As AI rack densities rise beyond the practical limits of air cooling, CETA System Co., Limited is positioning advisory-first operational intelligence as a way to connect liquid-cooling performance, energy discipline and outage-risk management across mixed-vendor data-centre environments.
The pressure is visible at Lake Mariner in Buffalo, New York, where a former coal plant has been converted for AI infrastructure. The campus targets 750 MW of power demand at full buildout and currently provides 500 MW of capacity across four data halls.
Its regional grid mix is described as approximately 89% zero-carbon electricity, while its cooling deployment includes direct-to-chip systems and rear-door heat exchangers integrated with existing power and building-management infrastructure.
The broader operating issue is simple: GPU workloads are concentrating more power into each cabinet than conventional air systems were designed to manage. Rack densities have moved from enterprise-standard 5 kW to 10 kW configurations towards deployments exceeding 30 kW to 100 kW.
Survey data cited in the sector shows rack density rising from 7 kW per rack to 16 kW per rack across the latest reporting window, with 79% of respondents expecting further increases due to AI and high-performance computing demand.
Individual GPU units can consume more than 300W, while fully populated GPU server racks can require 132 kW of power. Configurations expected over the near term could reach 240 kW per rack. Air-based cooling effectiveness deteriorates above 20 kW, while cooling and air-conditioning systems can account for up to 45% of total data-centre energy consumption.
For investors and operators, that translates directly into financial exposure. Concentrated thermal loads increase the risk of hot spots, server throttling and hardware failure. Downtime costs are cited at an average of $110,688.4 per hour, while high-density environments have reduced thermal ride-through time during power interruptions.
Lee Tsz-Hin chief executive officer of CETA System Co., Limited, starts with the economics: “Density changes the economics of resilience as much as the engineering.” He adds: “When thermal margins narrow, operators need better visibility before a fault becomes a capacity, availability or cost event.”
Lake Mariner illustrates how liquid cooling is being deployed to manage that shift. ChilledDoor heat exchangers, in-rack manifolds and coolant distribution units are being coordinated with existing alarm infrastructure to sustain higher thermal loads.
Direct-to-chip cooling reduces thermal resistance at processor and accelerator sources, supporting heat removal from GPU-dense systems under continuous high utilisation.
Brownfield conversion adds another layer of complexity. Older facilities designed around 5 kW to 10 kW racks must address floor loading, physical infrastructure and support requirements when moving towards configurations exceeding 50 kW per rack.
The attraction is that former industrial sites can offer transmission access, utility capacity, zoning and road infrastructure across large parcels. A proposed AI data centre at the former Tonawanda Coke site, for example, is described as a $2.2 billion, 500,000 sq ft project requiring approximately 300 MW of electricity, supported by a planned $55.3 million electrical substation.
CETA System frames the operational response around vendor-agnostic integration rather than wholesale hardware replacement. Monitoring platforms that connect building management systems, DCIM tools and plant controls can help operators correlate rack-level cooling, power consumption, server utilisation and equipment condition across mixed-vendor sites.
That integration is becoming more important as cooling becomes both an availability issue and a compliance issue. Data centres consume approximately 1.5% of global electricity, or 415 TWh annually, with projections indicating consumption could exceed 945 TWh by 2030 due to accelerated computing for AI.
Cooling infrastructure alone can represent up to 40% of total data-centre power consumption, making thermal optimisation a direct lever for cost control and disclosure readiness.
The advisory-first model is central to CETA System’s positioning. AI-generated setpoint recommendations and early-warning alerts remain subject to operator approval and hard constraints, preserving human oversight while giving teams earlier signals on cooling demand, energy use and asset-health deviations.
“Automation is most useful when it supports disciplined decision-making,” Lee says. “In critical infrastructure, recommendations must remain transparent, constrained and reviewable by the people responsible for availability.”
Operational intelligence platforms also support reporting discipline. Open protocols such as BACnet, Modbus, SNMP and application programming interfaces enable equipment integration, while data normalisation helps correlate units, naming conventions and timestamps.
Energy-impact reporting can quantify cooling savings in kilowatt-hours, cost and carbon metrics, while asset-health dashboards track downtime-risk reduction over time.
That matters as operators face stricter reporting requirements and more demanding customers. Mandatory energy-performance reporting applies to large data centres in Europe, while Germany’s efficiency rules set tightening Power Usage Effectiveness targets through 2030.
Sustainability reporting obligations have also expanded, increasing the value of auditable operational data. Manual compliance management raises workload and risk, particularly where mechanical and electrical teams lack rack-by-rack visibility into power and temperature.
For data-centre buyers, hyperscalers and pre-IPO investors, the strategic point is that cooling software is no longer a peripheral facility tool. It affects capacity planning, procurement discipline, resilience and the credibility of sustainability disclosures.
“The strongest platforms will be those that make engineering data usable for commercial decisions,” Lee adds. “Energy, uptime and compliance are now linked operating variables.”
CETA System argues that the sector’s move towards liquid cooling will require equally strong investment in operator-supervised intelligence, especially across brownfield sites and mixed-vendor estates where thermal stability, capital efficiency and reporting discipline must be managed together.
About CETA System
CETA System Co., Limited is a Hong Kong-incorporated technology company founded in 2017, delivering artificial-intelligence solutions for data-centre infrastructure.
Its vendor-agnostic platform combines HVAC and chiller-plant energy optimisation with predictive maintenance for critical assets including UPS systems, generators and chillers, integrating with existing building-management and DCIM environments through an advisory-first deployment model for colocation, enterprise and hyperscale operators across the Asia-Pacific region and beyond.
- Website: https://cetasystem.com
- Registered business: CETA System Co., Limited (Hong Kong BRN 67731517; CRN 2533166)
