Scott Bergs, CEO of Kirkwood IG: Fiber and the AI Data Center Buildout

As data centers race to support AI workloads, liquid cooling, and accelerated construction timelines, operators are facing new challenges around reliability, efficiency, and infrastructure visibility. Many facilities still rely on commercial-grade instrumentation across critical cooling and utility systems, creating blind spots that can increase operational risk and reduce confidence in performance. In this episode, we explore why leading operators are adopting industrial-grade instrumentation to gain more accurate insight into flow, temperature, pressure, and fluid quality across their facilities. The conversation covers the critical measurement points that impact uptime, how liquid cooling is changing infrastructure requirements, and the role measurement technology plays in helping operators commission faster, reduce risk, and confidently scale for future growth. Listeners will walk away with practical guidance for evaluating instrumentation strategies, selecting technology partners, and preparing their facilities for the next generation of data center demands.

As AI data centers follow available power into new markets, network infrastructure is moving much earlier into the site selection process. In this episode of the Data Center Frontier Show, DCF Editor-in-Chief Matt Vincent speaks with Scott Bergs, CEO of Kirkwood Infrastructure Group, about how hyperscale and high-density compute are changing the physical network requirements behind data center development. For years, fiber could often be addressed after land and power were secured. Bergs says that model is breaking down. Hyperscale and neo-cloud campuses increasingly require multiple physically diverse, high-capacity, low-latency network paths — infrastructure that may be difficult or impossible to add late in the development cycle. That means connectivity planning now has to begin alongside power, roads, permitting and other site infrastructure, sometimes before the ultimate tenant is even known. Bergs also explains the evolution of Kirkwood Infrastructure Group from the team behind DF&I, which built dark fiber infrastructure across Northern Virginia and Maryland. Kirkwood is now expanding across the Southeast and into the Midwest as large data center campuses move into markets where power may be available but communications infrastructure remains relatively thin. The conversation gets into the physical implications of that expansion. Bergs describes customer demand progressing from lit services to dark fiber, then dedicated cables and increasingly dedicated conduit capacity. High-density applications are driving fiber counts from 864 fibers to as many as 6,912 per cable, while conduit itself is becoming a strategic asset for security, routing control and future network expansion. Vincent and Bergs also discuss hollow-core fiber, including its potential latency advantages and the engineering tradeoffs posed by its larger form factor and different implementation requirements. Bergs’ larger point: network developers need to build enough physical pathway capacity today to accommodate whatever fiber technology becomes dominant tomorrow. The episode also examines several emerging infrastructure bottlenecks, including rights-of-way, permitting pressure on agencies such as the U.S. Army Corps of Engineers and transportation departments, and growing strain on the fiber manufacturing supply chain. Perhaps the clearest indication of how much the network map is changing comes from inter-campus connectivity. Bergs says links that once might have stretched two to 30 miles can now extend 250 miles or more. “What might have traditionally been thought of as a long-haul segment or path for us is just another inter-campus connectivity corridor,” he says. The conversation closes with another increasingly important part of infrastructure development: community acceptance. Bergs argues that early, factual engagement is becoming as important as early engineering. “It’s very difficult to fight emotion with facts,” he says. “But you can sometimes prevent negative emotion with positive facts if they’re presented first.” Listen to the full conversation for a detailed look at how power, fiber, conduit, permitting and community engagement are converging in the next generation of AI data center development.

AI Infrastructure’s New Site Selection Equation: Power, Fiber and Permitting AI data center development may begin with power, but it increasingly depends on whether connectivity can be built at the same scale and on the same schedule. In this episode of the Data Center Frontier Show, Matt Vincent, Editor-in-Chief of Data Center Frontier, speaks with Jeff Wabik, CTO of DC Blox, about how AI infrastructure is changing the relationship between data center development and fiber networks. Wabik explains how DC Blox’s site-selection strategy has evolved over the past decade—from finding “dirt close to eyeballs,” to securing more land, to chasing available power, and now to evaluating whether massive, diverse fiber routes can be constructed alongside new campuses. The conversation explores: Why 864-count fiber is now among the smallest deployments DC Blox commonly sees How hyperscalers are becoming major builders of terrestrial and subsea network infrastructure Why new conduit systems may include 10 to 14 ducts from the outset How fiber permitting can take 12 to 18 months and influence routing decisions Why DC Blox may route around jurisdictions with histories of permitting delays Fiber lead times stretching to 70–80 weeks for large-count cable Skilled-trades and operations workforce shortages Why community engagement around water, noise, environmental impacts and tax benefits is moving much earlier in the development process Wabik describes the current AI infrastructure buildout as “beautiful insanity”—an era in which power, connectivity, permits, equipment, labor and community acceptance increasingly have to come together at the same time. “A data center without connectivity is an expensive warehouse.” Listen to the full conversation for a ground-level look at how fiber is becoming part of the critical path for AI data center development.

The AI infrastructure race is largely about getting more computing into data centers faster. But increasingly, operators also need a plan for getting yesterday’s hardware back out while it still has value. In this episode of the Data Center Frontier Show, recorded live at the third annual DCF Trends Summit in Reston, Virginia, DCF Contributing Editor Doug Black speaks with Josh Humm, Data Center Solutions Manager at Dynamic Lifecycle Innovations, about how accelerated AI hardware cycles are changing IT asset disposition, or ITAD. Humm says traditional enterprise infrastructure might remain in service for three to five years. Newer GPU systems, by comparison, can face refresh cycles of just 18 to 24 months. That compressed timeline is colliding with another AI-era reality: the equipment itself is getting heavier, more specialized and more difficult to remove. AI systems can include liquid-cooling manifolds, proprietary configurations and units weighing 5,000 to 6,000 pounds, requiring specialized rigging and decommissioning procedures. At the same time, valuable processors, memory, storage and networking components can depreciate quickly once equipment is taken offline. “The faster we can get the materials out of your building, the more it’s worth, the more we can return to your program,” Humm says. The result, he argues, is that ITAD should become part of lifecycle planning rather than something operators begin thinking about only when hardware reaches end of life. The conversation examines how operators can design decommissioning workflows into facility operations, maintain defensible chains of custody, securely destroy data and determine whether retired equipment should be resold whole, harvested for components or recycled. Humm also discusses the risks created by vendor handoffs across onsite decommissioning, transportation, processing, remarketing and recycling. He recommends scrutinizing providers for data-security and environmental certifications, downstream transparency and the ability to scale as AI refresh projects grow larger. The economics can be substantial. Humm describes a recent project involving an approximately 8- to 10-MW enterprise data center in Colorado whose owner was migrating from on-premises infrastructure to the cloud. Dynamic removed racks and equipment, wiped hard drives onsite and shredded drives that could not be successfully sanitized. After roughly three months, Humm says the project returned more than $17 million net to the customer — about $15 million more than expected. That outcome highlights a larger issue emerging around AI infrastructure: decommissioning is not necessarily just a disposal cost. In a strong secondary market for memory, processors and other components, disciplined asset disposition can return capital to the next hardware cycle. Black and Humm close with two questions operators should ask prospective ITAD partners before a major refresh begins: Can I trust you? And can you scale with me? As GPU infrastructure turns over faster, those questions are likely to become a much larger part of data center operations.

As AI workloads continue to reshape the data center landscape, operators are looking for practical ways to evolve existing infrastructure without overbuilding for tomorrow. In this conversation, we explore how intelligent rack power infrastructure can help data centers support both traditional and AI workloads while improving visibility, efficiency and scalability. The discussion covers the role of intelligent monitoring, open integration and rack-level solutions in creating power infrastructure that can adapt as data center requirements continue to change.

AI is changing more than compute—it is changing how the entire physical infrastructure of the data center must be designed. In this episode, we'll explore why organizations should stop thinking about cabinets, power and cooling as separate decisions and instead view them as an integrated system. We'll discuss how this approach improves efficiency, simplifies deployment and creates a more flexible foundation for high-density AI environments.  Along the way, we'll examine the evolving role of the IT cabinet, the industry's transition from air cooling to hybrid and liquid cooling, and the practical questions organizations should be asking as rack densities continue to climb. From rear door heat exchangers to long-term thermal management strategies, listeners will gain practical insights into building infrastructure that's ready for the next generation of AI workloads. 

Community acceptance may be emerging as one of the most consequential constraints on data center growth. On this episode of the Data Center Frontier Show, DCF Editor in Chief Matt Vincent speaks with Buddy Rizer, Executive Director of Loudoun County Economic Development, and Adam Waitkunas, founder of Milldam Public Relations, about the rising political and community resistance confronting data center development — and what the industry needs to do differently. After two decades working with the world’s largest data center market, Rizer said the current level of opposition is unlike anything he has seen before. “The pushback is universal,” he said. “The talking points are fairly standard from community to community, and it has really become next to impossible to do business.” Rizer believes the industry’s next major constraint may no longer be purely physical. “We thought it was going to be power, but it may be community acceptance and political durability.” The discussion examines how distrust can overwhelm even strong factual arguments. Rizer noted that Loudoun County’s more than 250 data centers collectively use less than 10% of its water system and have generated enormous tax benefits, yet those figures increasingly struggle to break through. “You can’t change how people think until you change how they feel,” Rizer said. Waitkunas argues that developers need to begin community engagement much earlier and give residents a meaningful role in shaping Community Benefit Agreements. Those agreements should go beyond financial contributions to schools, parks or community programs and include operating commitments involving issues such as noise, traffic and other local impacts. Rizer agreed that simply complying with regulations no longer demonstrates partnership. “When the noise ordinance says 55 and they come in at 54.5, that to me, that’s not partnership.” The guests also discuss industry “unforced errors,” perceptions of secrecy surrounding anonymous LLCs and project filings, the growing influence of organized opposition groups, and the risk of companies waiting until a moratorium is already underway before attempting serious community outreach. For Rizer, developers increasingly need to manage two distinct measures of success. “Economic success and community trust are two different balance sheets,” he said. “Industry has to invest in both.” The conversation also explores the role of local governments, particularly smaller communities encountering hyperscale development for the first time; the value of bringing officials and residents through operating facilities that resemble what is actually being proposed; and whether the data center industry ultimately needs a formal standard for community engagement. Both guests expect conditions to become more difficult before they improve. Waitkunas anticipates more moratoriums without significant changes in industry behavior, while Rizer warned that data centers could increasingly become central issues in local political campaigns. “AI hasn’t really created this conversation,” Rizer said, “but it definitely has put it on fast forward.” The central question for the industry is increasingly straightforward: not whether data centers have impacts, but whether developers can understand those impacts, mitigate them, communicate them honestly and build enough community trust to sustain the infrastructure expansion now underway.

Optics is no longer a supporting accessory in the data center network. As AI infrastructure advances from 400G and 800G toward 1.6-terabit connectivity, optical components are consuming a larger share of network cost, power and operational risk. In this episode of the Data Center Frontier Show, DCF Editor in Chief Matt Vincent speaks with Bill Gartner, Senior Vice President and General Manager of Cisco’s Optical Systems and Optics business, about how AI is changing the strategic role of optics. Gartner explains that optics represented roughly 10% of a network port’s bill of materials at 10G. At 400G and above, the optics can cost more than the switch port itself. Reliability has also become critical: A single unstable link can force GPUs operating in parallel to stop, return to a checkpoint and restart. According to data Cisco has seen from hyperscale customers, link flaps can reduce GPU infrastructure efficiency by as much as 40%. The conversation maps the AI network across three distinct tiers: Scale-up: Connections within the rack, carrying approximately 500 times the bandwidth of a traditional WAN environment. Scale-out: Connections between racks, commonly using 400G and 800G pluggable optics. Scale-across: Coherent optical connections between data centers as AI clusters expand beyond the power limits of a single facility. Gartner also discusses Cisco’s 1.6T roadmap, routed optical networking, coherent pluggable optics and the emerging debate around co-packaged and near-packaged optics. These architectures promise lower power consumption and greater density, but introduce new questions involving interoperability, replacement and operational resilience. Looking ahead, Gartner emphasizes that optics is not constraining AI network growth. It is enabling clusters to scale across racks, campuses and geographically distributed data centers, while the coming inference wave shifts the industry’s focus toward cost and power efficiency.

AI infrastructure is rewriting the energy playbook. In this episode of the Data Center Frontier Show, Sage Geosystems CEO Cindy Taff joins DCF Editor-in-Chief Matt Vincent to discuss how the explosive growth of AI has shifted data center priorities from long-term decarbonization goals toward the immediate challenge of securing enough power, quickly and in the right location. Taff explains why aggregate generation capacity means little if electricity cannot reach a data center when it is needed. With transmission constraints and grid interconnection timelines stretching for years, hyperscalers and developers are increasingly exploring behind-the-meter generation and participating directly in energy infrastructure development. The conversation examines Sage Geosystems’ next-generation geothermal technology, which targets hot dry rock rather than the naturally occurring underground water resources required by conventional geothermal projects. By engineering subsurface reservoirs, Sage aims to make firm, dispatchable geothermal power available across a much wider geographic footprint. Taff also discusses how Sage combines geothermal generation with energy storage and pressure management, while addressing the water losses and high operating energy requirements associated with traditional enhanced geothermal systems. The episode explores geothermal’s evolution from a primarily “green” energy resource into strategic infrastructure for AI. Unlike intermittent renewables, geothermal can provide firm, 24/7 power and potentially be developed close to major loads. Taff argues that the oil and gas industry’s drilling expertise and existing infrastructure could provide a powerful scaling advantage. A hypothetical 5-gigawatt geothermal buildout over five years would require roughly 500 wells annually—a small fraction of current U.S. oil and gas drilling activity. The central takeaway: gigawatt-scale AI campuses may dominate the headlines, but they will still be built 50 to 100 megawatts at a time. For the next generation of data center development, Taff says the industry’s most important metric is becoming clear: time to power now matters more than cost or total capacity.

This conversation is about how the demands on data centers are changing and what that means for the systems that support them. As AI and high performance computing continue to scale, cooling is no longer a background function. It is central to whether these environments operate efficiently and reliably. Liquid cooling is becoming more common because it can handle the heat loads that air cooling cannot. But as systems move in that direction, the margin for error becomes much smaller. These are precision environments. Everything has to work as expected, and small issues can have larger consequences than people anticipate. Valves are a good example of something that is often overlooked but plays a critical role. They control flow, manage pressure, and help protect the integrity of the system. If they are not selected correctly, they can introduce problems that are difficult to detect early but show up later as inefficiencies or risk. One of the biggest points Eddie will make is that these systems depend on exact specifications. Engineers are not looking for something that is close. They need a valve that matches the system requirements exactly, whether that is flow performance, pressure characteristics, materials, connections, or physical dimensions. If something does not match, it can create integration issues, reduce efficiency, or delay the project. At the same time, the pace of data center construction is accelerating. Projects are moving quickly, and delays are not easily absorbed. That means availability and lead time are part of the technical decision, not just an operational detail. If the right solution is not available when it is needed, it creates risk for the entire build. This creates a real challenge for engineers, buyers, and OEMs. They need highly specific solutions, but they also need them delivered quickly and consistently. It is not enough to have a product that performs. The supplier has to be able to meet the spec, support the application, and deliver on time. Another important part of the conversation is how performance is evaluated. Published specifications do not always reflect real operating conditions. Systems do not run at a single point. They run across a range of flows and conditions. That is where the idea of usable Cv becomes important. It reflects how the valve actually performs in the system, not just how it performs in an ideal scenario. There is also growing awareness around hidden inefficiencies. Pressure drop, turbulence, and potential leak paths can all impact system performance. In high-density environments, these factors can reduce cooling effectiveness, increase energy usage, and introduce long-term reliability concerns. What this all points to is a shift in how components are selected. Valve selection is not a secondary decision. It is part of the overall system strategy. Getting it right helps protect uptime, maintain efficiency, and keep projects on track. Getting it wrong can introduce risks that are difficult and expensive to correct later. The goal of the conversation is to give people a clearer understanding of what matters most as they design and support modern cooling systems. It is about making better decisions upfront so systems perform the way they are intended to over time.