DEV Community: Da

Zerra’s 1.44 GW Western Downs Plan Shows How AI Data Centers Are Becoming Energy Projects

Da — Mon, 24 Aug 2026 18:58:21 +0000

Zerra DC has filed plans for a very large data center campus in Queensland’s Western Downs region. The August 24 market source describes the proposed Western Downs Digital Park as a 1.44 GW development on a 725.5 hectare site around 37 kilometers northwest of Dalby. Data Center Dynamics reported on August 24, 2026 that the project could be delivered across four phases and sits close to a major substation and several gas and renewable generation assets.

Current reporting uses different capital cost figures depending on the assumed scope of full buildout, so the power scale is the more useful anchor for understanding the project. At 1.44 GW, the proposed campus belongs in the category of infrastructure that must be planned alongside regional energy systems rather than treated as another commercial building. Sensaka’s guide to data center construction cost shows why projects of this size depend on much more than server procurement.

Site selection is increasingly about access to an energy ecosystem

The proposed site is close to the Braemar substation and several nearby generation facilities. That geography is central to the project story. AI data centers can require such large blocks of electricity that developers are increasingly looking for locations where generation, transmission and land can support expansion together.

A large grid connection on paper is not enough. Developers need to understand how much capacity can be delivered, when it can be energized and what upgrades are required. They also need backup and power quality strategies that can handle sudden workload changes from dense GPU systems.

Sensaka’s data center power calculator works at rack scale, but the principle is identical: planning should be based on usable electrical capacity with continuous load headroom rather than a headline maximum.

Phased construction is a response to infrastructure uncertainty

Data Center Dynamics reported that the campus is planned in multiple phases. That is a practical way to align building construction with customer commitments, power availability and equipment delivery. A 1.44 GW campus is unlikely to appear as one fully operational block on a single day.

Phasing also reduces some financial risk. Developers can bring capacity online as infrastructure becomes available rather than committing the entire capital program before demand is proven. The tradeoff is operational complexity because each phase can have different equipment generations, cooling designs and network requirements.

This is where AI era data center capacity planning becomes important. The useful unit is deployable capacity after power, cooling, network and operational constraints are considered together. A future phase may have land reserved while still lacking the electrical or thermal systems needed to support actual AI racks.

Cooling design will decide how much of the electrical capacity becomes compute

Large AI campuses convert enormous amounts of electricity into heat. The proposal has been reported as using primarily air cooling with closed loop water systems and other measures. The final thermal architecture will matter because cooling overhead changes how much of the site’s electrical capacity can be delivered to IT equipment.

High density accelerator deployments may also push individual halls toward more advanced liquid cooling even if other parts of the campus remain air cooled. That creates a mixed facility where operators have to monitor different thermal systems and understand how each affects capacity.

Cooling is also connected to local resource questions. Water use can influence public support and permitting, while more electrically intensive cooling can increase demand on the grid. No design eliminates every tradeoff.

Gigawatt scale campuses are becoming regional infrastructure

A project measured in more than one gigawatt affects more than its owner. It can influence transmission planning, generation investment, local construction markets, water infrastructure and community expectations. It may also attract suppliers and other digital infrastructure around the site.

That makes transparency important. Developers need credible schedules for how quickly power demand will ramp. Utilities need to know which phases are committed. Communities need understandable information about water, noise, land use and employment. Investors need to distinguish between announced capacity and capacity likely to generate revenue.

The Western Downs proposal is another sign that the AI infrastructure race is expanding beyond traditional data center hubs. The winning locations may be those that can combine land, power, permitting and operating capability in one development path. At 1.44 GW, Zerra’s proposal is best understood as both a data center project and an energy infrastructure project.

Originally published on the Sensaka blog.

The Dutch Government Is Treating Data Center Sustainability as an Operations Problem

Da — Mon, 24 Aug 2026 18:57:45 +0000

The Dutch government is taking a more coordinated approach to reducing the environmental impact of its own data centers. A government update published in late July 2026 said four public sector data center organizations completed the first phase of the Verduurzamen Overheidsdatacenters project, which is intended to make government data center sustainability work faster, more consistent and more measurable.

The participating organizations include government data centers supporting Rijkswaterstaat and the Justice Information Services Organization, Haaglanden, the Tax Administration and ODC Noord. These facilities underpin services such as DigiD, tax filing and the digital systems used to manage infrastructure including bridges and locks. The government's own explanation therefore frames sustainability as part of the future reliability of public digital services, not simply as an environmental reporting exercise.

This is a useful operating model because efficiency has to be measured before it can be improved. Sensaka's explanation of Power Usage Effectiveness shows why facility level energy ratios can provide a baseline while still leaving important questions about rack utilization, hardware condition and stranded capacity unanswered. The Dutch project is moving in the same direction by emphasizing shared metrics and data driven management rather than isolated sustainability claims.

Government data centers face the same AI pressure as commercial facilities

The Dutch update notes that energy demand is rising as digital services expand and AI applications become more common. That matters because public infrastructure cannot avoid the same physical limits affecting commercial operators. More compute requires more electrical capacity, more cooling and more careful lifecycle management.

The difference is that public sector facilities also have a policy role. The Netherlands has climate neutrality goals for 2050 and government organizations are expected to act as a model buyer and operator. That creates pressure to reduce carbon emissions, comply with energy efficiency requirements and improve transparency around energy, water, renewable power and waste heat.

The project therefore combines operational and policy objectives. It is intended to identify savings, share knowledge between data centers and create a common structure for deciding which sustainability actions should be implemented across the government estate.

Hardware replacement is being questioned more carefully

One of the more interesting findings around the program is that server age does not always predict energy efficiency as neatly as procurement cycles assume. Dutch reporting on the project says an analysis of nearly 3,800 servers found cases where systems around nine years old performed similarly in energy efficiency to systems only two years old.

That does not mean old servers are generally as efficient as new ones. Workload type, processor generation, utilization, memory, power supplies and performance requirements all matter. The important point is that replacement decisions can be improved when they use measured performance instead of a fixed age rule.

Extending the useful life of suitable hardware can reduce embodied carbon, avoid unnecessary procurement and lower electronic waste. Keeping inefficient equipment for too long can produce the opposite result by wasting electricity and occupying valuable rack capacity. The right answer therefore depends on telemetry and workload context.

This is a stronger sustainability model than assuming that every hardware refresh is automatically green because the new device has a better specification sheet.

Shared metrics make sustainability operational

The government project is moving toward common sustainability agreements and metrics across the four data centers. That is important because sustainability programs often fail when every facility measures performance differently. One site may focus on total electricity. Another may emphasize PUE. Another may report renewable sourcing or equipment age. Without a shared model, management cannot easily compare performance or decide where investment will have the greatest effect.

Common metrics can make the conversation more concrete. Teams can track facility overhead, IT load, utilization, hardware lifecycle, cooling performance, renewable energy and other factors over time. The goal is not to force every data center to produce identical numbers, because the facilities may support different workloads. The goal is to make the differences understandable enough to support decisions.

The Dutch initiative is therefore interesting beyond the public sector. Sustainable data center operations are becoming less about a single efficiency project and more about continuous management. Energy, hardware, architecture, procurement and software choices all influence the final footprint.

As AI increases computing demand, that approach becomes more important. The most sustainable data center is unlikely to be the one with the best slogan. It will be the one that can show what it consumes, why it consumes it and how operational decisions improve the result over time.

Originally published on the Sensaka blog.

Texas Data Center Interconnection Pause Extends Toward December as ERCOT Audits the AI Boom

Da — Mon, 24 Aug 2026 18:57:09 +0000

Texas is moving deeper into its review of large data center projects connected to the state power system. Utility Dive reported on August 21, 2026 that the Electric Reliability Council of Texas intends to complete a broad audit by December 10. The review follows Governor Greg Abbott’s August 3 call for a moratorium on new data center interconnections until questions about electricity, water and public financial support can be answered.

The headline number is extraordinary. ERCOT’s interconnection queue contains about 474 gigawatts of requests, and Abbott said roughly 90 percent of new power requests are associated with data centers. That does not mean 474 gigawatts of projects will be built. It shows how difficult it has become for utilities to separate serious demand from speculative or duplicated requests. Sensaka’s guide to data center capacity planning describes the facility level version of the same problem: nominal capacity is less useful than capacity that can actually be powered, cooled and operated.

The audit is about project credibility as much as electricity

ERCOT is reviewing hundreds of large proposed facilities and asking developers for more information. Utility Dive reported that around 300 data centers of 75 MW or larger are navigating the Batch Zero process. Community impact reviews will also apply to data centers and crypto facilities of 25 MW and above.

The state wants to know how much electricity projects expect to use, whether they plan to provide their own generation, how much water they need and whether they depend on public incentives. Those questions turn a data center proposal into a broader infrastructure commitment. A campus is no longer evaluated only on land, financing and customer demand. It must also explain its relationship with the grid and the surrounding community.

This makes power modeling central to development. Sensaka’s data center power calculator illustrates the basic relationship between device load, current and continuous operating limits. At utility scale the same discipline applies, although the numbers involve substations, transmission and generation rather than rack circuits.

The queue shows the difference between requested and usable capacity

Large interconnection queues can exaggerate future demand because developers may submit requests for several possible sites or reserve capacity before financing and customers are fully committed. Utilities still have to plan for the possibility that some projects are real, because transmission and generation can take years to build.

Texas is therefore trying to establish which loads are credible enough to shape long term planning. The delay matters because a data center can complete other parts of its development while waiting for certainty about energization. That creates financial risk for developers and their suppliers.

Inside a facility, a similar mismatch produces stranded capacity. A hall may have open rack positions but no remaining power or cooling headroom. Sensaka’s overview of data center management connects power, cooling, assets and operational capacity because each limit changes what the building can actually support.

Water and community impact are moving onto the critical path

Texas is also asking data centers to explain water consumption and whether they will rely on supplies needed by local communities. This matters because cooling design can shift the balance between water and electricity use, especially in large AI deployments.

Community impact is becoming harder to separate from technical design. A project may need new transmission infrastructure, backup generation, water systems and road access. Residents may care about noise, land use and utility prices as much as the developer cares about latency and fiber routes.

For data center companies, those concerns should be treated as project dependencies. Cooling architecture, site design and power strategy can influence approval timelines. Transparent operating data can also become useful evidence when regulators want to understand whether promised efficiency or demand management measures are actually working.

December will not necessarily end the planning problem

ERCOT’s target is to deliver a comprehensive report around December 10, but Utility Dive reported that the original Batch Zero study timeline is already unlikely to hold. Grid officials said they do not yet know how much the audit will shift the next stages of interconnection review.

That uncertainty is the larger lesson. AI investment can move faster than transmission planning, power plant construction and public approval. Developers may have capital and customers while the physical infrastructure underneath the project remains unresolved.

Texas is still likely to remain one of the most important United States data center markets. The current pause shows the condition attached to that growth: projects will increasingly need to prove that their requested capacity is credible, supportable and compatible with regional infrastructure. In the AI era, the ability to demonstrate usable capacity may become as important as the ability to announce it.

Originally published on the Sensaka blog.

Schwarz Group’s €11 Billion Lübbenau Data Center Makes European Sovereign AI Physical

Da — Mon, 24 Aug 2026 18:41:25 +0000

The €11 billion data center investment by Schwarz Digits in Lübbenau, Brandenburg is receiving renewed attention as construction advances. The project itself was announced earlier. In November 2025, Schwarz Digits said it was building one of Europe’s most modern data centers on the site of a former power plant, with an initial 200 MW grid connection and planned completion by the end of 2027. Recent reporting from Tagesspiegel shows cranes and construction activity now reshaping the former industrial site.

The strategic importance goes beyond the size of the investment. Schwarz Digits is the IT and digital division of the Schwarz Group, whose businesses include Lidl and Kaufland. The Lübbenau facility is intended to support cloud and AI capacity, including the STACKIT cloud platform. For Europe, this turns digital sovereignty from an abstract policy goal into a physical infrastructure question. Sensaka’s Europe data center map provides context for how facilities and operators are distributed across the region.

Sovereign cloud still depends on ordinary physical constraints

Discussions about sovereignty often focus on where data is stored, which legal jurisdiction applies and who controls the software platform. Those questions matter, but sovereign infrastructure still requires electricity, cooling, network capacity, land and skilled operators.

The Lübbenau project is a useful example because it is being built at the scale where those physical systems become strategic. A 200 MW initial connection is not simply an IT specification. It is a major regional power relationship. The facility also has to support dense AI systems whose power and cooling profiles can differ significantly from conventional enterprise computing.

Sensaka’s guide to data center power planning illustrates the facility level discipline behind those large headline numbers. Operators ultimately need to translate available electrical capacity into real equipment load, redundancy and continuous operating headroom.

The former power plant location shows how infrastructure geography is changing

Lübbenau was historically associated with energy production. Reusing a former power plant site for digital infrastructure highlights a broader European trend in which data center developers look beyond established metropolitan clusters for locations with land, grid access and industrial infrastructure.

That matters because traditional data center markets can face long grid connection times and higher land costs. AI infrastructure is power intensive enough that proximity to available electrical infrastructure can influence site selection as much as proximity to a large city.

The shift also creates new regional development questions. A large facility can bring construction activity and long term technical jobs, but it also requires public infrastructure and local acceptance. Data center developers increasingly need to demonstrate how the project will use power, manage noise and cooling, and fit into regional energy plans.

Heat reuse is becoming part of the operating story

Schwarz Digits said waste heat from the data center is planned to feed into Lübbenau’s district heating network from 2028. Heat reuse is attractive because data centers continuously convert electricity into heat, and using some of that energy locally can improve the broader efficiency story.

The practical value depends on operating conditions. The heat must be available at useful temperatures, the district network needs customers at the right times and the systems require reliable interfaces between the facility and the heating network. Heat reuse should therefore be treated as infrastructure with its own monitoring and availability requirements.

Sensaka’s guide to data center cooling systems explains how thermal design affects hardware health and facility efficiency. At AI densities, cooling architecture and heat rejection are becoming central design decisions rather than supporting details.

Europe’s AI competition is moving into capital intensive infrastructure

European policymakers frequently discuss the need for domestic cloud capacity and less dependence on foreign technology providers. The Lübbenau project demonstrates what that ambition costs when it becomes physical. Billions of euros are required not only for servers but also for the building, power systems, cooling, networking and the operational platform around them.

This creates a different competitive test. A sovereign cloud provider must offer useful services and competitive economics while also operating capital intensive facilities reliably. Data sovereignty can influence buyer preference, but it does not remove expectations around performance, uptime and price.

The Schwarz Digits investment is therefore important because it links a European strategic objective with a specific, visible construction project. Digital sovereignty depends on software and policy, but it also depends on whether Europe can build, energize, cool and operate large AI facilities at scale. Lübbenau is one of the places where that question is being answered in concrete and megawatts.

Originally published on the Sensaka blog.

Nvidia Server Price Hikes Show Memory Is Becoming an AI Data Center Cost Constraint

Da — Mon, 24 Aug 2026 18:40:48 +0000

Nvidia customers are reportedly being told to expect significantly higher prices for AI server systems delivered in early 2027. Reuters reported on August 22, 2026 that price increases above 15 percent are being communicated to major customers, with higher memory costs identified as the main driver. The report said systems based on Vera Rubin and Grace Blackwell platforms are among those affected and that the final increase will vary according to chip generation and memory configuration.

For data center operators, this is more than a procurement headline. AI infrastructure costs are created by a chain of dependencies that includes accelerators, high bandwidth memory, servers, networking, power delivery, cooling, storage and the building itself. A large change in one component can alter the economics of an entire deployment. Sensaka’s guide to data center cost is useful precisely because it treats IT equipment as one part of a broader capital and operating cost model.

Memory is becoming a first order infrastructure issue

AI systems depend heavily on fast access to large amounts of data. That makes memory capacity and bandwidth central to accelerator performance. As model sizes grow and inference workloads become more demanding, the amount and type of memory attached to each system can have a major effect on both performance and cost.

The reported price increases show how that pressure can move upstream into server pricing. A data center team may plan around a certain number of accelerators, only to find that the complete system cost changes because memory supply and configuration have become more expensive. At scale, a percentage increase that looks manageable on a single server can add millions of dollars to a cluster budget.

This is why total cost analysis should be updated continuously during long procurement cycles. Sensaka’s guide to data center TCO includes equipment, energy, cooling, staffing, software, maintenance, downtime and lifecycle assumptions. AI infrastructure planners should add component price volatility and delivery timing to the same model rather than treating the original hardware quote as fixed.

A more expensive server changes the value of usable capacity

When hardware costs rise, stranded capacity becomes more painful. An expensive GPU server that cannot be powered, cooled or connected on schedule is capital sitting idle. The problem can appear when a project has physical rack space but insufficient electrical headroom, when the cooling design cannot support the density, or when network and storage infrastructure are not ready for the cluster.

Sensaka’s guide to data center capacity planning for the AI era focuses on usable capacity rather than nominal capacity. That distinction becomes financially important when each deployed system costs more. A procurement team should know not only how many servers it can buy, but how many can be installed and operated within the available power, cooling and network envelope.

The same logic applies to phasing. If a facility can energize only part of a cluster in the first year, buying all hardware at once may expose the project to unnecessary depreciation and price risk. If component costs are rising, buying too late can also be expensive. Procurement therefore has to be coordinated with facility readiness instead of operating on a separate calendar.

Power and cooling costs do not disappear when hardware gets more expensive

Higher server prices do not reduce the operating demands of AI equipment. Dense accelerator systems still require substantial electrical capacity and thermal management. In fact, newer platforms can create additional infrastructure requirements that increase the cost of power distribution, cooling loops and monitoring.

Sensaka’s AI data center operations guide connects GPU health, power, cooling, networking and capacity as one operating model. That integrated view is important when hardware prices move because the business case should be based on productive compute delivered over time, not the purchase price alone.

Operators should also expect finance teams to ask harder questions about utilization. If server prices rise while electricity, construction and financing remain expensive, low GPU utilization becomes increasingly difficult to justify. Monitoring needs to show whether valuable systems are actually doing useful work and whether facility constraints are preventing full use of the installed hardware.

The AI infrastructure budget is becoming more sensitive to supply chains

The AI boom was initially framed around GPU scarcity. The next phase is showing that memory, advanced packaging, networking, power equipment and cooling systems can all become bottlenecks in turn. A server price increase driven by memory is a reminder that the infrastructure stack is only as predictable as its most constrained components.

For operators planning 2027 capacity, the practical response is to model multiple hardware price scenarios, keep facility readiness tied to procurement milestones and measure the cost of delays as carefully as the cost of equipment. A cluster that arrives at the wrong time can be expensive even if the purchase price is attractive. A cluster that arrives on time but exceeds the available power or cooling envelope can be worse.

The reported Nvidia price increases therefore matter because they push AI infrastructure planning further away from simple server counts. Memory supply, deployment timing, facility capacity and lifecycle economics are becoming one connected budget problem.

Originally published on the Sensaka blog.

ChatGPT Keeps Changing the Types of Pages It Cites: GEO Needs a Moving Benchmark

Da — Mon, 24 Aug 2026 18:34:36 +0000

Marketers looking for a permanent ChatGPT citation formula are likely chasing the wrong target. An August 23 analysis from NetContentSEO argues that AI search citation patterns are changing over time and that the types of pages performing well today may not play the same role a few months later. The article points to Peec AI citation datasets and a much smaller 20-query experiment conducted by NetContentSEO. One example cited in the analysis concerns machine-translated Reddit pages: their share of ChatGPT’s Reddit citations reportedly fell sharply between April and early June 2026, while Google AI products behaved differently. NetContentSEO is explicit that its own 20-query experiment is too small to establish a general trend. That caveat is important. The useful conclusion is not that one specific page type is suddenly “dead.” The useful conclusion is that AI search visibility needs longitudinal measurement because the retrieval system itself can change.

GEO cannot rely on a frozen list of winning sources

Traditional SEO changes constantly, but practitioners can still work with relatively durable concepts such as crawlability, relevance, links, internal architecture and user intent. AI search adds another layer of volatility. The answer engine may change models, retrieval methods, query expansion, source selection and citation presentation without giving marketers a clean version history of every ranking behavior. That means a tactic derived from one month of citation data can decay quickly. If a marketer observes that Reddit, listicles or product pages appear frequently and then builds the entire program around that format, the strategy becomes dependent on a pattern the marketer does not control. Mustard Seed’s guide to best practices for GEO is more durable because it focuses on making information clear, useful and easy for answer systems to interpret. Those principles survive source-mix changes better than chasing one temporary citation pattern.

The right benchmark is a repeated prompt set

One of the strongest ideas in the NetContentSEO article is methodological. Instead of constantly asking new questions, repeat the same meaningful questions over time and observe what changes. For a B2B company, that might mean maintaining a set of prompts around category education, comparisons, alternatives, implementation questions and buying criteria. Each prompt can be tracked across ChatGPT, Gemini, Perplexity, Copilot and Google AI experiences. The measurement should record more than whether the brand appears. Capture which competitors are mentioned, which domains are cited, which page types appear and how the answer frames the category. That creates a time series rather than a screenshot. A broader GEO versus SEO framework is useful because the two channels have different units of measurement. SEO often tracks rankings, impressions and clicks. GEO increasingly requires prompt-level observations, source analysis and answer context.

Different AI engines should be measured separately

The NetContentSEO analysis also reinforces another important point: source preferences can differ across platforms. A page performing well in ChatGPT may not perform the same way in Gemini or Google AI Mode. Even products from the same company can use different retrieval and citation systems. Marketers should therefore resist collapsing every engine into one “AI visibility score.” A blended score is convenient for reporting, but it can hide where the brand is actually strong or weak. The better model keeps the engine visible. A company might have strong ChatGPT visibility, weak Gemini visibility and high Perplexity citation frequency. Those differences can guide where content, technical access or authority work should be prioritized. Mustard Seed’s guide to what GEO means treats generative search as a distinct discovery environment. The implication is that measurement should preserve the characteristics of each environment rather than assume one universal ranking system.

Source volatility makes brand fundamentals more valuable

If citation patterns are unstable, the safest strategy is to invest in assets that remain valuable even when one source type loses visibility. Original research is useful to customers, journalists and AI systems. Clear product documentation helps buyers and retrieval engines. Real reviews and credible third-party coverage create independent evidence. Strong category explanations improve both SEO and AI discoverability. This is less exciting than finding a secret citation hack, but it produces a more resilient information footprint. A visibility program should still run experiments. It can compare structured pages, FAQs, product pages, glossary content, research and community distribution. The difference is that experiments should be treated as tests, not permanent laws. When a pattern appears, repeat it. Check whether it persists across engines and time. Only then should it influence significant content investment.

GEO teams need change detection, not only visibility reporting

Most AI visibility dashboards answer a current-state question: where does the brand appear today? The next maturity step is change detection. Which prompts changed? Which sources disappeared? Which competitors gained visibility? Did a specific domain suddenly become more or less important? Did the answer engine begin citing a different page type? Those changes can reveal emerging opportunities and prevent a team from continuing to optimize against an outdated pattern. Mustard Seed’s visibility revenue calculator connects discoverability with commercial assumptions. The same commercial discipline should be applied to citation monitoring. A source shift matters most when it affects prompts tied to meaningful buying decisions, not because a dashboard line moved. The NetContentSEO experiment is small, and its authors say so. Its larger question is still worth adopting: stop asking only which sources AI prefers and start asking how those preferences are changing. For GEO, that is a better operating model. The target moves, so the measurement system has to move with it.

Originally published on the Mustard Seed blog.

China Publishes Its First National Standard for Data Center Cold Plate Liquid Cooling

Da — Mon, 24 Aug 2026 18:34:00 +0000

China has published GB/T 48023-2026, titled Technical Specification for Data Center Cold Plate Liquid Cooling Systems, with implementation scheduled for February 1, 2027. Chinese industry coverage describes it as the country’s first national-level standard specifically for cold plate liquid cooling in data centers. An August 24 report from China Industry Development Research says the standard was jointly released by the State Administration for Market Regulation and the Standardization Administration of China. The report identifies Aite Network Energy as one of the drafting organizations involved in technical discussion, data validation and preparation of provisions. The significance goes beyond a new document number. Cold plate cooling is moving from specialized deployments into mainstream AI infrastructure, where differences in connectors, coolant loops, monitoring and operating procedures can create expensive integration risk. Sensaka’s guide to liquid cooling in data centers explains why operators have to manage flow, temperature, pressure, leak detection, pumps and coolant condition as part of the production environment.

Standardization matters when liquid cooling becomes operational infrastructure

Air cooling evolved over decades with familiar equipment classes, maintenance procedures and facility conventions. Cold plate liquid cooling is moving much faster because accelerator density is forcing operators to adopt it on compressed timelines. That speed creates a fragmented market. Different server designs can require different flow rates, temperatures, materials, connectors and coolant conditions. Facilities also have to decide where cooling distribution units sit, how loops are monitored, how leaks are detected and what happens during maintenance. A national technical specification can give designers, equipment manufacturers and operators a more consistent reference point. It does not make every implementation identical, and the available public report does not provide the full technical clauses of GB/T 48023-2026. The important signal is that cold plate cooling now has enough deployment importance to justify a formal national standard. For buyers, standards can reduce ambiguity in procurement. Requirements can be written against an agreed technical framework rather than relying entirely on proprietary vendor descriptions. That can also make acceptance testing and long-term maintenance easier to structure.

AI rack density is pushing cooling closer to the compute

The need for liquid cooling comes from basic physics. Higher server power creates more heat, and moving that heat through air becomes increasingly difficult as rack density rises. Cold plates remove heat closer to high-power components by circulating liquid through plates attached to processors or accelerators. This can reduce the amount of heat that must be handled by traditional room air systems, although most deployments still require a broader facility cooling architecture. Sensaka’s guide to data center cooling systems places direct-to-chip cooling alongside containment, free cooling and other facility approaches. The key operational point is that cooling becomes a chain. A cold plate can transfer heat efficiently, but pumps, heat exchangers, CDUs, facility water loops and heat rejection equipment still have to work correctly. As AI facilities deploy more liquid-cooled racks, these components become part of the availability path for compute. A pump failure or loss of flow can become as operationally important as a server hardware fault.

Monitoring requirements become more specialized

Cold plate environments add telemetry that many traditional monitoring stacks were not designed to treat as first-class infrastructure data. Operators may need to track supply and return temperature, differential pressure, flow rate, pump status, valve position, leak sensors and coolant quality. Those signals need to be connected with rack location, server identity, workload criticality and facility alarms. Sensaka’s guide to data center monitoring software describes monitoring across hardware, power, cooling, networks and services rather than isolating each subsystem. That becomes especially relevant with liquid cooling because thermal events can originate in several layers of the loop. A standard can help define technical expectations, but operations teams still need a data model that shows what equipment is connected to which loop and what business services depend on it. Without that context, a liquid cooling alarm becomes another isolated notification instead of an actionable infrastructure event.

Standards can reduce integration risk without eliminating design work

It would be a mistake to interpret GB/T 48023-2026 as proof that cold plate deployment is now simple. Standards provide common requirements, but facility conditions still vary. Operators have different water temperatures, redundancy targets, rack densities, floor layouts and maintenance practices. Server manufacturers may also implement cooling interfaces differently within the boundaries allowed by the standard. The useful outcome is a stronger baseline. Engineering teams can compare vendor designs against a recognized specification, identify deviations explicitly and build test procedures around known requirements. This is similar to the role standards play elsewhere in data center engineering. They reduce unnecessary variation and create common terminology while leaving room for facility-specific design.

AI data center operations now include the cooling loop

The larger trend is that cooling is moving deeper into IT operations. When racks rely on direct liquid cooling, facilities and IT teams can no longer treat thermal systems as background building services. Sensaka’s AI data center operations guide connects GPU health with power, thermal conditions, cooling loops, networking and capacity. That operating model is likely to become more common as standards make cold plate systems easier to procure at scale. The publication of GB/T 48023-2026 is therefore an industry maturity signal. Cold plate liquid cooling is becoming standardized infrastructure rather than a collection of project-specific experiments. The next challenge is operational consistency. Standards can define how systems should be designed and tested, but reliable AI capacity will still depend on whether operators can monitor the cooling path, detect degradation early and maintain every component without disrupting increasingly valuable compute.

Originally published on the Sensaka blog.

Frankfurt AI Data Center Shows Why Grid Connection Is Becoming the Real Project Gate

Da — Sat, 22 Aug 2026 11:42:10 +0000

A planned AI data center campus in Frankfurt (Oder), Germany has reached an advanced planning stage, yet its full scale still depends on one of the hardest resources in the data center market: grid capacity. An August 22 report from Märkische Oderzeitung highlighted the project while focusing on the need for a network connection. Public project information from Frankfurt DATA FFO AI describes a campus designed for up to 350 MW of power capacity for AI, cloud and other data intensive applications.

Regional development information says the project is planned in six modules and could involve total investment of up to €3.5 billion when fully built. The development team has identified a planned connection to the 380 kV transmission grid operated by 50Hertz as part of the infrastructure required for the campus. The site has advanced through local planning and permitting steps, but the electricity path remains central to how much computing capacity can actually be deployed.

The project is a useful example of a broader shift in data center economics. Sensaka's guide to data center power planning illustrates the same principle at facility level: nominal space does not become usable IT capacity until the electrical path can support the intended load. At hyperscale, that calculation extends far beyond the rack and into substations, transmission infrastructure, grid queues and utility planning.

Planning permission does not create megawatts

Data center development is often described through land area, investment value, building size or planned rack capacity. Those numbers can create the impression that a project is largely determined once planning approval is secured. AI infrastructure has made that assumption less reliable because large clusters need unusually high and concentrated electrical loads.

A campus can have land, a strong fiber position, municipal support and a viable building design while still facing a long path to full energization. Transmission upgrades may be required. New substations may need to be designed and permitted. Grid operators have to evaluate the effect of the load on surrounding infrastructure. Other industrial users may be competing for the same capacity.

This changes the sequence of project risk. In an earlier data center market, electricity could sometimes be treated as one workstream within a broader construction program. For the largest AI projects, grid access can become the workstream that determines the schedule of everything else.

AI makes the power curve steeper

The Frankfurt project is designed specifically around AI and data intensive applications, which matters because accelerator clusters compress more computing into each rack. Higher density can improve the productivity of floor space, but it also concentrates electrical and thermal demand. Power distribution, cooling systems and network design all have to scale with the compute layer.

This is why headline megawatts should be separated from energizable megawatts. A developer may have a long term design target of several hundred megawatts while only being able to bring capacity online in phases as the grid connection develops. The useful project question is therefore not simply how large the campus is intended to become. It is how much power can be delivered, when it can be delivered and under what conditions.

Battery storage and on site generation can help with resilience, peak management or transition planning, but they do not automatically replace a high capacity grid connection for a continuously loaded hyperscale campus. The underlying electrical system still has to support sustained operation at the scale the computing business requires.

Grid readiness is becoming a competitive asset

As AI infrastructure expands, locations with credible power pathways may gain an advantage over sites that appear attractive on land and tax economics alone. This can shift value toward regions with transmission capacity, faster interconnection processes, generation access and a realistic route to new substations.

It can also change negotiations between data center developers and public authorities. A municipality may want the investment and jobs associated with a large campus, but the project may require grid upgrades whose cost, timing and wider economic effects need careful allocation. The resulting development process becomes an energy infrastructure program as much as a real estate program.

Frankfurt (Oder) therefore offers a useful lesson well beyond Germany. The next wave of AI data center competition will not be decided by who can announce the largest campus. It will be decided by who can turn planned capacity into powered, cooled and operational compute on a schedule customers can actually use.

Originally published on the Sensaka blog.

75% of Americans Oppose a Data Center Near Home. The Industry Has a Trust Problem

Da — Sat, 22 Aug 2026 11:41:34 +0000

A new poll has turned the social side of the data center boom into a number that infrastructure developers cannot easily ignore. Heatmap reported on August 20, 2026 that 75% of registered voters surveyed would oppose a new data center near where they live, with more than six in ten saying they would strongly oppose it. The survey covered 2,045 registered voters across all 50 states and Washington, D.C. and was conducted by Embold Research from August 8 to August 13.

The direction of travel may matter more than the exact percentage. Heatmap has asked the same question several times over the past year and reported a sharp deterioration in local support. Other surveys use different samples and wording, so they produce different numbers. An Annenberg Public Policy Center survey published in August found 61% opposition to local data center construction, up from 49% earlier in the year. The figures are not identical, but the message is consistent: a growing share of Americans are uncomfortable with large data center projects close to home.

For infrastructure teams, this adds a new constraint to the familiar list of land, power, cooling, network access and construction. Sensaka's guide to data center capacity planning focuses on usable capacity across space, power and cooling. The latest polling suggests that community acceptance increasingly belongs in the same planning conversation because a technically viable site can still face political or permitting resistance.

The objection is increasingly about local resource tradeoffs

Public resistance cannot be reduced to a simple dislike of technology. Recent polling and reporting repeatedly point to electricity demand, water use, utility bills, noise, land use and the perceived distribution of economic benefits. A resident may support artificial intelligence in general while still questioning whether a nearby facility will increase pressure on the local grid or consume resources that are already scarce.

That distinction matters because many arguments used to defend data centers operate at the national level. Developers can point to economic growth, AI leadership, cloud capacity, digital sovereignty and construction investment. Local residents experience a different set of questions. They want to know who pays for grid upgrades, whether power prices could rise, what happens to water demand, how much permanent employment remains after construction and whether tax benefits justify the physical footprint.

The gap between national benefit and local cost is becoming a central problem for the industry. If the project narrative focuses only on compute capacity or investment value, it can sound disconnected from the concerns that determine whether a project earns local support.

Community acceptance is becoming part of site selection

Data center site selection has traditionally emphasized measurable engineering and commercial factors. Power availability, fiber routes, land cost, latency, tax incentives and climate conditions can all be scored. Community resistance is harder to model, but developers may increasingly need to treat it as a project risk rather than a communications issue that begins after a site has already been chosen.

That means due diligence should include local electricity conditions, water stress, housing and land pressures, existing industrial development, political sentiment and the credibility of promised community benefits. A county that looks attractive because it has inexpensive land and a nearby transmission corridor may become less attractive if residents believe the project threatens their energy costs or quality of life.

The industry also needs to distinguish between concerns it can address and objections it cannot simply message away. Better explanations may help when people lack information about how a facility operates. They will not solve a real grid constraint, an unfavorable utility cost allocation or a water problem. Trust improves when project design changes in response to valid concerns, not only when the public relations campaign becomes more polished.

Operators will need to show measurable local value

The next generation of successful projects may need a clearer local compact. That could include transparent power procurement, credible water strategies, noise limits, infrastructure investment, tax commitments, workforce development and public reporting after the facility opens. The exact package will differ by region, but the principle is straightforward: communities increasingly expect to understand the exchange they are being asked to make.

This will also put more pressure on operational data. Claims about efficiency are more persuasive when developers can show measured energy use, cooling performance and progress against stated targets. A promise made during planning has limited value if the operator cannot later demonstrate what happened in practice.

The 75% figure should not be treated as a universal measure of American opinion. It is one poll, and other surveys report different levels of opposition. It should, however, be treated as a warning. Data center growth now depends on more than finding power and land. The industry also has to earn permission to build.

Originally published on the Sensaka blog.

Spain's Cybersecurity Bill Shows What NIS2 Means for Infrastructure Operations

Da — Sat, 22 Aug 2026 11:32:34 +0000

A Spanish cybersecurity article published on August 21, 2026 has renewed attention on the country's proposed Law on Cybersecurity Coordination and Governance. The underlying draft is not new: Spain's Council of Ministers approved the preliminary bill in January 2025 as part of the national process for transposing the EU NIS2 Directive. The useful story in 2026 is that the legislation remains part of a wider shift from general cybersecurity guidance toward formal governance, incident reporting and operational accountability.

Official Spanish government material says the proposed law is intended to transpose Directive (EU) 2022/2555 and create a more coordinated national cybersecurity framework. The draft covers public and private entities in important sectors and creates a national coordination structure intended to improve cooperation across authorities and during major incidents. It also defines responsibilities around information security and the management of cybersecurity policies and incidents.

For infrastructure teams, the practical consequence is that cybersecurity evidence increasingly has to come from the environment itself. Sensaka's guide to network segmentation is one example of the operational layer behind regulation: an organization can state that sensitive systems are separated, but effective governance requires the network design, access controls and monitoring to make that separation real.

NIS2 moves cybersecurity closer to operations

NIS2 expands the set of organizations expected to manage cyber risk systematically and raises expectations around governance, incident handling, supply chain risk and continuity. National laws determine how those requirements are implemented and supervised within each member state, but the operational direction is consistent across Europe.

Security therefore becomes harder to isolate inside a specialist team. A cybersecurity policy may require vulnerability management, yet the organization still needs an accurate inventory of the servers, network devices and software that are in scope. An incident reporting obligation may define a timeline, yet the organization still needs monitoring that can detect the event quickly enough to start that clock. A continuity requirement may call for resilience, yet teams still need tested recovery paths for the systems that support essential services.

This is why regulation increasingly exposes weaknesses in basic infrastructure management. Unknown assets, stale diagrams, unmanaged firmware, weak network segmentation and incomplete dependency records are not separate housekeeping problems when they affect the organization's ability to understand or respond to cyber risk.

Governance needs named responsibility and usable information

Spain's draft creates a stronger governance model and includes the role of a person or body responsible for information security within covered entities. Assigning responsibility is important, but the accountable person still needs information that can support decisions.

A security leader cannot meaningfully approve a risk posture without knowing which assets are active, which vulnerabilities remain unresolved, which dependencies support critical services and whether protective controls are functioning. The quality of governance is therefore limited by the quality of operational visibility.

This is also where cybersecurity and IT operations converge. Infrastructure teams manage patching, device configuration, hardware lifecycle, backups, network changes and availability. Security teams manage threat exposure and control requirements. Regulation makes it increasingly difficult for those two views to remain disconnected because incidents rarely respect organizational boundaries.

Incident reporting raises the value of detection context

Reporting obligations sound administrative until an actual incident occurs. At that point, the organization needs to determine what happened, which systems were affected, whether critical services were disrupted, what data may be involved and whether the event meets a reporting threshold.

Those questions require context. An alert from a single endpoint may be insufficient if teams cannot connect it to the server's business role, network relationships and dependent services. A hardware failure may initially look unrelated to cybersecurity but still complicate containment or recovery. A third party outage may create a service disruption without any malicious activity at all.

The lesson from Spain's draft is broader than the final wording of one national law. European cyber regulation is making infrastructure governance more measurable. Organizations will increasingly be expected to know what they operate, how it is protected, how incidents are detected and whether recovery arrangements actually work.

That puts pressure on tools and processes that were once treated as background IT administration. Asset accuracy, configuration evidence, network visibility and tested recovery are becoming part of the compliance story because they are part of the resilience story.

Originally published on the Sensaka blog.

DORA's Hard Part Is Proving Resilience Works in Practice

Da — Sat, 22 Aug 2026 11:31:58 +0000

More than a year after the EU Digital Operational Resilience Act became applicable, the banking sector is moving into a harder phase. The policies exist. Management responsibilities have been assigned. The challenge is proving that the controls work when systems, suppliers and recovery processes are under pressure.

QA Financial reported on August 21, 2026 on a study of 23 banks conducted by risk and treasury consultancy Zanders. Almost all participating banks had assigned management responsibility for digital operational resilience and documented a strategy. The results became less consistent when the study looked at how those strategies were communicated, tested and translated into daily risk controls. Only 12 of the 23 banks reported a comprehensive stakeholder communication plan, and fewer than half had a formal ICT risk appetite statement approved by senior management.

This gap between documented control and operational evidence is directly relevant to infrastructure teams. Sensaka's guide to compliance software and infrastructure evidence examines the same problem from the technology layer: governance systems can record requirements, but organizations still need reliable evidence from the assets, configurations and operational systems those controls are meant to govern.

Testing maturity is still uneven

The Zanders study found that regular ICT resilience testing was generally established, but advanced methods were less consistently adopted. QA Financial specifically highlighted limited use of threat led penetration testing among institutions required to perform it. The report also described variation in how issues discovered during testing were escalated and how results were validated.

That matters because a test has little value if its findings do not create an accountable remediation process. A bank needs to know which service was tested, which dependencies were included, what failed, who owns the corrective action and whether the fix was retested. Evidence has to connect the scenario to the outcome.

The European Central Bank has been making a similar point. Its 2026 supervisory priorities emphasize ICT security, third party risk, incident response, change management and preparedness for disruption involving major cloud providers. DORA is therefore becoming less about showing that a testing program exists and more about showing that the program can expose and reduce real service risk.

Recovery plans need to cover complete business services

QA Financial reported that only 16 of the 23 banks said their business continuity and disaster recovery plans comprehensively covered all critical business functions. That is a significant distinction because recovery cannot be demonstrated by restarting a single server or application in isolation.

A critical banking service may depend on identity, networking, databases, middleware, payment systems, customer channels, storage, cloud services and external providers. Any one of those dependencies can prevent the business function from recovering even if the core application is technically online.

This is where operational resilience becomes an infrastructure mapping problem. Teams need to understand which physical and digital components support each critical service, what the acceptable recovery target is and how the service behaves when a dependency fails. Recovery exercises should prove that users can resume the function with correct data and acceptable performance, not merely that individual systems can boot.

Third party risk is becoming a system level problem

The study also found uneven maturity in technology supplier oversight. Most banks had third party risk frameworks, yet fewer than half had robust exit strategies or adequately accounted for geopolitical risks. Concentration and interconnectedness risks were also not consistently incorporated into assessments.

That weakness becomes more important as financial institutions rely on a smaller number of major cloud, SaaS and infrastructure providers. A supplier failure can affect several business services at once. A bank therefore needs to understand not only the risk of an individual vendor, but also how many critical processes depend on the same provider, region, identity platform or network path.

DORA has changed the standard of proof. It is no longer enough to show that the organization has a policy for resilience. The practical question is whether the bank can demonstrate that critical services can withstand disruption, that issues are escalated, that suppliers can participate in recovery and that remediation is verified.

That is a much more demanding test, but it is also a more useful one. A compliance document cannot restore a payment service. Operational evidence can show whether the organization has a realistic chance of doing so.

Originally published on the Sensaka blog.

Equinix Expands Bogotá BG2 as AI Demand Raises the Density Bar

Da — Sat, 22 Aug 2026 11:31:22 +0000

Equinix is expanding its BG2 data center in Bogotá as demand for artificial intelligence and cloud infrastructure grows across Latin America. DPL News reported on August 21, 2026 that the company invested an additional $28 million in the expansion, which adds 550 cabinets and 3 MW of electrical capacity. The original BG2 construction involved about $45 million and began operating in 2024.

The expansion is significant because Equinix is not simply adding conventional colocation space. BG2 is being prepared for higher density workloads, including GPU based AI infrastructure. DPL News reports that the facility uses liquid cooling to move heat away from high output components and that the site has been designed with a possible additional phase that could take electrical capacity to as much as 9 MW without expanding the physical footprint of the building.

That combination of more power in the same footprint captures the operational challenge created by AI. Sensaka's guide to liquid cooling in data centers explains why higher rack density is pushing operators toward cooling designs that can remove heat more effectively than conventional air systems alone. The Bogotá expansion shows that this is already becoming part of mainstream facility planning in growth markets.

AI changes the value of a cabinet

Traditional colocation growth could often be described through cabinet counts and floor space. AI makes those measures less informative unless they are paired with power density and cooling capability. Two cabinets can occupy the same physical area while supporting very different computing loads.

A rack full of conventional enterprise servers may operate within an established air cooled design. A rack containing modern accelerators can require substantially more power and produce far more heat. That means operators can no longer treat every available cabinet position as interchangeable capacity. The useful question is how much high density compute the electrical and thermal systems can sustain.

Equinix's decision to add capacity in Bogotá reflects this transition. The company is preparing the facility for customers that may need dense compute, private interconnection and access to cloud ecosystems at the same time. The commercial product is therefore becoming a combination of powered space, cooling capability and connectivity rather than simple square meters.

Bogotá is also an interconnection story

DPL News reports that Equinix's two Bogotá facilities connect with more than 40 network operators and more than 100 fiber routes reaching the sites. The company also points to private connectivity toward Miami and Atlanta as part of the broader role Bogotá plays in moving data between South America and North America.

This matters for AI because infrastructure performance depends on more than the accelerator itself. Training, inference and enterprise AI workflows can move large datasets between users, storage systems, clouds and model services. A data center that combines dense compute with a strong interconnection ecosystem can therefore provide a different kind of value from an isolated facility that has abundant power but weaker network reach.

The trend also supports distributed AI architectures. Not every workload needs to be concentrated in a single North American hyperscale campus. Inference and data sensitive workloads may benefit from being closer to regional users and enterprise data, especially when latency, regulatory requirements or data movement costs matter.

Sustainability claims will face more operational scrutiny

Equinix says BG2 operates with renewable electricity, primarily hydropower under supply contracts, and that the Bogotá sites also use smaller solar installations for common building services. Those measures sit within a larger industry challenge. AI demand is increasing total electricity consumption at the same time operators are under pressure to improve efficiency and reduce emissions.

The useful measure is therefore not whether a site can describe itself as efficient or renewable in general terms. Operators increasingly need to show how energy sourcing, cooling design, utilization and expansion interact as computing density rises. A facility can become more efficient per unit of compute while still consuming more total electricity because the amount of compute grows faster.

BG2 is a relatively compact example of that future. More capacity is being pushed into an existing location, with liquid cooling and interconnection becoming part of the design response. As Latin American demand grows, the competitive data center will increasingly be the one that can add dense capacity without losing control of power, heat and network performance.

Originally published on the Sensaka blog.