A briefing for engineers, mechanical contractors, data center owners, operators, OEMs, EPC firms, utilities, and executive teams on the developments shaping AI infrastructure decisions.
Executive Summary
This cycle reinforced a structural shift in AI infrastructure: the industry is moving from a technology-first buildout toward an infrastructure-constrained operating model. Power availability, permitting, water strategy, equipment qualification, and commissioning readiness are increasingly determining where capacity can be built and how quickly it can be brought online.
The Five-Minute Version
Grid access is becoming conditional. PJM is proposing new tools to address a 6.8 GW capacity shortfall and may require greater visibility and emergency flexibility from very large loads.
Behind-the-meter power is moving from contingency planning toward a core siting strategy, illustrated by federal AI campuses pairing gigawatt-scale data centers with dedicated generation.
Permitting risk has escalated from local opposition to statewide policy. New York’s one-year moratorium on qualifying large data center permits is a clear signal that power, water, and community impacts can now stop projects at the state level.
Liquid-cooling infrastructure is maturing into an OEM-qualified ecosystem. LG’s 600 kW CDU receiving NVIDIA quality certification shows that component qualification and interoperability are becoming procurement issues, not just engineering preferences.
Commissioning is becoming part of AI readiness. The growing use of liquid-cooled load banks reflects a need to prove thermal and hydraulic performance before expensive IT hardware is installed.
Water strategy is becoming a design and public-policy requirement. The industry’s response is broadening beyond efficiency claims toward air cooling, recycled water, closed loops, and infrastructure investment.
1. Power Constraints Are Becoming an Operating Rule, Not Just a Development Delay
POWER DELIVERY · GRID RELIABILITY
On July 28, Reuters reported that PJM Interconnection plans to file proposals with federal regulators after its recent capacity auction left a 6.8 GW supply shortfall despite high prices. PJM projects roughly 70 GW of additional demand by 2038 from large new users including data centers. Its proposals include a reliability backstop procurement process and a registry for major loads. PJM also raised the possibility that non-self-supplied data centers could be curtailed during grid emergencies, subject to state cooperation.
WHAT THIS MEANS ON THE GROUND
Power availability can no longer be treated as a binary site-selection question. Owners should ask how firm the power actually is, what curtailment rights may exist, and whether the facility can operate flexibly during grid stress. Cooling design matters here because every megawatt avoided in heat rejection, pumping, and mechanical chilling preserves more of a constrained electrical allocation for compute.
Derived Insight: The competitive advantage is shifting from simply securing megawatts to designing AI campuses that can prove grid compatibility, load flexibility, and efficient thermal operation.
2. Dedicated Power Is Becoming Part of the Data Center Site Plan
POWER STRATEGY · SITING
On July 20, the U.S. Department of Energy’s National Nuclear Security Administration selected Amentum to enter negotiations for a proposed 1 GW AI data center with dedicated on-site energy generation at the Savannah River Site in South Carolina. By July 31, the Department of Energy was also advancing redevelopment of the former Paducah Gaseous Diffusion Plant in Kentucky around a proposed 1.8 GW data center campus supported by large-scale on-site generation and battery storage.
WHAT THIS MEANS ON THE GROUND
Behind-the-meter and co-located generation are moving from exceptional solutions toward a repeatable development model for projects that cannot wait for conventional grid expansion. Mechanical and cooling teams should expect site architecture to be influenced by the operating characteristics of the power source, including ramping, redundancy, islanding, and heat-rejection requirements.
Derived Insight: Power generation, data center design, and thermal infrastructure are converging into a single site development discipline.
3. Permitting Risk Has Moved Up to the State Level
SITING · REGULATORY RISK
During this reporting period, New York’s one-year moratorium on qualifying large data center permitting became a national reference point. The state is pausing discretionary environmental permits for large facilities while it develops standards addressing energy use, water consumption, and community impacts. Reuters reported on July 20 that governments and cities elsewhere are also considering restrictions as concerns grow around electricity prices, water supplies, land use, and local infrastructure.
WHAT THIS MEANS ON THE GROUND
Developers should assume that water and power documentation will increasingly be required earlier in the entitlement process. Cooling architecture can become part of the project’s license to operate when it demonstrates lower water draw, lower auxiliary energy demand, and a smaller local infrastructure burden.
Derived Insight: A cooling strategy is becoming part of the permitting narrative. Engineering teams that can quantify resource impacts before public review begins will be better positioned than teams that treat cooling as a late-stage mechanical decision.
4. The Liquid-Cooling Supply Chain Is Moving Toward Qualified, Integrated Systems
LIQUID COOLING · OEM QUALIFICATION
LG Electronics announced in late July that its 600 kW-class coolant distribution unit received NVIDIA final quality certification. The company is positioning the CDU alongside chillers and cold plates as part of a broader chip-to-chiller offering. The development is another sign that liquid cooling is moving from custom project engineering toward repeatable, qualified infrastructure platforms.
WHAT THIS MEANS ON THE GROUND
Coolant compatibility, CDU performance, cold-plate materials, filtration, controls, and fluid specifications need to be evaluated as a system. As OEM qualification expands, procurement teams should expect approved-component lists, warranty requirements, and interoperability criteria to exert more influence over fluid selection.
Derived Insight: The coolant specification ecosystem is becoming more important because fluid decisions increasingly sit inside an OEM-qualified chain rather than an isolated facility loop.
5. Commissioning Is Becoming a Formal Part of AI Readiness
COMMISSIONING · RELIABILITY
Industry guidance published in July highlighted liquid-cooled load banks as a way to validate CDUs, facility cooling loops, and rack-level infrastructure under realistic loads before critical IT equipment arrives. The approach reflects a broader operational change: liquid cooling introduces thermal and hydraulic commissioning requirements that cannot be fully proven with conventional electrical load testing alone.
WHAT THIS MEANS ON THE GROUND
Commissioning plans should define cleaning, flushing, fill-water quality, coolant concentration, air removal, flow balance, thermal validation, baseline fluid sampling, and acceptance criteria before startup. Owners should preserve baseline chemistry and performance data so later fluid analysis can be compared against a documented starting condition.
Derived Insight: In high-density AI facilities, commissioning quality is becoming a direct contributor to schedule certainty and long-term reliability.
6. Water Strategy Is Shifting From Efficiency Metric to Infrastructure Commitment
WATER STEWARDSHIP · COMMUNITY ACCEPTANCE
Water remained a central issue during the reporting period. Google outlined a broader approach that includes air or recycled-water cooling in water-scarce regions, local water-infrastructure investment, increased transparency, and alternative water sources. At the same time, the New York moratorium explicitly tied data center growth to concerns about both energy and water impacts.
WHAT THIS MEANS ON THE GROUND
Owners should quantify not only annual facility water use, but also peak seasonal demand, water source, treatment requirements, discharge pathways, and the effect of the chosen heat-rejection strategy. Closed-loop liquid cooling and dry heat rejection can materially change that conversation, but the design must still be evaluated against local climate, energy use, and operating economics.
Derived Insight: Water stewardship is becoming a commercial and permitting competency. The strongest projects will be able to explain where water comes from, how much is consumed, what is discharged, and how the design reduces local resource pressure.
| Power availability | HIGH CONSTRAINT | PJM capacity shortfall and rapid large load growth keep power at the top of the project risk stack. |
| Permitting / community acceptance | RISING RISK | State-level restrictions show that resource impacts can now delay projects before construction begins. |
| Water strategy | RISING IMPORTANCE | Water sourcing, cooling method, transparency, and local infrastructure investment are becoming part of project acceptance. |
| Liquid cooling adoption | ACCELERATING | OEM qualification of CDUs and integrated chip-to-chiller platforms points toward broader standardization. |
| Commissioning discipline | RISING IMPORTANCE | Thermal and hydraulic validation is moving earlier, before IT deployment. |
| Coolant specification influence | RISING | Interoperability, warranty, materials, and OEM qualification are increasing the importance of documented fluid requirements. |
What We’re Watching Next
PJM’s FERC filings and how large-load curtailment or self-supply rules evolve.
Additional state or municipal permitting frameworks modeled on New York’s approach.
Expansion of behind-the-meter generation and federal-site AI development.
Further OEM qualification of CDUs, cold plates, manifolds, and coolant requirements.
Commissioning standards for liquid-cooled AI halls, including load-bank validation and fluid baseline testing. Water-use disclosure requirements and commitments to closed-loop, dry, recycled-water, or non-potable cooling. Whether higher-temperature liquid loops continue moving from reference architecture into project specifications.
Sources & Further Reading
- Reuters, July 28, 2026, PJM plan to address swelling data center demand https://www.reuters.com/business/energy/largest-us-power-grid-moves-ahead-with-plan-meet-swelling-demand-2026- 07-28/
- U.S. Department of Energy / NNSA, July 20, 2026, Savannah River AI Data Center and Energy Project https://www.energy.gov/nnsa/articles/nnsa-selects-amentum-ai-data-center-and-energy-project-savannah-river-site
- Reuters factbox, July 20, 2026, authorities restricting data centers amid AI boom https://www.investing.com/news/stock-market-news/factboxauthorities-restricting-data-centres-amid-ai-boom-4800546
- Engineering News-Record, July 14, 2026, New York data center moratorium https://www.enr.com/articles/63315-new-york-orders-moratorium-on-data-centers
- Data Center Dynamics, July 8, 2026, liquid-cooled load banks and commissioning https://www.datacenterdynamics.com/en/whitepapers/simplifying-data-center-readiness-with-liquid-cooled-load-banks/
- Electronic Times, July 28, 2026, LG receives NVIDIA certification for 600 kW CDU https://en.etnews.com/20260728200003
- TechRadar Pro, July 27, 2026, Google water strategy for data centers https://www.techradar.com/ai-platforms-assistants/how-we-build-is-just-as-important-as-what-we-build-google-reveals how-its-going-to-solve-the-ai-data-center-water-problem-by-2030
- NVIDIA Blog, June 21, 2026, 113°F warm-water liquid cooling reference architecture https://blogs.nvidia.com/blog/liquid-cooling-ai-factories/
