AI needs power, water, and engineers.

AI can help predict a failing pump, improve a manufacturing process or choose a promising biological experiment. But the prediction still has to meet a physical system. A pump needs power. A factory needs materials, water and people who understand its operation. An experiment needs reliable measurements. The screen is only the visible end of a much larger engineering problem.

That connection was at the heart of my July 24, 2026 keynote for AIChE Day at Obafemi Awolowo University: Chemical Engineering at Two Frontiers. Africa’s physical infrastructure and its use of intelligent systems cannot be treated as separate conversations. In many settings, we have the opportunity—and the responsibility—to build them together.

The infrastructure is part of the intelligence

A data center is a useful example. We may describe it in terms of computing capacity or the services it supports. It is also an installation that needs electricity, cooling, communications and dependable operation. The surrounding grid, water systems and maintenance capacity matter to what that computing can actually deliver.

The same connection appears in water treatment, energy supply, fertilizer production and biotechnology. Software can help monitor or control a process. It cannot make an unreliable measurement trustworthy, remove a physical constraint or maintain equipment by itself. Those are engineering responsibilities, and adding AI makes understanding them more valuable.

There is no single African starting point. Countries and regions have different resources, infrastructure and industrial strengths. Some will develop large facilities; others will also need smaller, distributed systems. A design that works in one setting may need substantial adaptation in another. The useful question is what a particular community or industry needs, and what can be built and sustained there.

Keep the engineering foundations close

This is why I would encourage a student interested in AI to take the foundations of chemical engineering seriously. Accounting for matter and energy, understanding reactions, moving heat and fluids, separating materials and controlling a changing process remain central to the work.

A model can produce a convincing answer without establishing that the answer obeys those constraints. We need people who can ask what is being transformed, how fast it happens, what energy is required and whether the proposed operation is safe. Learning computational tools should deepen those questions and help us investigate them.

A digital twin brings the connection into focus. It combines a representation of a real process with information about that process. Physical models describe the mechanisms we understand. Methods that learn from data can help describe more complicated behavior. Using both gives us more ways to examine a prediction before relying on it.

Neither approach removes the need for judgment. A learned pattern may become unreliable under conditions the model has not seen. A physical model may simplify something important or depend on uncertain measurements. We should know what each contributes, where it is limited and what evidence would make us revise it.

Build the ability to adapt and maintain

The connection also matters in biological engineering. An enzyme prediction needs context about the reaction and the conditions being measured. A model of a metabolic network has to account for what a cell can physically do. Patterns in biological data become more useful when we can relate them to mechanisms and test the explanation.

For students, the practical direction is to develop depth in a physical or biological system alongside the ability to model, measure and automate it. Understand the process well enough to notice when an attractive output makes little sense. Learn enough computation to investigate the problem, communicate the evidence and make the work usable.

The wider goal is productive capability: people and institutions that can build, adapt, govern and maintain the systems they depend on. Adopting an advanced tool is one step. Developing the knowledge to assess it, improve it and keep it working is a much longer project.

That is the opportunity I wanted the keynote to leave with students. Bring physical understanding and computation into the same room. Start with a real need. Work through the constraints. Build something whose usefulness can survive beyond the demonstration.

Read the full keynote slides: Chemical Engineering at Two Frontiers.