AI is often described as a software revolution. That is true in the way people experience it. They see chatbots, coding assistants, search tools, trading algorithms and medical models. But beneath the screen, AI is becoming something more physical, expensive and politically important. It is turning into an infrastructure race. Every major technology era has needed new rails. Railways needed steel and land. Cars needed roads and oil. The internet needed fibre and servers. Cloud computing needed hyperscale data centres. AI needs power grids, chips, memory, data centres, cooling systems, fibre networks, cybersecurity and patient capital, all at once. The next phase of AI may not be decided only by who builds the smartest model. It may be decided by who can secure electricity, build data centres, access advanced chips and finance projects that take years to complete. This is why the AI debate can feel incomplete. Much of the public conversation still asks whether AI stocks are overvalued. Some may be. But the deeper question is whether the world has enough physical capacity to support the demand now being created. In many places, the answer appears to be no. Computational power (compute) is becoming a strategic resource. Banks will need it for fraud detection and risk modelling. Hospitals may use it for diagnostics and drug discovery. Schools may use it for personalised tutoring. Governments will want it for security, public services and administration. Manufacturers will use it for robotics, simulation and supply-chain planning. That makes AI infrastructure a matter of national competitiveness. Countries with reliable power, faster permitting, chip access and data-centre capacity will attract investment. Countries without those advantages may become dependent on others for the basic infrastructure of intelligence. The bottleneck is not only demand. It is supply. Power grids are already under pressure in many regions. New data centres need land, water, cooling, transmission lines, substations and local approval. High-end chips remain difficult to secure. Building the AI economy is starting to look less like a typical tech cycle and more like a heavy industry buildout. That brings finance into the centre of the story. AI infrastructure is too large and long dated to be funded by governments and banks alone. Public budgets are stretched. Traditional banks are not always well suited to financing long-term assets with short-term liabilities. Large data centres and power projects require capital that can wait, absorb complexity and earn returns over many years. That creates a bigger role for pension funds, insurers, sovereign wealth funds, infrastructure managers and private credit. A data centre leased for 15 or 20 years to a major tech company may look less like a speculative technology bet and more like modern infrastructure. Yet the social consequences should not be ignored. If AI raises productivity but most of the gains flow to those who own data centres, chips, platforms, power assets and financial claims on infrastructure, wages alone may not capture the upside. That would deepen one of the central tensions of the modern economy: work creates value, but ownership often captures it. The same pressure may reshape companies. Large firms with capital, data and compute access could pull further ahead. Small firms may become more productive by using AI tools. But midsized businesses without scale, infrastructure access or a clear niche may find themselves squeezed. The AI boom is not only a market story. It is an energy story, an industrial story, a labour story and a capital markets story. The first phase of AI was about models and applications. The next phase is about the rails beneath them. Power is now AI policy. Data centres are now industrial strategy. Compute is becoming a scarce resource. Private capital is becoming part of the tech stack. The world is not simply adopting AI. It is rebuilding around it. The question is who pays for that rebuild, who owns it and who gets to share in the gains. • Muchena is founder of Proudly Associated and author of ‘Artificial Intelligence Applied’ and ‘Tokenized Trillions’.