Solving the hidden bottlenecks of AI growth

Harqs Singh at InfraPartners explores the challenges associated with the supply chain and how prefabricated data centres are the way forward

 

AI is rapidly evolving and the pressure is on to build infrastructure at pace. Billions are being invested in AI-ready data centres, but there’s a looming risk that today’s deployments could quickly become outdated. The expected lifespan of data centre infrastructure is shrinking rapidly, with some facilities needing major upgrades within five years due to the speed of AI hardware evolution.

 

This raises an important question about how long these assets will provide strong returns, highlighting the need for infrastructure that not only performs today but remains flexible and upgradeable over time.

 

According to McKinsey, global demand for AI-ready data centres is set to grow by 33% annually through 2030, presenting enormous commercial potential.  As well as this, around 70% of projected data centre growth through 2030 is expected to be driven by AI, with a shift from training-heavy workloads today to inference-heavy demands as we get closer to 2030. For investors, infrastructure operators and technology providers alike, there is an urgent need to reassess where the bottlenecks to growth lie and what must be done to address them.

 

As the demand for AI infrastructure increases, it’s clear that key systems like data centres, power grids, and skilled labour need to evolve to keep pace with the technology and the growth required.

 

 

Labour and power challenges

Due to the rapid growth of the industry, the shortage of skilled labour capable of designing, installing and maintaining high-performance GPUs is intensifying. Hiring expertise in these fields can be difficult, particularly in emerging or remote regions where power is available today. As a result, many companies are heavily reliant on bringing in foreign, skilled labour which drives up project costs. Inevitably, this raises questions about long-term scalability and labour sustainability.

 

 At the same time, companies are facing supply chain challenges when sourcing critical components. Generators, transformers and power equipment are in such high demand that lead times often stretch from months into years. Wood Mackenzie reports that transformer lead times rose from around 50 weeks in 2021 to 120 weeks on average in 2024.

 

These delays, amid ongoing geopolitical tensions and shifting trade policies, are putting project delivery timelines at risk. The most significant challenge is the power itself. In many regions, grid capacity is either constrained or located in remote areas where a supply chain does not exist. 

 

These challenges highlight that the traditional approach to data centre construction needs to be re-examined. These builds are slow, expensive and geographically rigid, locking operators into fixed designs that are not flexible enough to meet future demand. Major cloud providers are already adjusting expectations, shortening hardware lifecycles and retiring AI servers earlier than planned to keep up with performance demands.

 

For investors, this is a growing risk. A billion-dollar data centre that can’t be upgraded without major disruption becomes a financial liability and has significant asset value risk.

 

 

Building for flexibility, not just scale

To address these challenges, the industry must adopt a new approach - one that emphasises speed, upgradeability, and scalability. Prefabricated data centres are emerging as a solution. These designs offer the flexibility and capital efficiency needed, all while condensing construction timelines. In addition, they support on-site power generation and integration of renewable or alternative energy sources, such as fuel cells.

 

Offsite manufacturing allows complex technical systems to be designed, tested and assembled under controlled conditions, reducing on-site risk, streamlining delivery and ensuring consistency across all deployments. Also, prefabrication allows for a parallel construction process. Site preparation can happen whilst the modules are constructed, accelerating deployment schedules. Prefabricated units and strategic warehousing are increasingly being used to bypass supply chain disruptions amid shortages of critical components and skilled labour onsite. Through this approach, operators can ship modules to locations where power is available, removing some of the logistical and labour constraints that often stall traditional builds.

 

By designing facilities that can be easily upgraded in terms of cooling technologies or power delivery, operators can preserve asset value, extend their useful lifecycle, and meet the needs of future technology. This approach is already gaining traction. Companies are delivering flexible, AI-optimised capacity quicker than before, enabling AI capabilities in constrained data centre markets.

 

 

Build fast, evolve faster

AI infrastructure is entering a new era where flexibility is the real competitive advantage. Those who will be leaders in the industry won’t only be the biggest or most specialised but will be those who can deploy the fastest. Agility is crucial as businesses demand infrastructure that can scale or pivot with shifting infrastructure needs. Governments fast-tracking sovereign AI and investors pushing for capital efficiency highlight the need for upgradeable, future-proof designs that can keep up with rapid refresh cycles.

 

To unlock AI’s full potential, we must think of data centres as intelligence factories, not just static buildings that hold servers.  It will be important to build and evolve in an agile way without costly downtime. This means integrating next-generation power and cooling, adapting to fluctuating energy profiles, making data centres grid-aware, and accommodating future hardware. AI’s future will depend on how we build and what we build with.

 

In a market where high demand drives an increase in performance and revenue per token, upgradeable infrastructure is no longer a nice-to-have; it’s the foundation for sustained returns. The companies that design with flexibility at their core will be the ones blazing the trail.

 

---

 

Harqs Singh is CTO at InfraPartners

 

Main image courtesy of iStockPhoto.com and quantic69