Why AI Infrastructure Is the Defining
Investment of the Decade

The $5T Infrastructure Supercycle

The AI industry has entered an unprecedented capital expenditure cycle. In 2025 alone, the five largest hyperscalers announced combined AI infrastructure spending exceeding $300 billion — a figure that would have been dismissed as fantasy three years prior. Microsoft has committed to $80 billion in annual AI data centre investment. Google has earmarked $75 billion. Meta is spending $65 billion. Amazon and Oracle are each deploying tens of billions more.¹

This is not a bubble. It is a structural repricing of compute as a strategic asset. Every enterprise AI deployment — from customer service chatbots to autonomous vehicle systems to drug discovery platforms — requires physical infrastructure: servers, networking, power delivery, and thermal management. The software layer captures attention; the infrastructure layer captures value.

For investors, the critical insight is the divergence between AI software and AI infrastructure economics. Software companies face commoditisation risk as open-source models proliferate and API pricing compresses. Infrastructure providers, by contrast, benefit from physical scarcity: you cannot download a megawatt or 3D-print a data centre. The barriers to entry are measured in billions of dollars, years of permitting, and grid connections that cannot be accelerated with venture capital.

Why Europe Is Different

The global AI infrastructure buildout is concentrated overwhelmingly in the United States. Approximately 60% of hyperscale data centre capacity is located in North America.² This geographic concentration creates a structural vulnerability for European enterprises, governments, and critical infrastructure operators who depend on AI capabilities but lack sovereign compute capacity.

Europe is responding. The regulatory and investment landscape has shifted dramatically in 2025–2026:

• EU AI Act:The world's first comprehensive AI regulation creates compliance requirements — transparency, auditability, data governance — that favour providers operating within EU jurisdiction with full regulatory alignment.
• InvestAI Initiative: The European Commission has announced a €20 billion facility to accelerate AI infrastructure investment across the EU, including direct support for sovereign data centre capacity.³
• CADA Framework: The Cloud Authority Data Act framework introduces sovereignty tiers for cloud infrastructure — sovereign, semi-sovereign, and non-sovereign — creating formal classification requirements that non-EU providers cannot easily satisfy.
• National Programs: France has committed €2.5 billion to its national AI strategy. Germany has allocated €3 billion for AI compute infrastructure. The Netherlands, Nordics, and Mediterranean nations are following with their own sovereign compute initiatives.

For investors, Europe's regulatory environment creates what is effectively a compliance moat. Non-EU providers — particularly US hyperscalers subject to the CLOUD Act — face structural disadvantages in serving European customers with sovereignty requirements. This is not a temporary market inefficiency; it is a regulatory design feature that will deepen as AI becomes classified as critical infrastructure.

The Power Bottleneck

Energy is the single greatest constraint on AI infrastructure deployment globally. A modern AI data centre consumes 50–100 MW of power — equivalent to a small city. The world's largest planned AI facilities will consume over 1 GW. Demand for data centre power is growing at 25–30% annually, while grid capacity expansion in most developed markets grows at 1–3%.⁴

The result is a global power bottleneck. In Northern Virginia — the world's largest data centre market — grid connection wait times now exceed four years. Similar constraints exist across Ireland, the Netherlands, and the UK. Developers with access to power are sitting on the most valuable asset in the AI economy.

The Mediterranean region offers a structural advantage. Countries like Cyprus, Greece, Spain, and Portugal combine three critical factors:

• Abundant solar irradiance: 300+ days of sunshine per year enabling cost-effective behind-the-meter solar generation. A Bring Your Own Power (BYOP) model — where the data centre operator deploys co-located solar — can reduce energy costs by 40–60% compared to grid-only power in Northern Europe.⁵
• Coastal cooling advantages: Proximity to the Mediterranean enables seawater-assisted cooling systems, reducing the energy overhead of thermal management by up to 30% compared to inland facilities.
• Grid availability: Mediterranean markets have significantly less data centre density than Northern Europe, meaning grid connections are available on faster timelines — months rather than years.

RISC-V: The Open Hardware Revolution

The AI infrastructure stack has a single point of failure: NVIDIA. With an estimated 80–95% share of the data centre AI accelerator market, NVIDIA's dominance creates supply-chain concentration risk, pricing power asymmetry, and exposure to US export controls that can restrict hardware availability with 90 days' notice.⁶

RISC-V — an open-source instruction set architecture governed by RISC-V International, a Swiss-domiciled non-profit — offers investors a fundamentally different risk profile:

• No export control exposure: The RISC-V ISA is open and internationally governed. While specific chip implementations may involve controlled manufacturing, the architecture itself is not subject to US BIS licensing restrictions.
• No vendor lock-in: Multiple companies — Tenstorrent, SiFive, Ventana Micro, and others — are building RISC-V silicon. This creates competitive dynamics that benefit buyers, unlike the NVIDIA monopoly.
• Full auditability: Open-ISA silicon can be audited at the instruction-set level, satisfying sovereignty requirements that proprietary architectures cannot.

The investment community is recognising this shift. SiFive closed a $400 million Series G in April 2026 at a $4.5 billion valuation. Tenstorrent, led by legendary chip architect Jim Keller, has raised over $700 million and is shipping the Galaxy inference server platform.⁷

For investors, RISC-V is not a speculative bet on an unproven technology. It is a structural shift in the economics of AI compute, backed by billions in venture funding, a maturing ecosystem, and regulatory tailwinds that favour open, auditable hardware over proprietary alternatives.

The Investor Checklist: Evaluating AI Infrastructure

Not all data centres are created equal. The transition from traditional enterprise hosting to AI-native infrastructure represents a fundamental change in facility economics, tenant profiles, and return characteristics. Sovereign AI infrastructure adds another layer of differentiation. The following comparison framework helps investors evaluate opportunities across the spectrum.

Sovereign AI infrastructure commands premium multiples for three reasons: longer contract durations reduce revenue volatility, regulatory compliance requirements reduce competitive pressure, and the strategic nature of sovereign compute makes these assets attractive acquisition targets for governments and civil-resilience-adjacent entities.

Return Profile: Infrastructure Stability with Technology Upside

AI infrastructure investments offer a distinctive return profile that combines the stability characteristics of traditional infrastructure with the growth upside of the technology sector. For institutional investors — pension funds, sovereign wealth funds, family offices — this hybrid profile addresses a portfolio gap that neither pure tech equity nor traditional infrastructure fills.

Contracted Revenue Base

Sovereign AI data centres operate under long-term contracts — typically 3–10 years — with government entities, civil protection organisations, and regulated enterprises. These contracts provide predictable cash flows similar to traditional infrastructure concessions, with annual escalation clauses typically linked to inflation indices or energy cost pass-throughs.

Inflation-Linked Pricing

Unlike software businesses where pricing power erodes with competition, AI infrastructure benefits from physical scarcity. As demand for compute grows faster than supply, pricing power accrues to operators with deployed capacity. Energy costs — the largest variable expense — are typically passed through to tenants, creating a natural inflation hedge.

Data Sovereignty Premium

Sovereign AI infrastructure commands a 20–40% pricing premium over non-sovereign alternatives, reflecting the compliance value, jurisdictional certainty, and supply chain independence that customers require. This premium is structural — driven by regulation, not sentiment — and therefore more durable than typical technology pricing advantages.⁹

Multiple Exit Pathways

AI infrastructure assets offer multiple exit strategies for investors:

• Strategic acquisition:Hyperscalers, telecoms, and sovereign wealth funds are actively acquiring data centre platforms. Recent transactions include Blackstone's $16 billion QTS acquisition and DigitalBridge's $11 billion Switch deal — both at 25–30× EBITDA multiples.
• Infrastructure REIT conversion: Mature data centre portfolios can convert to REIT structures, unlocking public market valuations and institutional capital at scale.
• Government concession: Sovereign AI assets may be acquired or concessionised by national governments as strategic infrastructure, at premium valuations reflecting strategic infrastructure value.

Frequently Asked Questions

What returns can I expect from AI infrastructure investments?

AI infrastructure investments typically target 15–25% unlevered IRR over a 5–7 year hold period, with stabilised cash-on-cash yields of 8–12% once facilities reach operational maturity. Sovereign AI assets may command higher returns due to the data sovereignty pricing premium (20–40% above non-sovereign alternatives) and longer contract durations that reduce re-leasing risk. Exit multiples for AI-native data centres have ranged from 20–35× EBITDA in recent transactions, significantly above traditional data centre multiples of 15–20×.

How does sovereign AI infrastructure differ from traditional data centers?

Sovereign AI data centres differ in three fundamental ways. First, they guarantee data residency and jurisdictional control — all data processing occurs within a specific legal jurisdiction (e.g., the EU) with no exposure to foreign government access requests such as those enabled by the US CLOUD Act. Second, they use auditable hardware — open-ISA silicon like RISC-V that can be inspected at the instruction-set level, unlike proprietary NVIDIA or Intel architectures. Third, they are designed for AI-native workloads — high power density, purpose-built cooling, and optimised networking for large language model inference and training.

Why is RISC-V important for AI infrastructure investors?

RISC-V eliminates three critical risk factors for AI infrastructure investors. Export control risk: unlike NVIDIA GPUs, RISC-V silicon is not subject to US BIS export restrictions, ensuring hardware availability regardless of geopolitical developments. Vendor concentration risk: multiple RISC-V silicon providers create competitive dynamics that reduce single-vendor pricing power. Supply chain risk: open-ISA hardware can be manufactured at multiple foundries globally, reducing dependence on any single fab. Additionally, RISC-V's dramatically lower unit costs ($110K per Galaxy server vs $3M for NVIDIA DGX) improve capital efficiency and accelerate time to positive unit economics.

What regulatory advantages does EU sovereign AI infrastructure offer?

EU sovereign AI infrastructure benefits from a layered regulatory moat: the EU AI Act requires transparency and auditability for high-risk AI systems (favouring open hardware); GDPR mandates data processing controls that are structurally easier to satisfy with EU-jurisdictional infrastructure; NIS2 imposes supply-chain security requirements on critical infrastructure operators; and DORA creates specific digital resilience standards for financial services. Together, these regulations create compliance requirements that non-EU providers cannot easily satisfy, effectively granting EU-based sovereign infrastructure providers a structural competitive advantage.

How do I evaluate an AI infrastructure investment opportunity?

Key evaluation criteria for AI infrastructure investments include: (1) Power access — does the operator have secured grid connections or behind-the-meter generation? (2) Hardware strategy — is the facility dependent on a single vendor (NVIDIA) or does it have supply chain diversification? (3) Regulatory positioning — is the facility located in a jurisdiction with sovereignty advantages? (4) Tenant pipeline — are there committed customers or letters of intent? (5) Management team — does the team have data centre development and operations experience? (6) Capital efficiency — what is the cost per MW of deployed capacity compared to industry benchmarks? (7) Exit strategy — are there clear paths to liquidity including strategic acquisition, REIT conversion, or government concession?