The Next Phase of Data Center Growth

AI-Driven Workloads, Investment Pressure and Delivery Risk

The rise of AI is fundamentally reshaping the computer era, driving an escalation in workload demands and redefining data center (‘DC’) requirements. Historically, enterprise computing evolved from on-premise mainframes to physically concentrated infrastructure and ultimately to the cloud, with hyperscalers dominating the landscape. Today, AI democratization and the infusion of AI into cloud workloads are triggering a change: datacenters must now deliver far higher compute densities, tighter total cost of ownership (‘TCO’) and increasingly latency-sensitive architectures. As a result, facility sizes continue to scale, with next-generation AI campuses reaching hundreds of megawatts or even gigawatt-class footprints. Compared to traditional cloud or colocation workloads, AI’s compute and power requirements introduce new pressures on infrastructure design, siting and economics.

Meeting surging AI demand requires unprecedented near-term capital expenditure, with required global DC investments set to outpace hyperscalers’ own annual CapEx trajectories. By 2027–2029, required AI-driven CapEx is projected to exceed hyperscaler CapEx by 140–160%, highlighting the need for new capital pools and accelerating the rise of alternative infrastructure providers.

Two trends are emerging:

Outsourcing at the chip layer is fueling the growth of “NeoClouds” or GPU-as-a-Service platforms — ranging from giants to regional emerging players, sovereign clouds and legacy B2B providers expanding into GPUs.¹
Speed-to-market is becoming a decisive competitive moat as platforms race to secure land, power and customer commitments before supply tightens further.

Despite rapid development, structural constraints — utilities and investors demand customer commitments to allow operators access to finance and power — limit the risk of oversupply and speculative build. Four critical areas determine the achievability of “ready-for-service” (‘RFS’) timelines: planning & permitting, power procurement and delivery, construction execution and customer contract visibility. Power is often the principal bottleneck, prompting operators to explore alternatives beyond traditional grid connections, including natural gas, nuclear options and batteries to bypass grid constraints and accelerate deployment.

Meanwhile, colocation and cloud markets continue to evolve alongside AI. Hybrid cloud strategies are expanding as enterprises balance public cloud scalability with the need to manage specific workloads locally, while sovereign cloud propositions are becoming nationally strategic. Retail colocation also stands to absorb demand from industries with regulatory, latency or deployment-specific requirements

In a rapidly shifting market, investors and operators need to ask the right questions is key to assessing platform risk and value.

What is the strength of the tenant contract and counterparty risk borne by the datacenter operator?
How realistic is each site’s RFS timeline, and how does it stack up against market build-outs over the next 24 months?
As chipsets evolve, what risk of obsolescence does the DC infrastructure face – and what upgrade capex is required?
Where is demand emerging, and from which locations can it be served most efficiently across workloads?
What is the likely price erosion as new hardware generations are released?
How will alternative power sources reshape workload siting and when will they start influencing operators’ buying criteria?

Ultimately, only those investors and operators who put their mind to these questions will be positioned to accurately price risk, capture emerging value and remain competitive as the data center landscape undergoes its next wave of transformation.

Click here to download the full report

Footnotes:

1: “GPU-as-a-Service: AI Demand Reshapes Data Centers and Fuels NeoCloud,” FTI Consulting (October 15, 2025) Growth