This article was originally published by WisdomTree on 25 Jun 2025. Written by Christopher Gannatti.

‍

Key Takeaways
‍

• As public cloud infrastructure players ramp up capital expenditures (projecting nearly $400 billion in 2025 alone), data centre infrastructure is emerging as a critical and underappreciated pillar of the AI economy.
• Amid macroeconomic uncertainty and big tech volatility, data centre REITs like Equinix and Digital Realty offer focused exposure to the physical compute layer, with rising demand from AI-native tenants.
  ‍

As questions emerge about the durability of the artificial intelligence (AI) megatrend – and as policy makers in Washington and Beijing escalate tariff threats that could reshape global tech supply chains – investors are increasingly debating whether the infrastructure build-out that defined the past two years will persist. While enthusiasm once cantered around semiconductors, foundation models and application layers, attention is shifting. Beneath the surface of these innovations lies something more tangible: the concrete, cooling and cabling of the modern data centre.
‍

Despite being the physical foundation of every AI training run and real-time inference, data centre real estate investment trusts (REITs) have remained second-tier considerations for many investors. Why consider Digital Realty or Equinix when you can buy Nvidia or Microsoft? Yet that framing may be shortsighted.
‍

Heading into the second half of 2025, data centre infrastructure stands not only as an essential piece of the digital economy, but also as a potential countercyclical opportunity. As hyperscalers¹ recalibrate their strategies, the facilities housing their compute clusters may offer more durable exposure to long-term AI demand than many expect. The next phase of AI investment may be less about speculative software and more about securing physical capacity, bandwidth and power.

‍

The invisible engine of AI
‍

Generative AI is devouring compute at a historic rate. Microsoft's Azure, Google Cloud and Amazon Web Services are all projecting unprecedented capital expenditures to support next-generation workloads. It'll be interesting for all of us to monitor how or if these figures change in the face of shifting economic tides in different countries:
‍

• Microsoft plans to spend ~$80B in FY25, up from $53B two years prior, with most of that targeting data centre capacity and AI-specific infrastructure. Importantly, Microsoft favours "fungible" CapEx that can support multiple use cases and prefers to own its infrastructure—but leases powered shells when needed.²
• Amazon is guiding for ~$100B in CapEx for 2025, with the majority directed toward Amazon Web Services (AWS), which crossed $100B in annual revenue in 2024.³
• Alphabet plans to spend $75B in 2025, citing a near-term shortage of AI data centre capacity as its primary motivation.⁴

Nvidia CEO Jensen Huang underscored this urgency at the GTC 2025 conference: Demand for AI compute has grown "100 times more than we thought we needed this time last year" due to the rise of agentic AI. Looking ahead, he projected that by 2028, annual capital expenditures on AI compute infrastructure could exceed $1 trillion – underscoring not just current scarcity but the scale of investment that may be required to keep pace with accelerating model complexity and global deployment.⁵

‍

So why not just own the hyperscalers?
‍

It's a fair question. Microsoft, Amazon and Alphabet dominate AI infrastructure, and they benefit from vertical integration, global reach and embedded distribution. But from an investment perspective, their data centre exposure is buried under other business lines – ad tech, retail, devices and software. For example – we have been watching the headlines regarding Alphabet's monopoly in advertising, its second such verdict.⁶ It's an important and developing story for Alphabet, but it is a distraction from the overall growth of data centres.

In contrast, data centre REITs and infrastructure developers offer something different: pure-play, metered exposure to the compute layer.

‍

Data centre REITs: Evolving for the AI age
‍

Far from passive landlords, REITs like Equinix (EQIX) and Digital Realty (DLR) are repositioning themselves to serve high-density AI workloads:

• Equinix has logged 22 years of consecutive quarterly revenue growth and is now targeting $9B in annual revenue. Through its xScale joint venture, it has leased more than 400 MW of hyperscale capacity to cloud providers while maintaining its premium retail colocation franchise.⁷
• Digital Realty has pivoted toward a full-spectrum model, combining hyperscale campuses with high-density colocation. Its recent land acquisitions in Texas and North Carolina are designed for 100MW+ scale developments. In Q4 2024 alone, DLR signed 166 new customers.⁸
• Iron Mountain grew its data centre revenue 25% in 2024 and is selective in hyperscale deals – it passed on a 130MW lease that didn't meet return on investment (ROI) thresholds, underscoring the financial discipline now shaping the market.⁹

These REITs are now investing in liquid cooling systems, graphics processing unit (GPU)-ready pods and modular expansions. Their tenant base increasingly includes AI labs, GPU startups and enterprise inference deployments.

‍

The new middle layer: CoreWeave and the rise of the AI cloud specialists
‍

Somewhere between the hyperscalers and REITs sits a rising breed of "AI-native" cloud infrastructure firms. The most prominent is CoreWeave:¹⁰

• Originally a crypto miner, CoreWeave now leases GPU infrastructure and operates more than 250,000 Nvidia GPUs across 32 data centres.
• Its revenue grew 737% year-over-year (YoY) in 2024 to $1.9B. Microsoft alone accounted for 62% of that revenue.
• It is targeting a $32B–35B initial public offering (IPO) and has raised more than $12B in debt and equity to fund expansion.

CoreWeave's approach is emblematic of a new market niche: hyperscalers and AI developers leasing GPU-rich capacity from specialized providers. The economics are risky, as CoreWeave carries high debt and client concentration, but the model proves there's room for vertical scaling outside of Big Tech.

‍

A global build-out: Segro, SoftBank and the new geography of compute
‍

• Segro, the UK REIT known for logistics, has entered the data centre space via a £1B joint venture with Pure Data Centres. This is their first "fully fitted" facility, designed to host AI workloads and pre-let to a hyperscaler.¹¹
• OpenAI and SoftBank are planning to invest up to $500B in new AI data centres globally via the Stargate project, starting in Texas and potentially expanding to the UK depending on regulatory and energy availability.¹²
  
• Huawei's CloudMatrix 384, powered by its Ascend 910C chips, is China's answer to the Nvidia GB200 platform. While less efficient per watt, it shows the global arms race in AI infrastructure is not constrained to the West.¹³

The result? Data centres are not just a U.S. or hyperscaler story. They are now a global competitive asset class.

‍

Conclusion: Owning the compute layer
‍

AI's future will be built in concrete and cooled by water. Whether via hyperscaler CapEx or REIT expansions, capital is flooding into the foundational layer of compute.

Data centre REITs offer a unique and underexposed angle: reliable cash flow, long-term leases and rising pricing power from AI-optimized capacity. For investors looking to own the infrastructure fuelling the AI economy – without the baggage of ad tech or retail – these assets may deserve a premium, not a discount. In short, don't overlook the real estate beneath the revolution. It's where the future gets built.
‍

¹_{Term used to reference the biggest computing infrastructure providers, most often Microsoft, Amazon and Alphabet.}
²_{Source: Rangan et al., "Investing aggressively ahead of AI opportunity while remaining prudent, focused on long-term ROI - MSFT NDR takeaways," Goldman Sachs Research, 24/03/25.}
³_{Source: https://www.aboutamazon.com/news/company-news/amazon-ceo-andy-jassy-2024-letter-to-shareholders}
⁴_{Source: Kendrick Cai, "Alphabet reaffirms $75 billion capital spending plan in 2025 despite tariff turmoil," Reuters, 09/04/25.}
⁵_{Source: Tae Kim, "Nvidia's Disconnect: An Improving Business With a Cheaper Stock," Barron's, 21/03/25.}
⁶_{Source: Tae Kim, "Google Is Ruled a Monopolist for Second Time in Latest Federal Trial," Barron's, 17/04/25.}
⁷_{Source: https://www.equinix.com/newsroom/press-releases/2025/02/equinix-reports-strong-fourth-quarter-and-full-year-2024-results}
⁸_{Source: https://www.digitalrealty.com/about/newsroom/press-releases/123308/digital-realty-reports-fourth-quarter-2024-results}
⁹_{Source: Dan Swinhoe, "Q4 2025 data center colocation results: Digital Realty, Equinix, and Iron Mountain," Data Center Dynamics, 27/02/25.}
¹⁰_{Source: Tabby Kinder & Robert Smith, "CoreWeave tests investor risk appetite with $7.5bn in looming debt repayments," Financial Times, 21/03/25.}
¹¹_{Source: Joshua Oliver, "Segro partners with Oaktree-owned Pure on 1 billion GBP data centre in London," Financial Times, 3/25/25.}
¹²_{Source: Hammond et al., "OpenAI and SoftBank weigh UK investment for Stargate AI project," Financial Times, 17/04/25.}
¹³_{Source: Patel et al., "Huawei AI CloudMatrix 384—China's Answer to Nvidia GB200 NVL72," SemiAnalysis, 16/04/25.}

‍
Le informazioni contenute nel sito mexem.com hanno uno scopo puramente informativo. Non devono essere considerate come consigli di investimento. L'investimento in azioni comporta dei rischi. La performance passata di un titolo non è un indicatore affidabile della sua performance futura. Consultare sempre un consulente finanziario o fonti fidate prima di prendere qualsiasi decisione di investimento.
‍
‍

Read the original article on WisdomTree.
‍