The Ai Power & Cooling Conference - March 2026

The team at Strategy Engineering Research Group would like to thank the speakers, supporters and attendees for making the recent AI Data Centers: Power and Cooling Conference such a valuable learning and networking opportunity. 

One of the clearest messages that came through during the conference was that water and power cannot be considered separately when planning large-scale AI data centre infrastructure.

For the first time at an AI data center conference, we had water and power discussed together at a strategic level, particularly in terms of optimal site selection, planning and resource adequacy.

What became clear is that many organisations now seem to have a plan for selecting technologies and scaling infrastructure — but what can still trip up multi-billion-pound projects is often something else entirely: ensuring resource adequacy, navigating regulatory realities, and staying ahead of community relations and local concerns. 

Several speakers brought real clarity to these issues.

Commissioner Catarina Gonzales from the Texas Commission on Environmental Quality gave an extremely well-received presentation on the regulatory perspective. She also highlighted the very real opportunities associated with utilising produced water for cooling, which sparked a great deal of discussion. 

We also heard from Commissioner David LaCarte of the Federal Energy Regulatory Commission, who discussed the realities of interconnection reform and behind-the-meter generation, providing strategic-level clarity on what the future may look like — but also on what still needs to happen for that future to actually materialise. 

Ning Ling, Chief Economist at the Bureau of Economic Geology in Texas, and Temple McKinnon, Director of the Texas Water Development Board, both helped bring real clarity to the topics of water availability, community relations and the broader issues that are at stake when planning large-scale infrastructure in water-constrained regions. 

On the power side, we heard a succession of very powerful presentations on how to plan for data center load growth.

Gordon Drake, Director of Market Design and Analysis at ERCOT, and John Marra, Director of Reliability Assurance at NERC, shared valuable perspectives on interconnection challenges, but also spoke in depth about what really matters from the perspective of grid operators. 

There was also a thorough evaluation of behind-the-meter power projects and the optimal configurations that can work well for specific sites.

For those involved in long-term strategic planning, this kind of discussion was extremely valuable. 

Cooling Technologies — Opportunities and Realities

The whole morning of Day Two was dedicated to the different emerging cooling technologies.

We heard about the challenges associated with liquid cooling, air cooling and immersion cooling, as well as some of the cutting-edge developments that are genuinely starting to make a difference in the sector. 

What was particularly valuable was that we did not simply talk about the technologies themselves, but about the pros and cons of the different cooling approaches and their commercial readiness.

There was also an important conversation about future-proofing.

Multi-billion-pound investments are now being made in specific cooling technologies, yet the pace of technological change is extremely fast. That creates a real challenge: how do organisations invest with confidence when the technology landscape is still evolving so quickly? 

We heard about the latest modular solutions and rack architectures that can help both hyperscalers and AI-orientated data centres future-proof their cooling strategies.

At the same time, speakers were very clear that there is no single silver-bullet solution.

What works at one site may not work at another.

It really depends on the local environment, infrastructure, water availability, power availability and operational requirements.

Developments in air cooling, liquid cooling and immersion cooling are constantly shifting the landscape, which made these updates particularly valuable for attendees. 

What Came Out of the Conference

All in all, the chair — Steve Thomas, Head of Infrastructure Resilience Initiatives at Strategy Engineering Research Group — summarised that the conference generated over 200 valuable takeaways, many of which will be included in the post-conference report and presentation/video package that is available to purchase on this website. 

If you would like to receive a free copy of the short "headline report", which is shorter in scope but captures many of the key insights, please email us, and we will make sure that you receive a copy. Alternatively, a shorter summary is available on the website — accessible via the top navigation bar.

A Final Thought

The good news is that many of the solutions are already out there.

Yes, the technologies are evolving. The chip architectures and cooling approaches will change and improve.

But what is really important - now - are the site planning and community engagement factors — the regulatory alignment, the community engagement, and the practical realities of delivering these projects in the real world. 

Once again, thank you to everyone who attended and helped make this such an enjoyable conference to run.

After a rescheduling due to the late January Texas and multi-state winter storm, it was a privilege to chair the AI Data Centers Power and Cooling Conference 2026 in Dallas.

The initiative was designed to bring together senior-level decision-makers from across the full ecosystem — including grid operators, regulators, water planners, cooling technology providers, data centre operators, and technologists — to address the real bottlenecks slowing AI infrastructure build-out.

Back in mid 2025, when we initially researched this conference, infrastructure constraints — including grid connection delays, water permitting challenges, and regulatory approvals — were already slowing growth, not only in Texas but also in states such as Virginia, Arizona, and California.

Since then, the conversation has evolved further, with state-level moratoriums now being proposed in certain regions.

A central theme of the discussion was that AI data centres must fundamentally rethink how they are powered. Grid-only solutions are unlikely to support projected demand. At the same time, identifying cooling and water management strategies that can scale under real-world constraints — across different states and site-specific conditions — is becoming increasingly critical.

This conference was never intended to solve every challenge facing the industry. However, speakers consistently fed back that it created a pioneering, cross-sector conversation that is helping to define the next phase of infrastructure delivery.

Regulatory Reality: Growth Within Constraints

From a regulatory enforcement perspective, Commissioner Katrina Gonzalez of the Texas Commission on Environmental Quality, alongside a strategic-level discussion with David Alcarte of the Federal Energy Regulatory Commission, provided essential clarity.

At both state and federal levels, the message was consistent: while regulatory predictability matters, AI data centre growth must operate within a framework of:

• Environmental compliance
• Statutory discipline
• Public legitimacy

Texas, for example,  may be “open for business,” but environmental compliance still applies in full. Projects must be professionally planned, rigorously justified, and carefully executed.

Economic development and environmental responsibility are not competing priorities — they are obligations that must be managed together.

A key takeaway is that developers and their advisors must engage individuals with deep, jurisdiction-specific expertise. A lack of local regulatory understanding can materially delay projects.

Equally important, community engagement should never be treated as “soft PR.” It is, in reality, schedule protection — a core component of the compliance strategy.

A strong case study illustrated this: a developer resolved a dispute by returning to the community, understanding local priorities (in this case, wetlands), adapting the project, and ultimately avoiding a prolonged delay.

The broader lesson is clear:

Local concerns are often more negotiable than assumed — but only if engagement happens early. The cost of not engaging is often significantly higher than the cost of adapting.

And perhaps the simplest, most powerful takeaway:

Communities never want to be surprised by a project appearing next door.

Water: From Constraint to Strategic Variable

From both a regulatory and public perception standpoint, water is emerging as one of the defining issues in data centre growth.

Commissioner Gonzalez highlighted that water sits at the heart of TCEQ’s mission — balancing economic development with water conservation.

Water concerns are now front and centre in public and political discourse, particularly in Texas, where community sensitivity around water usage is intensifying.

One of the most interesting developments is the increasing focus on produced water from the oil and gas sector.

Oil and gas operators face a growing disposal challenge, and through legislative and regulatory momentum, a framework is beginning to emerge in which treated produced water could become a viable industrial resource. However, significant questions remain—particularly around the cost and logistics of delivering this water to end users.

This presents a potential opportunity for data centres, particularly for non-potable applications.

However, in water-stressed or politically sensitive regions, developers must prioritise early engagement on water strategy, as misunderstandings can quickly harden into opposition.

Integrated System Thinking: Power, Water, and Trade-Offs

From a system modelling standpoint, Dr Ning Lin, Chief Economist at the Bureau of Economic Geology (University of Texas at Austin), reframed the challenge entirely.

AI data centre deployment is no longer a single-variable problem centred on power.

It is now a system-level optimisation challenge across:

• Power
• Water
• Land
• Infrastructure
• Community dynamics
• Policy

His team is developing advanced tools to model and optimise across these constraints, recognising the interdependencies rather than treating variables in isolation.

A critical insight is that water is not simply a constraint — it is part of a trade-off system.

Reducing water usage often increases power consumption. “Zero water” solutions are not always optimal.

Water must therefore be assessed across:

• Direct usage (on-site cooling)
• Indirect usage (power generation)

Two otherwise identical data centres can have materially different total water footprints depending on their design and sourcing strategies.

On community dynamics, Dr Lin clarified that opposition is not uniform:

• Some concerns stem from knowledge gaps
• Others reflect fundamental “not in my backyard” preferences

This reinforces that site selection remains critical, and data alone cannot resolve every conflict.

A further takeaway for future discussion is that water is not standardised — different stakeholders define and measure it differently, adding complexity to decision-making.

Planning Reality: Why Water Forecasting Is Behind the Curve

From a planning and forecasting perspective, Temple McKinnon of the Texas Water Development Board provided a critical reality check.

Texas does have a structured water planning system — but it was not designed for the speed, opacity, and scale of data centre demand we are now seeing.

The state water plan is produced every five years, based on regional inputs and long-term projections. However, this system has an inherent lag.

Current plans are based on:

• 2020 data
• Projections developed in 2023

Data centre growth has accelerated significantly since that window.

As a result:

• Data centres are not yet fully integrated into planning models
• Forecasting is constrained by data gaps and limited visibility

Additionally, planning operates at aggregated levels, not site-specific detail — meaning it cannot accurately model individual projects.

The system is catching up, but not fast enough.

For developers, this creates several important implications:

• Do not over-rely on public planning data
• Assess both direct and indirect water impacts  early
• Engage with multiple agencies, not just regulators
• Expect uncertainty in water availability projections

Planning tools should therefore be treated as directional inputs, not definitive answers.

Grid Reliability: From Assumption to Risk Management

From a system reliability standpoint,  John N. Moura  of the North American Electric Reliability Corporation reframed how risk should be understood.

The system is entering unfamiliar and riskier territory:

• Demand is accelerating faster than anticipated
• The generation mix is becoming less predictable
• Traditional reliability metrics are becoming less meaningful

Historically, NERC risk maps were largely low-risk. Today, they show widespread high-risk zones — including Texas.

His analogy was clear:

A cardiologist does not predict the exact timing of a heart attack — but identifies risk factors.

Similarly, NERC is not predicting outages — it is identifying rising system risk.

The implications are significant:

• Reliability can no longer be assumed
• Demand flexibility is becoming essential
• Data centres are both a demand driver and a potential system risk

Market Design & ERCOT Reality: Building in Real Time

From a market design and grid operations perspective, Gordon Drake of ERCOT brought the discussion into immediate operational reality.

ERCOT is effectively redesigning its system in real time to accommodate unprecedented demand growth driven by data centres.

This creates a fundamental tension:

• Load is arriving faster than infrastructure can  be built
• The generation mix is shifting toward renewables
• Firm capacity is less available at critical times

As a result:

 net load — not peak demand — is now the key operational metric.

For developers, this has material implications.

Connection is no longer just about securing capacity. It increasingly requires demonstrating:

• Flexibility
• Responsiveness
• Alignment with system needs

In practice:

• Demand response is becoming essential
• Load flexibility is increasingly a prerequisite for connection
• Market participation is becoming part of infrastructure strategy

The underlying message is clear:

You are not connecting to a static grid — you are connecting to a system that is being actively redesigned around you.

What Actually Matters Now

Stepping back, five consistent themes emerged:

1. AI data centre growth is no longer constrained by technology — it is constrained by infrastructure, approvals, and coordination.

2. Power and water decisions are fundamentally interdependent.

3. Site selection is now a system-level optimisation challenge, not a single-variable decision.

4. Community engagement is a critical delivery function, not a communications exercise.

5. Planning systems and public datasets are lagging reality.

Put simply:

The industry does not have a technology problem.
It has an execution problem under constraint.

What This Means in Practice

For developers, operators, and investors, the implications are clear:

• Treat grid connection as a negotiated capability, not a guaranteed right
• Define water strategy before site announcement, not after
• Build community engagement into the project timeline, not as a reaction
• Model total resource footprint, not just design efficiency
• Invest in flexibility — both technically and commercially
• Engage across multiple stakeholders, not just regulators
• 
  

This is not simply about building data centres.

It is about building infrastructure that can operate within — and adapt to — constrained, interdependent systems.

Closing Reflection

This was a deliberately focused conference that shifted the tone of the conversation:

Away from:

• Pure growth ambition
• Technology-led optimism

And towards:

• Real-world constraints
• Delivery complexity
• System-level thinking

Speakers consistently fed back that this was one of the first forums where:

• Power and water were discussed together at a strategic level
• Regulatory, technical, and community perspectives were aligned
• Trade-offs were openly acknowledged

And that, in itself, represents progress.

Because the next phase of AI data centre growth will not be defined by: 

who can build the fastest.

But by:

who can deliver reliably, responsibly, and at scale within real-world constraints.

Steve D Thomas, Chair, SERG

The full Chair’s Report — including detailed analysis of every presentation, full Q&A coverage, all presentation materials, and the complete video — is available for purchase and download on this site.