LONDON — In the high-stakes, hyper-accelerated world of artificial intelligence infrastructure, the traditional rules of commercial negotiation have been thoroughly rewritten. For decades, data centre development was defined by rigorous cost control, aggressive bidding wars, and the optimization of capital expenditure. Today, however, that playbook is obsolete.
According to industry leaders speaking at the Capacity Europe 2026 conference in London, the metric that dictates whether multi-million-dollar AI projects succeed or fail is no longer how much an operator is paying, but whether they can secure a manufacturing slot to get built at all.
During the pivotal "Scaling Under Scarcity: Delivering Projects Amid Shortages" panel on the AI Build Stage, engineering, procurement, and design experts warned that soaring demand, complex supply chain bottlenecks, and a frantic race for electrical power have triggered a fundamental shift. Speed of delivery has entirely eclipsed cost as the ultimate competitive advantage, forcing developers to order mission-critical infrastructure years in advance and completely invert traditional construction workflows.
Executive Overview
The rapid, unrelenting deployment of generative AI has thrust the digital infrastructure sector into an era of unprecedented scarcity. Hyperscalers, enterprise networks, and colocation providers are scrambling for physical real estate, high-density power, and specialized electrical equipment to support clusters of advanced graphics processing units (GPUs).
However, the physical world cannot scale at the speed of software.
As key equipment lead times stretch past the two-year mark and grid operators struggle to keep pace with soaring demand, data centre builders are adopting radical new strategies. From ordering backup generators before architectural designs are finalized to shifting from product-specific requirements to performance-based flexibility, the industry is transitioning into a posture of permanent crisis management.
At Capacity Europe 2026, panellists from Greenscale, Start Campus, Telehouse, and Kentik laid bare the realities of building under scarcity—highlighting a landscape where missing a ready-for-service (RFS) deadline carries far graver financial penalties than paying a premium for hardware.
Detailed Chronology: The Anatomy of a Supply Squeeze
The warning signs of the current infrastructure bottleneck have accumulated over several years, driven by the exponential leap in compute density required for large language models (LLMs) and foundational AI models.
The Long Shadow of Lead Times
By early 2026, the global supply chain for heavy electrical equipment had reached a breaking point. During the panel discussion on October 14, 2026, industry veterans detailed just how severely lead times have blown out.
- Generators: Backup power generation units now routinely require an ordering window of up to two years in advance. In some extreme instances, procurement teams face immediate, tight deadlines—such as committing to a multi-million-dollar generator order by the end of the week—merely to lock in a production slot that will materialize 24 months later.
- High-Voltage Transformers & Switchgear: Essential for stepping down and distributing power across hyperscale campuses, high-voltage transformers and gas-insulated switchgear now demand lead times of 12 to 18 months.
Ozgur Duzgunoglu, Head of Engineering and Design at Telehouse, highlighted the precarious nature of securing these components. In a market where a single hyperscaler can suddenly step in and scoop up hundreds of units in a single order, procurement teams cannot afford to wait. They must aggressively chase manufacturers simply to protect their fragile production slots from being pushed to the back of the queue.
The Six-Month AI Cycle vs. 18-Month Construction
Compounding the equipment crisis is a brutal temporal mismatch. Nelson Pinto de Magalhaes, Director of PMO and Procurement at Start Campus—the firm developing a massive hyperscale campus in Portugal—pointed out that the technological paradigm of AI shifts roughly every six months.
In stark contrast, even a fast-tracked, heavily optimized data centre takes approximately 18 months to build from the ground up. This temporal gap forces developers to design and order infrastructure for AI workloads that may evolve significantly before the facility’s doors even open.
Supporting Context & Metrics: The Mechanics of Scarcity
The bottlenecks facing the data centre industry extend far beyond factory-floor hardware. A convergence of grid constraints, skilled labor shortages, and evolving architectural demands has created a multi-layered matrix of risk.
1. The Grid Connection Bottleneck
While equipment shortages are deeply visible, power availability remains the ultimate governor of digital growth. Securing grid connections in tier-one European markets like Frankfurt can routinely take five to seven years, creating a systemic delay for operators eager to deploy capacity.
This friction has pushed some developers to extreme measures. Justin Ryburn, Field CTO and VP of Global Solutions Engineering at Kentik, noted instances in the United States where data centre customers run their diesel generators continuously before the grid connection is officially ready. Once local power finally arrives, the facility switches over to grid power, treating fossil-fuel generation as an expensive interim bridge to avoid delayed service launches.
The risks of grid failure are no longer theoretical. Oracle recently issued a high-profile force majeure notice regarding Project Jupiter—its massive Stargate campus in New Mexico—specifically to protect itself against potential delays in the site’s power supply ahead of its targeted 2028 operational launch.
2. The Labour and MEP Shortage
Equipment is only half the battle. As panellists emphasized, getting the hardware, the installation crews, and the commissioning teams to land in the exact same operational window is an immense logistical hurdle.
Mechanical, Electrical, and Plumbing (MEP) trades face a severe structural deficit. Clients, contractors, designers, and equipment suppliers are all competing for the exact same pool of specialized human talent. Steve Walton, CEO of 3C Datalec, recently echoed this sentiment, noting that the engineer and tradesperson shortfall is just as restrictive as the shortage of transformers and switchgear.
3. The Squeeze on Smaller Players
While hyperscalers possess the balance sheets to absorb price hikes and secure multi-year manufacturing blocks, smaller players face a radically different market dynamic. Kentik, which operates its own bare-metal infrastructure within leased data centre cages, has been informed by providers that space is increasingly being sold in bulk—entire floors at a time. Consequently, mid-sized technology firms are being forced to re-evaluate their architectures, considering whether to re-engineer their software to run natively on hyperscale clouds that already possess pre-built capacity.
Official Statements and Industry Insights
The paradigm shift was captured by candid insights from the industry’s leading practitioners during the Capacity Europe sessions:
-
On Shifting Priorities:
"We’re worried about cost and lead times, where I believe we should be worried about capacity."
— Miriam van Kooperen, Senior Director of Supply Chain, Greenscale -
On "Procure First, Design Around It":
"You can’t really compress a factory to produce more."
— Nelson Pinto de Magalhaes, Director, PMO and Procurement, Start Campus -
On the True Value of Time:
Each day saved in the delivery schedule brings revenue forward and dramatically improves returns on borrowed capital. Cost frequently becomes secondary to hitting the target date.
— Ozgur Duzgunoglu, Head of Engineering and Design, Telehouse
The New Mantra: Procure First, Design Around It
To survive this climate, leading developers are completely restructuring how they contract with suppliers. Historically, procurement followed a linear path: design the facility, draft precise specifications, solicit competitive bids, and negotiate the lowest price.
Today, that model has collapsed. Greenscale’s Miriam van Kooperen explained that operators are now forced to secure manufacturing capacity "without knowing exactly what it is we will be buying." This requires establishing preferred vendor lists early, setting strict cut-off dates where designs must be prematurely frozen, and prioritizing production slots above commercial discounts.
Start Campus has taken this philosophy a step further by elevating procurement above design. By integrating their teams so that procurement leads the charge, they have adopted the core corporate mantra: "Procure first, design around it."
Rather than viewing financial savings through the lens of unit cost, progressive project managers are evaluating savings in terms of days. In an AI economy where speed-to-market generates tens of millions in early monetization, saving weeks on a construction schedule easily outweighs minor savings on equipment invoices.
Strategic Adaptation: Standardisation and Prefabrication
To mitigate the risks of single-vendor dependencies and rigid supply chains, engineering teams are rethinking how data centres are conceptualized and built.
Performance-Based vs. Product-Based Specifications
In the past, design teams insisted on rigid, product-specific equipment lists. Van Kooperen noted that she once championed strict product standardisation, but the realities of supply chain volatility have forced a pivot toward performance-based specifications. By defining the performance required rather than the exact brand of the hardware, operators can seamlessly swap out components or pivot to alternative suppliers when a preferred vendor runs out of factory capacity.
Geographic and Modular Flexibility
Telehouse applies a modular design philosophy, utilizing standard layouts with flexible sub-packages that can be deployed anywhere from Frankfurt and Paris to international markets like Thailand. However, they maintain relationships with four or five primary suppliers for core packages, acknowledging that local supply chains dictate reality—a UPS manufacturer viable in Europe may not be the optimal choice in Tokyo or Osaka.
Furthermore, prefabrication has transitioned from a cost-saving measure to a time-saving imperative. Coolant distribution units (CDUs) for advanced liquid-cooling systems are now routinely manufactured off-site and delivered for rapid plug-and-play installation. While factory testing helps iron out flaws, compressed build timelines mean that commissioning phases are frequently squeezed into narrow windows of a few short weeks.
Future Outlook: Power, People, and 2029
Looking three years ahead to 2029, the consensus among industry leaders is sobering. The structural bottlenecks currently choking AI infrastructure delivery are not temporary post-pandemic anomalies; they are the new structural baseline of the digital economy.
- Power as the Permanent Constraint: Experts expect electrical power—both generation and transmission—to remain the primary bottleneck for the foreseeable future. The clean energy transition, combined with the insatiable power draws of next-generation AI accelerators, will continue to strain national grids.
- The Talent War: The scarcity of skilled labour will intensify. As data centre developments expand into secondary and tertiary locations to chase available land and power, those regions will frequently lack experienced local construction trades. The competition for MEP engineers, commissioning specialists, and project managers will remain fierce.
- Location as a Strategic Lever: Geographic arbitrage will become increasingly critical. Developers who successfully identify and secure regions with pre-existing, robust power infrastructure will bypass the multi-year grid queues that currently plague major European and North American hubs.
Conclusion
As Capacity Europe 2026 made abundantly clear, the AI era has permanently altered the economics of digital infrastructure. In a market where manufacturing lines are running at full tilt and lead times stretch across years, financial capital is no longer the scarcest resource—time and production capacity are. For data centre developers navigating this high-pressure landscape, the mandate is clear: secure your manufacturing slots, lock in your power, and design around the constraints, or risk being left behind in the global race for artificial intelligence dominance.
