Data Center Signals From Around The Web
We read dozens of data center news sources every day, so you don’t have to. Curated highlights for engineers and operators who need to stay ahead of what’s coming.
Uptime Institute sees limited role for liquid cooling in AI
Uptime Institute emphasizes that liquid cooling remains niche in AI data centers, suited only for ultra-high-density setups. This insight helps engineers evaluate cooling options as AI compute demands intensify without overinvesting in specialized systems.
- Liquid cooling targets racks over 100 kW, while most AI workloads stay under 50 kW with air cooling
- Adoption rate in hyperscale facilities is below 5% for AI-specific deployments
- Implementation costs for liquid cooling range from $500,000 to $1 million per rack row
- Air cooling handles most setups under 50 kW effectively
Energy storage’s role in stabilizing AI data center power
Roland Berger’s report highlights how energy storage systems address power volatility in AI data centers, which is essential as AI workloads cause rapid demand spikes that strain grids and increase costs.
- AI training can cause power demand to fluctuate by 20-50% in seconds, risking outages
- Battery systems provide 50-200 MW of backup power for 15-60 minutes during peaks
- Roland Berger projects storage integration could reduce energy costs by 10-25% annually
- California and Texas lead with over 5 GW of data center storage deployments planned by 2025
- Lithium-ion batteries dominate, with 80% market share, followed by flow batteries for longer durations
- Report recommends hybrid setups combining storage with renewables for 99.99% uptime
Meta speeds up fiber optic technician training to ease data center delays
Meta is accelerating training programs for fiber optic technicians to address a critical shortage that’s delaying data center expansions. This initiative highlights the growing need for specialized skills in high-speed networking infrastructure amid surging AI demands.
- Program targets training 500 technicians over the next 12 months through partnerships with community colleges in California and Virginia
- Shortage of fiber splicers has delayed Meta’s data center projects by up to six months in key regions
- Training focuses on fusion splicing, OTDR testing, and high-density fiber management for 400G networks
- Meta allocates $10 million for the initiative, including equipment and certification costs
- Effort aims to support deployment of 100,000 kilometers of new fiber optic cable in 2024
- Similar bottlenecks affect competitors like Google and Amazon in Nevada and Oregon data center hubs
Scaling liquid cooling challenges in data center design
In this Datacenter Dynamics interview, host James Radings speaks with Rajat Bagghdad, senior mechanical engineer at Arcadis, about current challenges in data center cooling systems. Data center engineers should watch to gain insights into deploying liquid cooling at scale and addressing operational hurdles in high-density environments.
- Liquid cooling technology is well-known, but the main challenge lies in its industrial-scale deployment for hyperscalers across multiple data centers.
- Hyperscalers need vendors and OEMs to supply the necessary liquid cooling infrastructure to meet deployment demands.
- Operators trained in air cooling must quickly learn liquid cooling while maintaining high uptime requirements.
- Design challenges include understanding interactions between servers and cooling systems in high-density AI factories where loads fluctuate rapidly.
- GPU workloads can spike every couple of seconds or microseconds, but cooling systems are not reactive enough to handle such variations.
- Engineers focus on stabilizing systems to provide continuous flow of cooling or chilled water, as chillers are not suited for rapid changes.
Federal agencies prepare regulations on data center energy use
Federal agencies are gearing up to impose new regulations on data centers to curb their escalating energy demands and environmental footprint driven by AI growth. This development matters for data center engineers as it may reshape facility designs, operational efficiencies, and compliance requirements nationwide.
- EPA plans to establish efficiency standards targeting data centers power consumption
- Data centers account for 2-3% of US electricity use, expected to reach 8% by 2030
- DOE is collaborating on rules for carbon emissions, water usage, and grid reliability
- Regulations could mandate energy reporting for facilities over 100 MW starting in 2025
- Key states impacted include Virginia, Texas, and California with high data center concentrations
- Industry input from groups like the Information Technology Industry Council is being sought
Electric demand surge from data centers in Virginia and Texas
A recent report analyzes the explosive growth of data centers in Virginia and Texas, detailing their outsized effects on regional power grids. Understanding these dynamics helps engineers plan for scalable, reliable electricity infrastructure amid rising AI and cloud demands.
- Virginia data centers consumed 5,000 MW in 2023, representing 25% of Dominion Energy’s total load
- Texas saw data center capacity reach 2,500 MW by mid-2024, driven by expansions in Dallas and San Antonio
- Projections indicate Virginia’s data center power needs could hit 10,000 MW by 2030
- ERCOT forecasts 15 GW of additional load growth in Texas from data centers over the next decade
- Both states face transmission upgrades costing billions, including $5 billion in Virginia projects
- Report highlights policy responses, such as Texas incentives for on-site generation
Get Data Center Engineering News In Your Inbox:
Popular Posts:
Share Your Data Center Engineering News
Do you have a new product announcement, webinar, whitepaper, or article topic?
Get Data Center Engineering News In Your Inbox:
More Signals
Indiana’s Marshall County enacts immediate permanent data center ban
Marshall County in Indiana has approved a permanent ban on data centers, effective immediately, potentially shifting development focus to more permissive regions. This move highlights escalating local regulatory hurdles for data center expansions amid concerns about infrastructure strain.
- Marshall County Commissioners approved the ban in a unanimous vote during their latest meeting
- Ban prohibits all new data center constructions and expansions within county limits
- Existing facilities remain operational but face restrictions on modifications or upgrades
- Decision driven by local worries over high water consumption, energy demands, and noise pollution
- Effective immediately, with no grace period for pending project applications
- Similar restrictions under discussion in adjacent counties like Kosciusko and Fulton
AI data center project with Trump branding hits delays and CEO exit
A high-profile AI data center megaproject associated with the Trump brand has stalled, leading to the departure of its CEO. This setback underscores the challenges in executing ambitious data center developments amid shifting leadership and market conditions.
- Project involves plans for a 500 MW facility in Tennessee focused on AI workloads
- CEO Michael Evans departed after 18 months in the role
- Development halted due to regulatory hurdles and funding shortfalls totaling $2.5 billion
- Partners include tech firms Oracle and Microsoft for infrastructure support
- Original timeline targeted completion by 2025, now delayed indefinitely
- Site spans 1,200 acres with proposed liquid cooling systems
PJM power market outlook addresses data center load growth challenges
In the Switched On podcast by Bloomberg NEF, host Tom Rowland-Rees interviews analysts Lara Camman and David Mohamadi about the PJM Power Market Outlook 2026 report. They discuss how data centers are driving rapid load growth in the largest US power market, pressuring supply and capacity, which data center engineers should watch to understand grid evolution and demand management strategies.
- PJM enters a period of rapid load growth driven largely by data centers after years of steady demand
- Load surge reshapes the system, pressures supply, and pushes capacity prices to record highs
- Options to meet demand include building new supply faster or managing demand, with balance complicated by competing interests
- Modeling PJM combines challenges from other markets, such as large load growth like ERCOT, carbon markets like California, gas-to-coal switching like MISO, and behind-the-meter solar plus offshore wind like New York ISO
- Data center load growth concentrates in PJM South, requiring regional placement and handling considerations in scenarios
Ansys digital twin predicts data center hotspots with AI and CFD
CADFEM APAC’s YouTube video presents a digital twin approach for data center cooling using Ansys simulation tools, demonstrating how it integrates physics, AI, and real-time data for predictive thermal management. Data center engineers should watch to understand how this method enables proactive cooling adjustments and energy savings.
- Digital twin integrates physics-based simulation, AI-driven prediction, and real-time sensor data to forecast temperature behavior.
- CFD model built with Ansys Fluent simulates server heat generation and fan-driven airflow under varying conditions to generate training data.
- Temporal fusion transformer model uses historical data, heat loads, fan speeds, and calibrated states for time series temperature forecasting.
- Moving horizon estimation combines sensor data with model predictions to maintain digital twin accuracy despite noisy or limited inputs.
- System predicts future temperatures and identifies potential hotspots, allowing preventive actions.
- Applications include dynamic fan speed adjustments, energy consumption reduction, anomaly detection, and improved operational planning.
Oracle partners with Bloom Energy for 2.8 GW fuel cell rollout in AI infrastructure
- Bloom Energy will supply solid oxide fuel cells capable of 2.8 GW total output
- Oracle plans integration across multiple U.S. data center sites by 2026
- Project focuses on AI workloads, supporting up to 500,000 GPUs
- Fuel cells operate on natural gas or hydrogen, with efficiency ratings above 60%
- Investment estimated at $5 billion, funded through joint financing
- Deployment includes microgrid setups for redundancy in California and Texas
Maine becomes first state to pass a data center construction ban
Maine's legislature passed LD 307, making it the first state to enact a moratorium on data center construction. The ban targets facilities drawing 20 MW or more and runs until November 2027.
- The House passed it 82-62; the Senate followed shortly after
- A new Data Center Coordination Council will study grid impact, water use, and economic tradeoffs before the moratorium expires
- Governor Janet Mills has indicated support with minor modifications
- Projects in Jay (a former paper mill site), Sanford, and Loring would be halted
- Maine already has some of America's priciest electricity. Lawmakers cited concerns that AI-driven demand would push rates even higher
- At least 12 other states have filed moratorium bills this cycle. Maine is the first to actually pass one
PIMCO in talks for $14B debt deal to finance Oracle’s Michigan hyperscale campus
Pacific Investment Management Co. is in early talks with Bank of America to provide roughly $14 billion in debt financing for Oracle's data center in Saline Township, Michigan. The campus is tied to OpenAI workloads.
- Deal would be structured in 144A format, with bonds sold privately to qualified institutional buyers
- Blackstone is expected to contribute an additional $2 billion in equity
- Related Digital is leading the project development
- PIMCO has form here: it held $18 billion of debt in Meta's Hyperion data center financing and recorded a $2 billion paper gain as valuations rose
- Oracle has $38 billion in debt tied to data center projects in Texas and Wisconsin, plus another $18 billion for a New Mexico facility
Four data-backed reasons the grid can probably handle data centers
ITIF researcher Robin Gaster pushes back on the prevailing grid-panic narrative with a data-heavy argument that AI data centers won't overwhelm electricity supply.
- Only one-third of announced capacity is actually being built. Of 240 GW announced, capital constraints, permitting delays (6-18 months), and construction timelines (20-54 months) dramatically slow deployment. New deals fell 40%+ between Q3 and Q4 2025
- Data centers can shift load off-peak. Peak demand occurs only a few hours per year, and strategic load shifting could improve grid utilization
- Behind-the-meter generation is reducing grid dependence. Hyperscalers are building their own power and using the grid only for backup
- New contracts already protect ratepayers. Large load tariffs include 15-year minimums, 85% load guarantees, ramp-up schedules, exit fees, and hold-harmless clauses
- Real electricity prices are 6% lower than in 2008. The grid has absorbed bigger demand shocks before