Weather & agriculture
AI-supported forecasting and advisory services for farmers.
Source · Atlantic Re:think / Google
Published by Oshen
The InfrastructureAge
From Prompt to Power: Understanding Artificial Intelligence, Infrastructure, and Global Competition
By Hassan Mohamed
Artificial intelligence is reshaping the world’s infrastructure. Understanding the systems behind it matters more than following the headlines.
Scroll to exploreChapter 00
Executive Summary
What this report argues, at a glance.
Artificial intelligence has moved from research laboratories into daily life. People use it to write software, study documents, translate languages, generate media, support customer service, and conduct research. Businesses and governments are also exploring how it can improve services and productivity.
Behind this adoption is a large physical system. AI depends on computer chips, data centres, electricity, cooling, networks, engineers, and large amounts of capital. Global corporate AI investment reached an estimated $581.7 billion in 2025. U.S. private AI investment alone reached $285.9 billion.
Electricity demand is also rising. Global data-centre electricity use increased by 17 percent in 2025. Consumption by AI-focused data centres rose by 50 percent.
These figures do not prove that AI is a bubble. They show the scale of the wager being made.
Some believe AI will become a general-purpose technology, similar to electricity or the internet. Others question whether adoption, revenue, and productivity gains will grow quickly enough to justify current spending. This report does not choose either position in advance.
ANALYSIS The question is not whether AI matters, but which parts of today's build-out will still stand ten years from now.
“These figures do not prove that AI is a bubble. They show the scale of the wager being made.”
Chapter 00
Introduction
Every generational technology begins with the same question: is this real, or is it a mirage?
Dot-com era
- 1990sInternet expansion accelerates worldwide.
- Mar 2000Nasdaq peaks; dot-com correction follows.
- 2005–2020Internet becomes economic infrastructure.
AI era
- 2022Generative AI reaches consumers at scale.
- 2025–2026Historic AI infrastructure build-out.
- FutureLong-term outcome remains unknown.
History rhymes, but rarely repeats. The dot-com correction did not end the internet, it cleared the field for what followed. The current AI cycle is similar in energy but different in its physical footprint, its capital intensity, and the speed at which capability is compounding.
Chapter 02
The Real Cost of Artificial Intelligence
What holds this system up, and what does it consume?
The AI Infrastructure StackLayer 01
FACT Global data-centre electricity use grew 17 percent in 2025; AI-focused facilities grew 50 percent, according to the International Energy Agency.
Compute is not free, and neither is the grid it depends on. Water, land, permits, and transformers are becoming as strategic as chips.
- Renewable energySolar · Wind · Hydro
- Transmission gridHigh voltage
- SubstationStep-down
- Regional data centreFacility
- Cooling systemsThermal loop
- GPU clusterAccelerators
- InferenceModel runtime
- AI responseDelivered
Every prompt begins upstream, at the power plant. The chain compresses continental infrastructure into a few seconds of response.
FIG. 03 · DATA-CENTER ELECTRICITY DEMAND · 2020–2030
Global electricity consumption by data centers, historical and projected.
Flourish visualization placeholder
Chart URL to be added at publication.
Global electricity demand from data centres is projected to increase substantially through 2030 as AI workloads, cloud services, and digital infrastructure expand.
FIG. 04 · ELECTRICITY DEMAND GROWTH BY MAJOR SECTOR · 2025
Annual change in global electricity consumption across major end-use sectors.
Flourish visualization placeholder
Chart URL to be added at publication.
Electricity demand is rising across major sectors. Data centres and electrified transport are among the fastest-growing contributors, while buildings and industry still account for the largest absolute demand.
“The cost of building AI and the price of using it are moving in different directions.”
Chapter 04
Two Different Paths to AI Power
The United States and China are pursuing the same technology through very different systems.
FIG. 06 · AI INFRASTRUCTURE RACE — UNITED STATES VS CHINA
Two ecosystems building the same technology
The United States and China pursue AI leadership through different combinations of capital, industry, and state coordination — but the underlying inputs are the same.
United States
- Private capital
- Frontier AI models
- Cloud platforms
- NVIDIA / AMD ecosystem
- Frontier research
- Venture capital
China
- State-backed investment
- Manufacturing scale
- Huawei AI accelerators
- Energy expansion
- Industrial deployment
- Open-source AI ecosystem
Both require
- Electricity
- Semiconductors
- AI accelerators
- Data centres
- Networking
- Talent
- Capital
- Research
The race is not decided by a single model or benchmark. It runs on electricity, silicon, capital, and people — inputs both countries need in growing quantities.
FIG. 05 · GLOBAL AI INFRASTRUCTURE INVESTMENT · 2020–2030
Estimated annual investment in AI compute, data centers, networking, energy systems, and supporting infrastructure.
Flourish visualization placeholder
Chart URL to be added at publication.
Investment in AI infrastructure is expected to accelerate as governments, cloud providers, semiconductor manufacturers, utilities, and private investors expand computing capacity for AI training and inference.
Source: McKinsey & Company; Goldman Sachs Global Investment Research; International Energy Agency, Energy and AI, 2025.
Values shown after the latest reported period are estimates. The chart combines publicly available forecasts and research synthesis rather than a single standardized annual dataset.
FIG. 06 · LEADING HYPERSCALER INVESTMENT IN AI INFRASTRUCTURE · 2026E
Estimated capital expenditure supporting AI compute, cloud infrastructure, networking, and data-centre expansion.
Flourish visualization placeholder
Chart URL to be added at publication.
The largest cloud and technology companies are increasing capital expenditure as they expand AI computing capacity, data centres, networking, and related infrastructure.
Source: Amazon, Alphabet, Microsoft, Meta, and Oracle annual reports, earnings presentations, company guidance, and publicly available analyst estimates.
The figures are estimates and are not perfectly comparable. Companies report capital expenditure using different fiscal periods and definitions, and not all reported spending is exclusively attributable to artificial intelligence.
FIG. 07 · THE AI INFRASTRUCTURE VALUE CHAIN
From physical infrastructure and computing capacity to models, applications, and economic sectors.
Flourish visualization placeholder
Chart URL to be added at publication.
Artificial intelligence begins with physical infrastructure. Electricity, semiconductors, networking, cooling, and data centres support computing systems that enable models and applications across the economy.
Source: Oshen Research synthesis based on the International Energy Agency, Stanford AI Index, McKinsey & Company, and publicly available industry documentation.
This is a conceptual systems map. Link widths illustrate relationships within the AI infrastructure stack and must not be interpreted as measured financial, energy, material, or computing flows.
Chapter 05
Africa: Consumer, Supplier, or Builder?
Adoption alone does not translate into economic power. Ownership does.
The AI value chain
Critical minerals
Cobalt, copper, and rare earths that anchor the global hardware supply chain.
Four connected constraints
Compute
Limited domestic access to accelerators and hyperscale capacity.
Data
Fragmented, under-governed, and often not representative of local realities.
Talent
Pipelines exist but competition for engineers is global.
Governance
Frameworks vary widely across the continent's 54 countries.
Case studies · Across parts of the continent
Local-language systems
Speech and text models built for African languages.
Source · Global Center on AI Governance
Digital finance
Credit scoring, fraud detection, and payments infrastructure.
Source · Various operators
Public services
AI applied to health triage, education, and administration.
Source · Government & NGO deployments
Data labour
African workers supporting data-labelling and content moderation.
Source · Independent journalism
“Africa's long-term position will depend on ownership as much as adoption.”
Chapter 07
Key Takeaways
Six points to carry out of this report.
01
Every AI response depends on physical infrastructure, not software alone.
02
The cost of building AI and the price of using it are moving in different directions.
03
A transformative technology can exist alongside an overheated investment cycle.
04
The global AI race is becoming a competition between complete ecosystems.
05
Africa's long-term position will depend on ownership as much as adoption.
06
Public trust will depend on safety, accountability, transparency, and meaningful participation.
Three possible futures
Infrastructure boom
AI use grows quickly enough to justify today's spending.
Capital correction
AI remains useful, but expected revenue arrives too slowly to support every investment.
Transformation through correction
AI becomes essential, but the market first removes weaker companies and unrealistic expectations.
Sources
Source library
- Used in: Executive Summary, Introduction, The Machine, Bubble, Two Paths
- Used in: Executive Summary, The Real Cost
- Used in: Who Makes the Rules
- Used in: Who Makes the Rules
- Used in: Africa
- Journalism · 2024Bridging the Gap: AI's Potential in Africa's Data DesertAtlantic Re:think and GoogleUsed in: Africa
Cite & share
Hassan Mohamed,
The Infrastructure Age
From Prompt to Power: Understanding Artificial Intelligence, Infrastructure, and Global Competition.
Oshen Independent Research Report, 2026.