Jun 12, 2025 · 36m · catalyst
The state of play of data center development
0:00 / 0:00
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
Host Shail Khan interviews Digital Realty CTO Chris Sharp to explore the realities of data center development during the AI boom, contrasting speculative industry hype with genuine grid bottlenecks, high-density engineering, and utility partnerships.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Shayle holds 33.4% of the talking time here. How this is scored →
Shayle as informed peer 5.9 Guest teaching 5.5 Guest disagreement 1.2 Shayle pushing back 1.4
speaking balance: gold is Shayle, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 22:17 Rejection of distributed small-footprint data centers
Chris firmly pushes back on Shayle's hypothesis that developers could easily split large gigawatt workloads into dozens of distributed 20-megawatt data centers, citing operational complexity and multi-model peak capacity needs.
Hardest push from Shayle ▶ 15:59 Challenging the equipment lead time bottleneckShayle drills into equipment supply chain concerns, questioning whether transformer lead times actually delay projects given that utility interconnection queues take even longer.
Biggest teaching moment ▶ 31:35 Liquid cooling thermal dynamics enforce backup powerChris educates Shayle on the physical constraints of high-density chips, explaining that liquid-cooled hardware requires three-nines mechanical reliability because cooling pump failures instantly cause billions in hardware damage.
Shayle holds their own ▶ 28:43 Deconstructing behind-the-meter generation strategiesShayle demonstrates deep power sector fluency by clearly delineating between islanded off-grid bridge power and negotiated utility behind-the-meter interruptible tariff structures.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Shayle as informed peer | Guest teaching | Guest disagreement | Shayle pushing back | Why |
|---|---|---|---|---|---|---|
| Geographic Concentration and Cloud Availability Zones | 6 | 5 | 1 | 1 | Shayle probes whether the new wave of AI development is dispersing data centers away from traditional hubs. Chris explains that while frontier model training expanded regionally, inference workloads pull infrastructure back to established cloud availability zones. | |
| AI Workloads, Throughput Constraints, and Data Gravity | 7 | 6 | 1 | 2 | Shayle asks whether non-latency-sensitive inference could be sited in remote areas with cheap stranded power. Chris explains that high data throughput requirements, multi-model reasoning pipelines, and data gravity make tier-one proximity necessary. | |
| Master Planning and Regional Grid Bottlenecks | 5 | 5 | 1 | 1 | Shayle asks how close prime regions like Northern Virginia are to tapping out of power. Chris shares Northern Virginia vacancy figures and explains the five-year master planning required to coordinate large gigawatt-scale additions with utility distribution grids. | |
| Development Timelines and Supply Chain Lead Times | 7 | 5 | 1 | 3 | Shayle asks whether long utility interconnection queues render long transformer lead times moot as a bottleneck. Chris acknowledges the queue duration while outlining Digital Realty's proactive vendor-managed inventory and noting gas turbine backlogs extending to 2029. | |
| Data Center Densification and Sizing Realities | 5 | 6 | 2 | 1 | Shayle explores the trend toward massive gigawatt data center announcements. Chris calls out speculative 'braggawatts' hype, explaining that actual customer requirements vary between contiguous 100MW GPU clusters and modular 5MW enterprise inference blocks. | |
| Designing Contiguous Mega-Halls for Peak AI Demand | 6 | 7 | 3 | 2 | Shayle asks whether building distributed twenty-megawatt sites is viable to ease grid interconnection bottlenecks. Chris pushes back, noting that multi-model inference pipelines and peak reliability requirements necessitate contiguous large-scale halls. | |
| Navigating Interconnection Queues and Speculative Bubbles | 6 | 4 | 1 | 1 | Shayle highlights how speculative interconnection requests overwhelm utilities trying to separate real projects from noise. Chris agrees, outlining how established developers maintain utility credibility through creditworthy customer ramp commitments. | |
| Evaluating Bridge Power and Behind-the-Meter Generation | 7 | 5 | 1 | 2 | Shayle distinguishes between temporary off-grid generation and negotiated behind-the-meter utility agreements with interruptible tariffs. Chris validates Shayle's framing, noting developers prefer structured utility partnerships over becoming independent power producers. | |
| Backup Power Architectures, SLAs, and Liquid Cooling Demands | 6 | 8 | 1 | 1 | Shayle asks if data center architectures can drop diesel backups by relying on software failovers. Chris reveals that high-density liquid cooling demands triple-redundant reliability because cooling pump failures immediately ruin costly accelerators. | |
| Innovations in Efficiency, Next-Gen Hardware, and AI's Societal Impact | 4 | 4 | 0 | 0 | Chris outlines upcoming advancements in Power Usage Effectiveness, next-generation chip architectures with Nvidia, and real-world AI applications like the Gefion supercomputer in Copenhagen. |
Statements from this episode (17)
Sharp: AI inference, not training, will drive future data center growth
“I would say that there's been a lot of growth in that, but we see that kind of leveling out where we really see the consumption of AI or inference that's driving that regional specific growth going forward. And I think that's where it's more embedded. And a lo…”
Sharp: Text-to-text AI workloads can service North America from three markets
“Yeah, we see a lot of text to text scenarios where that workload can be deployed, you know, in two or three markets throughout North America and service the entire market.”
Sharp: Advanced reasoning models force AI infrastructure into tier-one hubs
“As you see bimodal, some of these more advanced reasoning models where a token isn't just generated against a prompt, and then you consume it, and it's done. A token may be generated inside of an AI world and go through multiple models to ensure that, you know…”
Sharp: Embedding inference into training clusters boosts AI chip utilization and ROIC
“The expense to stand this infrastructure up is, these chips are not cheap, right? And so, being able to utilize that over a longer horizon of workload and driving that utilization is also a form factor of, if I have it installed for training, and training can …”
Sharp: Northern Virginia data centers have a 0.5% vacancy rate
“I mean, some of the most recent stats, it has .five percent vacancy rate, which is phenomenal.”
Digital Realty is building a gigawatt-scale data center by Dulles Airport
“When we say we're going to need a gigawatt, like with what we're building right now, right next to the Dulles Airport, that they have an understanding of that power requirement”
Sharp: Transformer lead times for data centers exceed 50 weeks
“Some of the, you know, broader infrastructure constraints are transformer lead times are 50 plus weeks right now.”
Sharp: Gas turbine delivery backlog extends past 2029
“Some markets were always investigating, you know, different types of, you know, solutions via gas turbines and, Even those are backlog plus 20, 29, right?”
Sharp: Gigawatt announcements are often just 'braggawatts' ignoring real demand
“There's a lot of noise, like one of the things we've been joking about is a lot of braggawatts. Oh, I have a gigawatt. I have a gigawatt. And there's just so much noise out there that you really want to get underneath the workload and the durability of the com…”
Sharp: AI inference deployments require 5-megawatt blocks, unlike training arrays
“What we're really seeing is inference. Can come in, in like, five-ish megawatt blocks, and you can solve for it a bit differently. Now, the densification is still there, and then the private AI pieces, there's hot spots where it can be, you know, a couple of m…”
Sharp: Hyperscalers prefer contiguous 100 MW halls over distributed mini-centers
“That's the outlier of what we're seeing is that to operationalize, they would really like it to be more of a contiguous scenario where they do logically think about them as five megawatt chunks, and they want a bit of resiliency... What we're seeing is having …”
Sharp: Power and financing do not exist for top AI growth projections
“Where, you know, the power, the amount of power, and even the amount of financing required to meet these upper end projections, it doesn't exist, right?”
Sharp: Data centers built for spiky training runs will face major failures
“There's going to be some probably written about very big challenges and failures where they wanted a total capacity block, but they couldn't support the densification, or they were just building for the spike. It was very spiky, but the longer term utilization…”
Sharp: Pure off-grid bridge power remains an outlier in data centers
“Yeah, I think there's some outliers. Very few of it gets covered in the press because nobody wants to really go on to the market where you, the grid can't meet the customer ramp demands today. I mean, full stop... But yeah, it is an outlier.”
Sharp: Behind-the-meter data center power is merely a temporary stopgap
“Doing behind the meter, you know, becoming a power generation capability, I think is a short term gap. Nobody would be investigating that if we didn't have the constraint.”