Glossary: Sovereign AI, Carbon-Aware Computing & African Infrastructure

Africa's datacenter market is growing 20%+/year, driven by cloud adoption and digital transformation.

An AI agent is an autonomous system that uses LLMs to reason, plan, and execute multi-step tasks.

An embedding is a vector representation of data (text, images) that captures semantic meaning.

AI inference is the process of using a trained model to make predictions on new data.

AI training is the process of optimizing model parameters using labeled data and compute resources.

An API is a set of protocols allowing software components to communicate with each other.

Attention allows AI models to focus on relevant parts of input when generating output.

Audit logging records all system activities for security, compliance, and forensic analysis.

B2B SaaS is software delivered as a service to business customers via cloud, billed by subscription.

Backpropagation computes gradients of the loss function with respect to model weights for training.

Capacity factor is the ratio of actual energy output to maximum possible output of a power plant.

CAPEX is capital expenditure — money spent on long-term assets like hardware and buildings.

Carbon intensity measures CO2 emissions per unit of electricity consumed, expressed in gCO2/kWh.

A carbon offset is a credit for reducing emissions elsewhere to compensate for your own emissions.

Carbon-aware GPU computing schedules workloads based on renewable energy availability to minimize carbon emissions.

CFC is Morocco's international financial center, offering tax incentives for financial and tech companies.

CCME is Morocco's advisory council for the Moroccan diaspora (MRE).

Cloud native is an approach to building and running applications that exploits cloud computing.

Colocation is a datacenter service where customers rent space, power, and cooling for their own servers.

COP is the annual UN climate change conference where nations negotiate emission reductions.

CRAC units are specialized air conditioners that cool datacenter air and maintain humidity.

CUDA is NVIDIA's parallel computing platform and programming model for GPU acceleration.

Data privacy is the right of individuals to control how their personal data is collected and used.

Data residency is the physical location where data is stored, often mandated by law.

Data security is the protection of digital data from unauthorized access, corruption, or theft.

Data sovereignty is the principle that data is subject to the laws of the country where it is stored.

A datacenter is a facility housing computer systems and networking infrastructure for computing and storage.

Datacenter redundancy is the duplication of critical components to ensure continuous operation if one fails.

Datacenter uptime is the percentage of time a facility is operational, typically expressed as a percentage over a year.

DevOps combines development and operations to deliver software faster and more reliably.

Digital sovereignty is a nation's ability to control its digital infrastructure, data, and technologies.

Distributed training splits AI model training across multiple GPUs or nodes for faster convergence.

Docker is a platform for packaging applications and dependencies into portable containers.

An edge datacenter is a smaller facility located close to end-users for low-latency applications.

Encryption converts data into code to prevent unauthorized access, using cryptographic algorithms.

Energy storage systems (batteries, pumped hydro) store electricity for use when generation is low.

ESG is a framework for evaluating organizations on environmental, social, and governance criteria.

Fine-tuning adapts a pre-trained model to a specific task using domain-specific data.

A foundation model is a large pre-trained AI model that can be adapted to many downstream tasks.

The GDPR is the EU's data protection regulation, setting global standards for privacy.

GPU as a Service provides on-demand GPU access via cloud APIs, billed per second or per hour.

GPU cloud is a cloud computing service that provides access to graphics processing units (GPUs) on-demand for AI, ML, and HPC workloads.

A GPU cluster is a network of GPUs working together for distributed AI training and HPC workloads.

Gradient descent is an optimization algorithm that minimizes loss by iteratively updating weights.

Green hydrogen is hydrogen produced via electrolysis powered by renewable energy.

Hot aisle/cold aisle is a datacenter layout that separates cool intake air from hot exhaust air.

IaaS provides virtualized computing resources — servers, storage, networking — via cloud.

InfiniBand is a high-speed networking standard used in HPC and AI GPU clusters.

IaC manages infrastructure through code rather than manual processes, using tools like Terraform.

Kubernetes is an open-source container orchestration platform for deploying and scaling applications.

An LLM is an AI model trained on massive text data to understand and generate human language.

Law 09-08 is Morocco's data protection law, governing the processing of personal data.

LCOE measures the average cost of electricity generation over a power plant's lifetime.

Liquid cooling uses liquid instead of air to cool datacenter servers, enabling higher density and efficiency.

LLM serving is the deployment of large language models for inference via APIs or applications.

MENA cloud computing is a $20B+ market growing 25%/year, with Morocco as a key hub.

A microgrid is a localized energy grid that can operate independently or connected to the main grid.

MLOps is the practice of deploying, monitoring, and maintaining ML models in production.

Quantization reduces model precision (FP32→FP16→INT8) to decrease memory and increase speed.

Morocco's AI strategy aims to position the country as an African AI hub by 2030.

Morocco Digital 2025 is the Kingdom's national digital transformation strategy.

Morocco's startup ecosystem is growing rapidly, with 300+ startups and $500M+ funding in 2024.

Morocco Tech is the Kingdom's initiative to position Morocco as a digital hub for Africa and MENA.

MRE are Moroccans living abroad — 5.8 million people transferring $13B+ annually to Morocco.

Multi-cloud uses multiple cloud providers to avoid vendor lock-in and optimize costs.

Net zero means balancing emissions produced with emissions removed, achieving carbon neutrality.

Network bandwidth is the maximum data transfer rate of a network connection, measured in Gbps or Tbps.

Network latency is the time for data to travel between source and destination, measured in milliseconds.

Noor Ouarzazate is the world's largest concentrated solar power complex, in southern Morocco.

The NVIDIA A100 is a datacenter GPU built on the Ampere architecture, widely used for AI training and HPC.

The NVIDIA B200 is a next-generation Blackwell GPU with 192GB HBM3e memory for trillion-parameter models.

The NVIDIA H100 is a datacenter GPU built on the Hopper architecture, designed for AI training and inference at scale.

The NVIDIA H200 is the successor to the H100, featuring 141GB HBM3e memory for even larger AI model training.

NVLink is NVIDIA's high-speed GPU interconnect providing up to 900 GB/s bandwidth between GPUs.

Open banking allows third-party financial services to access bank data via APIs, with customer consent.

OPEX is operating expenditure — ongoing costs for running a business, like cloud subscriptions.

PaaS provides a cloud platform for developing, running, and managing applications without infrastructure management.

Power density is the amount of electrical power delivered per rack in a datacenter, measured in kW/rack.

A PPA is a long-term contract to buy electricity directly from a power generator at a fixed price.

PUE is the ratio of total facility power to IT equipment power, measuring datacenter efficiency.

PyTorch is an open-source deep learning framework developed by Meta, widely used in research and production.

RAG combines a retrieval system with an LLM to ground responses in external knowledge.

RDMA (Remote Direct Memory Access) allows direct memory access between GPUs without CPU overhead.

Compliance means adhering to laws, regulations, and industry standards relevant to your business.

Renewable energy comes from natural sources that replenish — solar, wind, hydro, geothermal.

ROI measures the profitability of an investment relative to its cost.

SaaS delivers software via cloud subscription, accessed through web browsers without installation.

A server rack is a standardized frame that holds IT equipment in a datacenter, typically 42U or 48U height.

An SLA is a contract defining the guaranteed level of service, including uptime, performance, and support response.

Solar energy is power generated from sunlight using photovoltaic panels or concentrated solar power.

Sovereign AI is a nation's capability to develop, deploy, and control AI infrastructure within its borders.

Tamwilcom is Morocco's public financing institution for startups and SMEs.

Tanger Med is Morocco's mega-port and free zone, the largest port in Africa and a logistics hub.

TCO is the complete cost of owning and operating an asset over its lifetime.

Tensor Cores are specialized GPU hardware units that accelerate matrix operations for AI workloads.

TensorRT is NVIDIA's high-performance deep learning inference optimizer and runtime.

A Tier III datacenter has redundant power and cooling with 99.982% availability (N+1 redundancy).

The Transformer is a neural network architecture using self-attention, the basis for modern LLMs.

A UPS provides backup power during outages, ensuring continuous datacenter operation.

A vector database stores and queries high-dimensional vectors for AI similarity search.

Wind energy is power generated from wind using turbines, onshore or offshore.

Zero Trust is a security model that never trusts any user or device, always verifies.