trained with Senegalese universities. Open weights.
70B
Swahili-70B
Swahili, English
Flagship East African model. 70B parameters. 8-node H100 training cluster.
8B
Amazigh-8B
Tamazight, Tachelhit, Tarifit
First open-weight Amazigh LLM. IRCAM partnership. Tifinagh + Latin scripts.
// 07Reliability
Target: 99.98% uptime
Every Harch Intelligence hub is designed with redundant everything: power (2N UPS + diesel backup), cooling (N+1 chillers), network (3 submarine cables, 2 terrestrial), and security (24/7 SOC, biometric, mantrap). Target: 99.98% uptime SLA with service credits. Tier-III design with 2N power, N+1 cooling.
Operations centre · 24/7 NOC
// 08Network
5 hubs, 3 countries, 1 sovereign fabric
From Dakhla on the Atlantic to Casablanca on the Mediterranean, Harch Intelligence spans 5 hubs in development, connected by 4 submarine cable systems. Sub-50ms latency to 1.2B Africans. Zero egress fees between Harch regions.
Dakhla
Casablanca
Tanger
Dakar
Abidjan
Planned hub· 5 planned
Submarine cable landing· 4 cables
Planned 2027–2029· 4 cities
submarine fabric
- ACE — Africa Coast to Europe · 12.8 Tbps · Casablanca + Dakar landing
- MainOne — West African loop · 4.96 Tbps · Casablanca + Dakar landing
- Africa Coast to Europe 2 — expansion 2026 · 16 Tbps · Tanger + Abidjan
- 2Africa — 37 African landings · 48 Tbps · Dakhla spur 2027
// 09Carbon
100% renewable, 0.02 gCO2 per kWh
Harch Intelligence is powered by Moroccan solar and Senegalese wind. Carbon intensity 0.02 gCO2/kWh — 89% below the global datacenter average of 0.45 gCO2/kWh. Every training job reports its carbon footprint. ESG-compliant AI by default.
Energy mix · live
0.02
gCO2/kWh — 89% below industry average
Energy mix · live
100%
Renewable energy (solar + wind)
Energy mix · live
1.15
PUE — 30% below industry average
Energy mix · live
87
Job-level carbon footprint reporting
100% renewable
Solar 62%
Wind 38%
Energy mix · live
// 10Comparison
Harch Intelligence vs US cloud
| Criterion | Harch | US Cloud (AWS/GCP) |
|---|---|---|
| H100 price/hr | $1.80 | $3.50 |
| Data sovereignty | African soil | US CLOUD Act |
| Energy | 100% renewable | Mixed |
| PUE | 1.15 | 1.5-1.6 |
| Egress fees | 0 between African regions | $0.09/GB |
| Latency from Africa | <50ms | 150-300ms |
| Compliance | Moroccan Law 09-08 | US jurisdiction |
| Carbon footprint | 0.02 gCO2/kWh | Variable |
// 11Plans
Choose your GPU plan
Three plans for every AI workload. All include HarchOS, data sovereignty, and renewable energy.
On-Demand
Pay per hour
$1.80/hr H100
1-8 GPUs
- H100, H200, A100 instances
- Scale in 90 seconds
- PyTorch, JAX, vLLM pre-installed
- HarchOS dashboard
- Carbon footprint per job
- No commitment
Best value
Reserved
Best for teams
$1.08/hr H100
8-256 GPUs
- Everything in On-Demand
- 40% discount vs on-demand
- Reserved capacity 1-3 years
- Priority support SLA
- Custom images
- Volume discounts
Dedicated
Private cluster
Custom
256+ GPUs
- Everything in Reserved
- Dedicated hardware
- Private network fabric
- Custom SLA (target 99.99%)
- On-site engineer
- Multi-year contract
// 12Innovation
Beyond conventional cloud
African LLMs. Optical computing. Quantum-ready. The future of AI infrastructure is being built today — by us.
African LLMs
Pre-trained LLMs for African languages (Arabic, Wolof, Swahili, Hausa, Amazigh). Open weights. HarchOS fine-tuning pipeline. 47 languages target 2027.
Optical computing
Silicon photonics for inter-GPU communication. 10x bandwidth, 50% lower latency. Pilot 2028. Partnership with Lightmatter.
Quantum-ready
Quantum annealing for optimization workloads. Partnership with D-Wave. Pilot quantum-classical hybrid pipeline 2028.
Edge inference
HarchOS Edge — deploy models to edge devices (phones, IoT, vehicles). Sub-10ms latency. On-device fine-tuning. Pilot 2027.
// 13Case studies
Sovereign AI in production
From Moroccan banks training fraud detection models to Senegalese startups building African-language LLMs, HarchOS powers sovereign AI across the continent.
Get started free// 14Testimonials
Trusted by African AI teams
“We trained our fraud detection model on HarchOS. 40% cheaper than AWS, and the data sovereignty compliance was instant — no legal review needed.”
Yassine El Mansouri
CTO, Moroccan Bank of Commerce
“HarchOS lets us train African-language LLMs on African compute. Latency from Dakar is 12ms. This changes everything for Senegalese AI.”
Awa Diop
Founder, Noolon AI Senegal
“The carbon footprint per job reporting is incredible. We can show our ESG board exactly how much CO2 each training run used. Game changer.”
Laila Benali
Head of AI, OCP Group
// 15Flow
From signup to first model
Five steps. Ninety seconds to first GPU. Two weeks to production inference.
01
Signup
Create HarchOS account. Verify identity (KYC per Moroccan law). Add payment method.
$ harchos signup --email you@startup.ma --verify-kyc
02
Provision
Choose GPU type, quantity, region. Scale from 1 to 1024 GPUs in 90 seconds.
$ harchos gpu provision --type H100 --count 8 --region dakhla
03
Connect
SSH, Jupyter, or HarchOS SDK. PyTorch, JAX, vLLM pre-installed. Import your data.
$ ssh trainer-0.dakhla.harchos.africa -J bastion.harchos.africa
04
Train
Launch training jobs. Real-time telemetry. Auto-scaling. Spot instances for 70% savings.
$ harchos job submit --script train.py --gpus 8 --spot
05
Deploy
Deploy models to HarchOS Inference endpoints. Auto-scaling. Pay per token. Sub-50ms latency.
$ harchos deploy --model falcon-40b-finetune --replicas 4
harchos-cli · session log
$ harchos signup --email you@startup.ma --verify-kyc
step 01 — Signup
$ harchos gpu provision --type H100 --count 8 --region dakhla
step 02 — Provision
$ ssh trainer-0.dakhla.harchos.africa -J bastion.harchos.africa
step 03 — Connect
$ harchos job submit --script train.py --gpus 8 --spot
step 04 — Train
$ harchos deploy --model falcon-40b-finetune --replicas 4
step 05 — Deploy
model deployed · https://inference.harchos.africa
// 16Simulator
See your GPU cost
Enter your monthly GPU hours. See cost vs US clouds.
2,000GPU-hr / mo
10010K20K
Harch / mo
$3,600
Monthly GPU hours needed
AWS / mo
$7,000
vs US cloud
3-yr save
$122,400
3-yr · 3-yr save
Estimate based on $1.80/hr Harch vs $3.50/hr AWS p5.48xlarge. Custom quote within 48h.
// 17Sectors
Solutions by sector
Fintech
Healthcare
Agriculture
Government
Research
E-commerce
Media
Defense
// 18Applications
Sovereign AI for every workload
From LLM training to inference, from computer vision to scientific computing — HarchOS supports every AI workload. African data, African compute, African sovereignty.
01
LLM training
Train LLMs up to 70B parameters on African data. Multi-node training with NCCL over InfiniBand. Mixed precision, FSDP, DeepSpeed pre-configured.
02
Inference endpoints
Deploy models to autoscaling endpoints. vLLM, TGI, Triton supported. Pay per token. Sub-50ms latency from African edge.
03
Computer vision
Train and deploy CV models. Pre-built pipelines for agriculture (crop disease), security (ANPR), industrial (defect detection).
04
Scientific computing
HPC workloads for research institutions. Slurm scheduler. MPI, OpenMP. Partnership with Moroccan universities.
05
RAG & agents
Build RAG pipelines with African knowledge bases. Vector databases (pgvector, Weaviate). Agent frameworks (LangChain, LlamaIndex).
// 19Why Harch
The Harch Intelligence difference
African data sovereignty
Your data never leaves African soil. Moroccan Law 09-08 compliance. No US CLOUD Act exposure. No GDPR conflicts. Sovereign AI by design.
100% renewable energy
Powered by Moroccan solar and wind. PUE 1.15. Carbon intensity 0.02 gCO2/kWh. ESG-compliant AI. Carbon footprint per job reported.
48% cheaper than US clouds
H100 from $1.80/hr vs $3.50/hr AWS. No egress fees between African regions. Reserved capacity 40% off. Spot instances 70% off.
Harch Corp backing
Building in Public conglomerate. Strategic AI infrastructure for Morocco and Africa. We're building the continent's AI future.
// 20Process
From signup to first model
01
Signup
Day 1Create HarchOS account. Verify identity (KYC per Moroccan law). Add payment method.
02
Provision
Day 1Choose GPU type, quantity, region. Scale from 1 to 1024 GPUs in 90 seconds.
03
Connect
Day 1SSH, Jupyter, or HarchOS SDK. PyTorch, JAX, vLLM pre-installed. Import your data.
04
Train
Day 2+Launch training jobs. Real-time telemetry. Auto-scaling. Spot instances for 70% savings.
05
Deploy
Week 2+Deploy models to HarchOS Inference endpoints. Auto-scaling. Pay per token. Sub-50ms latency.
// 21FAQ
Frequently asked questions
NVIDIA H100 (80GB), H200 (141GB), A100 (80GB). L40S for inference. All with NVLink and InfiniBand. H200 available Q1 2026.
// 22Resources
Documentation & downloads
Architecture specs, API docs, case studies — everything for due diligence.
HarchOS architecture
PDF · 3.8 MBWhite paper on sovereign AI platform architecture. PDF, 24 pages.
API reference
PDF · 6.2 MBFull HarchOS SDK and REST API documentation. PDF, 86 pages.
Data sovereignty guide
PDF · 2.4 MBMoroccan Law 09-08, AU Convention, GDPR comparison. PDF, 18 pages.
OCP Group case study
PDF · 2.8 MBHow OCP trains fertilizer demand models on HarchOS. PDF, 10 pages.
// 23Geography
5 hubs across Africa
1,798 GPUs planned. 5 hubs. 3 countries. Submarine cable connected.
Dakhla
Dakhla
Flagship Hub
512 GPUs
Casablanca
Casablanca
HQ + Edge
256 GPUs
Tanger
Tanger
Submarine Landing
384 GPUs
Dakar
Dakar
West Africa Hub
256 GPUs
Abidjan
Regional Hub
192 GPUs
Lagos
Planning 2027
512 GPUs target
Nairobi
Planning 2028
512 GPUs target
Cape Town
Planning 2028
256 GPUs target
Accra
Planning 2029
256 GPUs target
HARCH · INTELLIGENCE
Free $500 credit. H100 instances in 90 seconds. Data stays in Africa.Ready to build sovereign AI?
H100 from $1.80/hr
Target: 99.98% SLA
Law 09-08
100% renewable
Backed by Harch Corp