At the IMF and World Bank annual meetings in Bangkok, IMF chief Kristalina Georgieva warned on October 16 that AI’s risks could exceed its opportunities without safeguards and an ethical framework. Reuters also reported concerns about the electricity and digital infrastructure required for broad adoption, the possibility of job displacement and the financial risk if heavy AI investment fails to deliver expected productivity gains. These issues connect model development to real-world costs.
Why computing needs a physical footprint
Training and serving AI models depends on chips, data centres, networking, cooling and reliable electricity. A model can feel like a piece of software to a user, but its service depends on physical infrastructure and long-lived capital investment. The cost profile varies widely by model and workload, so claims about AI’s total energy demand should be grounded in measurements and scenarios rather than a single dramatic figure.
The return-on-investment question
Companies are investing because they expect AI to automate tasks, improve products and raise productivity. Those benefits are not evenly distributed, and pilots do not always translate into sustained results. If an organisation builds capacity ahead of demand, it may face underused infrastructure. If adoption succeeds, it still needs to account for ongoing compute, electricity, data governance, security and human review in its cost model.
Access and distribution matter
Infrastructure is not equally available everywhere. Reliable power, network connectivity, skilled staff and affordable access influence which businesses and communities can benefit. The IMF discussion highlighted these prerequisites alongside regulation. Policymakers and companies therefore need to consider not only model performance but also access, energy planning, workforce transitions and whether productivity benefits flow beyond the largest buyers.
A more practical governance agenda
Good governance makes technical promises testable. Organisations can document what a system is used for, assess high-impact workflows, evaluate security and privacy risks, monitor outcomes after release and assign people who can intervene. At the infrastructure level, transparent reporting about resource use and realistic capacity planning help avoid assuming that every new capability will automatically justify every cost.
What to watch next
The useful question is not whether AI will be transformative; it is where it creates durable value and under what constraints. Look for measured productivity improvements, evidence about operational costs, grid and cooling plans, and stronger ways to test systems before deployment. AI investment decisions are more credible when the business case includes electricity, security, governance and possible failure—not just headline model capabilities.
Sources and further reading
The links below provide the official listing or reporting used for this guide. Release times and service availability can change; check the source for the latest details.
