Platform · OrchestrationRafay · Strategic Delivery Partner
GPU PaaS & AI Factory orchestration
Turn raw GPU capacity into governed, self-service AI cloud services — the operating layer for neoclouds, sovereign clouds, telcos and enterprises.
- Services you can launch: Kubernetes, SLURM-as-a-Service, bare-metal GPUs, VMs, workbenches and containers from one catalog
- Token Factory: token-metered, OpenAI-compatible model APIs with usage tracking, chargeback and monetization
- Multi-tenancy: hard and soft isolation across customers, business units, projects, quotas and policies
- Sovereign operation: in-region, private or fully air-gapped environments with data kept inside defined borders
- Observability & Day-2: unified estate visibility, synthetic monitoring, AI-assisted root-cause and automated remediation workflows
GPU PaaSToken FactorySLURM-aaSMulti-tenancySelf-service portalsCost & chargebackNVIDIA reference architecture
MAK capabilityMAK is Rafay’s strategic and regional delivery partner across MEA — certified platform engineers delivering design, deployment, tenant onboarding and Day-2 managed operations, with production deployments live in Africa and the GCC.
Solution · Agentic AIMAK-built platform
MAK AI Agentic — sovereign agentic AI platform
A sovereign, model-agnostic platform that turns enterprise and government knowledge into working agents — deployable on-prem or fully air-gapped.
- Sovereign data layer: ingestion, chunking, embeddings and vector search over your own corpora, entirely in-country
- Retrieval-augmented generation: grounded answers with citations, access-aware retrieval and freshness control
- Agent orchestration: multi-step tool-calling agents that query systems, trigger workflows and hand off to humans
- Bilingual by design: Arabic and English understanding, generation and voice front-ends
- Enterprise connectors: document stores, ticketing, ERP and line-of-business systems, with guardrails and full audit trail
RAGVector searchAgent orchestrationArabic + EnglishOn-prem / air-gappedModel-agnosticKiosk & voice front-ends
MAK capabilityDesigned, built and supported in-house by MAK — deployed for public-sector and national programmes where data may not leave the country.
Solution · SecurityMAK-built layer
Secure AI operations layer
The governance and security envelope around an AI factory — so a multi-tenant GPU platform can be trusted with regulated and classified workloads.
- Zero-trust access: identity-aware RBAC across clusters, tenants, notebooks and model endpoints
- Workload isolation: namespace, node and GPU-level segmentation with policy enforcement at admission
- Secrets & key management: vaulted credentials, rotation and encryption in transit and at rest
- Model & prompt governance: content guardrails, prompt/response logging, jailbreak and data-exfiltration controls
- Audit & compliance: immutable audit trails, posture reporting and SOC/SIEM integration for regulator-ready evidence
Zero trustRBACPolicy-as-codeGuardrailsAudit trailSIEM integrationAir-gap ready
MAK capabilityDelivered as part of MAK’s platform engagements — architected against national telecom and public-sector security requirements.
Partner · NetworkingAviz Networks · MEA
AI-optimized, vendor-agnostic networking
Open networking software for the AI era — build the fabric, see every packet, and run the NOC with agentic AI, on any NOS, switch or ASIC.
- ONES (Open Network Enterprise Suite): orchestrate and operate SONiC fabrics for data centre, edge and DCI — plus ONES for the NVIDIA AI Factory
- SONiC distributions & 24/7 TAC: Aviz Certified Community SONiC and Broadcom Enterprise SONiC, vendor-neutral to cut CapEx and lock-in
- Deep network observability: Aviz Packet Broker, Service Nodes (including on NVIDIA BlueField-3 DPUs), virtual ASN/TAP and Flow Vision
- Fabric Test Automation Suite: automated validation of fabrics before and after go-live
- Network Copilot™ & AI agents: agentic AI for NetOps — natural-language troubleshooting and autonomous operations
- AI factory fabrics: NVIDIA Spectrum-X, InfiniBand, BlueField and NVL72 designs, simulated in NVIDIA DSX Air digital twin before deployment
SONiCONESSpectrum-XInfiniBandBlueField-3 DPUPacket BrokerNetwork CopilotDSX Air
MAK capabilityMAK brings Aviz to the Middle East & Africa as regional integrator — fabric design, validation and NetOps for NEO clouds, telcos and public-sector AI factories.
Partner · Inference privacyProtopia AI · MEA
The inference privacy layer for AI factories
Encryption ends where AI begins: at inference, data is plaintext at the model host. Protopia removes that exposure — so sensitive data and shared GPUs can finally meet.
- Stained Glass Transform™: replaces raw input with model-specific randomized representations usable only by the target model — no change to the model or serving stack
- Post-training, not re-training: applied in a post-training step so the unmodified model stays accurate
- Private multi-tenant inference: secure tenancy without dedicated hardware isolation — more tenants and tokens on the same fleet
- SafeClaw™: build AI agents on sensitive data with zero plaintext exposure, including private tool-calling for agentic workflows
- Economics: ends large-granularity GPU carve-outs that leave AI factories idle — higher utilization and materially lower TCO
- Fit: sovereign AI, neoclouds, defence and government, financial services and regulated enterprises
Stained Glass TransformSafeClawPrivate inferenceMulti-tenantNo hardware isolationAgentic tool-callingNVIDIA AI Factory aligned
MAK capabilityMAK introduces Protopia into MEA sovereign and neocloud builds — the privacy layer that lets regulated data run on shared, monetizable GPU infrastructure.
Partner · AI-native storageScality · WEKA · VAST · DDN
Storage engineered for AI workloads
Data is the bottleneck in an AI factory. MAK designs the storage tier to match the GPU tier — from training throughput to sovereign, cyber-resilient capacity at exabyte scale.
- Scality RING: S3 object plus file (NFS/SMB/FUSE) on a patented MultiScale architecture — independent scaling of capacity and performance to hundreds of petabytes, with fine-grained placement down to site, rack and drive
- Scality CORE5 & ARTESCA: end-to-end cyber resilience, S3 Object Lock immutability and ransomware protection; ARTESCA delivers Kubernetes-native, immutable S3 from small footprints upward
- Scality for AI: metadata-driven retrieval for chunked documents and embeddings, with integrations across vector databases, RAG frameworks and training pipelines
- WEKA: ultra-low-latency parallel storage for large-scale training and inference pipelines
- VAST Data: disaggregated all-flash for unified file, object and database-style AI access
- DDN: proven HPC and AI parallel file systems for the most demanding GPU clusters
- Engineering: GPUDirect Storage, checkpoint sizing, data-pipeline and tiering design to keep GPUs saturated
S3 objectParallel file systemGPUDirect StorageImmutabilityVector & RAG dataMulti-tenantExabyte scale
MAK capabilityMAK specializes in AI-native and AI-workload storage — sizing, benchmarking and integrating the data tier as part of every AI Factory build.
Services · NVIDIANVIDIA Inception
NVIDIA AI Factory services — NEO cloud & enterprise
End-to-end NVIDIA engineering: from reference architecture and cluster bring-up to tuned, production-grade multi-tenant AI platforms.
- Design: AI Factory reference architectures sized for training, fine-tuning and inference — accelerated systems, rack power/cooling and scale-unit planning
- Fabric: Quantum InfiniBand and Spectrum-X Ethernet design, rail-optimized topologies, BlueField DPU offload and multi-rail RDMA
- Bring-up & validation: cluster provisioning, driver/firmware baselines, NCCL and multi-node benchmarking, burn-in and acceptance testing
- Software stack: NVIDIA AI Enterprise, NIM inference microservices, NeMo pipelines, GPU Operator, MIG/time-slicing and container runtimes
- Operations: DCGM-based telemetry, GPU health and utilization management, capacity and scheduling policy, patch and lifecycle management
- NEO cloud enablement: turning accelerated capacity into sellable, multi-tenant services with metering and tenant isolation
Reference architectureInfiniBand / Spectrum-XNCCL tuningNVIDIA AI EnterpriseNIM & NeMoMIGDCGMAcceptance testing
MAK capabilityMAK is an NVIDIA Inception member with certified engineers delivering accelerated-computing builds across research, government and operator environments.
Services · Telco-gradeNational operator programmes
Telco-grade AI platform engineering
MAK operates at national-operator scale — where an AI platform must satisfy procurement, security, sovereignty and SLA scrutiny before a single GPU is racked.
- Large-scale RFP engineering: multi-workstream technical responses with clause-by-clause compliance matrices running to several hundred requirements
- Platform architecture: sovereign multi-tenant GPU PaaS, inference services, HLD/LLD and integration design alongside prime contractors and OEMs
- Security architecture: zero-trust, tenant isolation, key management and regulator-facing controls built into the platform design
- Delivery governance: acceptance criteria, test plans, phased go-live, knowledge transfer and documented handover
- Managed Day-2: SLA-backed operations, capacity and tenant management, and continuous optimization after go-live
Multi-workstream RFPCompliance matrixHLD / LLDSovereign multi-tenantAcceptance criteriaSLA operations
MAK capabilityDelivered on active telecom-operator engagements in the GCC. Customer names are withheld under commercial confidentiality and shared on request under NDA.
Solution · Enterprise & mid-market4–24 GPU segment
Private Cloud AI — enterprise-grade, right-sized
Not every organization needs a national AI factory. MAK packages the same engineering into a compact, private AI platform that lands in weeks, not quarters.
- Pre-validated stack: accelerated compute, fabric, storage and platform software delivered as one tested, supported system
- Right-sized: engineered for the four-to-twenty-four GPU segment, with a clean growth path as demand scales
- Inference-first: tuned for private model serving, RAG over corporate data and agentic workflows rather than frontier training
- Self-service on day one: Kubernetes, VMs and notebooks with quotas, chargeback and departmental multi-tenancy
- Data stays home: on-premises or air-gapped, satisfying residency, privacy and sector regulation
- Commercially flexible: capital purchase or a managed, subscription-based consumption model with a single support contact
Private AI platform4–24 GPUsInference-firstRAG & agentsDepartmental tenancyOn-prem / air-gappedManaged option
MAK capabilityVendor-neutral by design — MAK selects the accelerated compute platform on merit for each customer, then owns integration, validation and lifecycle support.
