Much of the conversation around artificial intelligence understandably focuses on models, compute power, and software innovation. Yet after participating in discussions at the Rome Conference on AI, Ethics & Governance, I came away with a growing conviction: the next chapter of AI is as much an infrastructure story as it is a technology – and, increasingly, a societal one.
As AI adoption accelerates, we are beginning to confront questions that extend beyond computing power alone. How will we generate the energy needed to support AI at scale? How will we manage increasingly concentrated heat loads? How will we use water responsibly? How will communities experience the infrastructure being built around them?
These are not secondary considerations. They are becoming core design questions.
A Moment of Choice
We have seen challenges like this before.
Over the past several decades, industries around the world have worked together to transform refrigeration systems—improving performance while significantly reducing environmental impact. That progress did not happen because of a single breakthrough. It required collaboration across manufacturers, policymakers, equipment providers, and technology developers. It required the willingness to adopt new technologies and rethink established systems.
I believe AI infrastructure is approaching a similar moment.
As chip architectures become more powerful and workloads more demanding, energy density continues to increase. That translates directly into higher heat loads, greater cooling requirements, and increasing pressure on energy and water systems.
We have reached a choice point. We can continue to incrementally adapt legacy systems, or we can rethink how the infrastructure supporting AI is designed.
Cooling is one clear example. Air cooling has served the industry well, but for the performance levels AI now requires, air alone is reaching its limits in data centers. Many operators are moving toward liquid cooling, including single-phase direct-to-chip approaches. But as rack densities continue to rise, we will need to keep moving toward more advanced solutions, including two-phase liquid cooling.
In many cases, the barrier is no longer technical feasibility. It is risk tolerance and inertia in existing infrastructure designs. Delaying adoption may feel lower risk in the short term, but it can also increase cost and complexity over time as infrastructure struggles to keep up.
That transition matters because the decision to adopt new cooling technologies can do more than support performance. It can help reduce total cost of ownership, lower energy requirements, and dramatically reduce water use.
Those outcomes matter beyond the walls of a data center. AI’s impact is experienced through electricity demand, water use, siting decisions, and the relationship between infrastructure and the communities that host it.
As AI infrastructure grows, there is also a growing expectation that the entire value chain—from technology to materials—operates responsibly. Trust will depend on performance, transparency, and responsibility advancing together.
Efficiency May Be the Most Underrated Innovation Opportunity
I suspect one of the largest opportunities over the next decade may lie in efficiency.
The opportunity is significant. Better workload allocation, smarter cooling, eliminating wasted compute, and making the most out of every electron of energy can drive meaningful efficiency gains. The challenge, however, is not technology. In many cases, the challenge is adoption.
We need to think beyond incremental improvements. The biggest gains often come when industries are willing to leapfrog to new approaches rather than continuously optimizing legacy systems.
In cooling, for example, advanced two-phase liquid cooling technologies offer opportunities to reduce energy consumption, nearly eliminate water use, reduce noise, and support significantly higher compute densities than were previously possible.
The question is not whether these technologies will play a larger role in the future, but how quickly we can scale when we know rack densities will start to exceed 500kW within 2027.
Performance and speed are often prioritized over efficiency, even when more efficient solutions exist. That is why this ultimately becomes a leadership and systems question. Do we design for performance and efficiency, so they are inseparable, or do we continue to treat them as competing priorities?
Why Policy Matters
This is also why policy discussions are increasingly important.
We know it is difficult to manage what we do not measure, and today we are not measuring the full system. Energy use is only part of the picture. Water use, land use, and community impact must also be considered if we want infrastructure that is both high-performing and resource-efficient.
The next step is better disclosure and a broader set of metrics that reflect real-world resource use and community impact. Policy can play an important role by setting clear expectations, so efficiency and transparency are built in from the start. That helps us move from a partial picture to a more complete one.
Smart policy can help establish clear expectations around efficiency, resource use, transparency, and community benefit in ways that encourage innovation while delivering better outcomes. We are already starting to see this happen. In Singapore, policymakers reopened data center development with stricter efficiency requirements, while Europe is expanding reporting requirements beyond energy consumption alone to include broader resource-use metrics.
In the United States, similar momentum is emerging through bipartisan liquid cooling legislation introduced in both the House and Senate, signaling growing recognition of the need for more efficient, next generation data center infrastructure.
In major technology transitions, progress depends on more than technical readiness. It also depends on the ability of policy, industry, and equipment manufacturers to move together. That was true in the evolution of refrigerants, and I believe it will be true as AI infrastructure evolves.
Building What Comes Next
We have seen industries navigate transitions like this before. The organizations and countries that lead are rarely the ones that wait for perfect conditions. They are the ones willing to embrace innovation, align incentives, and build for what comes next.
The decisions we make today will help determine whether AI develops on infrastructure that is merely larger—or infrastructure that is smarter, more efficient, more resilient, and better aligned with the communities it serves.
This is an opportunity worth getting right. And it will take leadership to get there.
Denise Dignam is the President and Chief Executive Office of The Chemours Company, a global chemistry company with a vision to deliver Trusted Chemistry that makes people’s lives better and helps communities thrive.
