Investors and operators modelling AI infrastructure capacity have spent the past two years focused on semiconductor supply chains, GPU allocation and server lead times. That framing is now incomplete. On 6 October 2026, Finland ordered a temporary halt to work on two Google data centres pending environmental impact assessments, according to reporting by BBC News (bbc.co.uk) and The Guardian (theguardian.com). On the same day, The Verge (theverge.com) reported that Google had signed a nuclear power purchase agreement for its power-hungry data centres.
The juxtaposition is not coincidental. It is a signal that the binding constraints on AI capacity build-out have shifted from silicon to sites, permits and electrons. For investors underwriting data centre returns, the Finland halt is a case study in jurisdictional and permitting risk that deserves to be modelled explicitly rather than treated as a tail event.
What happened in Finland
Finland's decision to halt work on two Google data centres pending environmental impact assessments is notable for where it occurred. Finland has been one of Europe's more attractive data centre markets, benefiting from cooler climates that reduce cooling costs, relatively abundant low-carbon electricity and a stable regulatory environment within the EU. A stop-work order in that context is not a signal of host hostility to technology investment. It is a signal that environmental approval processes are being applied with more rigour as the scale and resource intensity of AI data centres increases. The precise duration of the halt, the conditions for resumption and the capital already committed to the sites are not fully specified in the available reporting. What is clear is that the projects cannot proceed to completion on their original timeline while the assessment is outstanding. That has direct implications for capacity delivery dates, depreciation schedules and the revenue assumptions built into any underwriting model.The nuclear pivot and what it reveals
Google's nuclear power purchase agreement, reported by The Verge, points to the second half of the same problem. AI data centres are power-intensive, and the grid connections, renewable energy certificates and short-term contracts that sufficed for earlier generations of facilities are increasingly inadequate for the load profiles and uptime requirements of AI training and inference. Nuclear power offers firm, low-carbon baseload generation, which is why hyperscalers have been pursuing it. But nuclear procurement is slow, capital-intensive and subject to its own regulatory approvals. A power purchase agreement improves the long-term energy outlook for a portfolio; it does not resolve a near-term permitting halt at a specific site. The two developments should be read together as evidence that energy strategy and site-level approvals are separate risk factors that require separate modelling.Why permitting is now a first-order investment variable
For most of the past decade, data centre investors could treat permitting as a background process with predictable timelines in developed markets. That assumption is weakening. Environmental impact assessments, water usage reviews, grid connection queues and local planning objections are all becoming more material as facilities grow in size and visibility. The Finland case illustrates three specific risks that belong in a diligence checklist. First, timeline risk: a halt pending assessment can delay revenue recognition by quarters, not weeks. Second, cost risk: delays increase carrying costs, financing costs and the risk of contract penalties. Third, jurisdictional risk: a stable EU member state can still impose stop-work orders, which means the risk premium attached to any single site should reflect the specific approval regime, not just the country's general business climate.What operators and investors should model
A more robust underwriting approach would separate AI infrastructure risk into four distinct layers. The first is supply chain risk, covering chips, servers and construction materials. The second is power risk, covering grid capacity, generation contracts and energy price exposure. The third is permitting and regulatory risk, covering environmental assessments, local planning and EU-level rules. The fourth is execution risk, covering construction, commissioning and operational performance. The Finland halt and the nuclear PPA sit in different layers. Treating them as a single narrative about AI infrastructure misses the point. Investors who model them separately will be better placed to price the risk and to structure contracts that allocate delay costs appropriately.Commercial impact
For data centre operators, the immediate commercial impact is a reminder that site selection must weigh regulatory approval timelines alongside power availability and land cost. A site with cheap power but a slow or uncertain environmental approval process may carry a higher effective cost than a more expensive site with a clearer path to construction. For investors, the impact is a potential repricing of assets with concentrated exposure to jurisdictions where environmental scrutiny is increasing. That does not mean avoiding the EU or Finland. It means demanding more granular disclosure on permitting status, approval conditions and contingency plans before committing capital. For energy suppliers and nuclear developers, the Google PPA reinforces that hyperscalers are willing to sign long-term contracts to secure firm low-carbon power. That creates opportunities, but also raises questions about how quickly new nuclear capacity can be delivered relative to AI demand growth.Risks and unknowns
The duration of the Finland halt is unknown. The conditions under which work can resume are not fully specified in the available reporting. The financial terms of Google's nuclear power purchase agreement are not public. The extent to which other hyperscalers face similar permitting scrutiny in EU markets is not established by the supplied sources. There is also uncertainty about how environmental assessment requirements will evolve. If regulators tighten rules further, the cost and timeline assumptions in current data centre models may prove optimistic. Conversely, if approval processes are streamlined, the current halt may prove to be a temporary disruption rather than a structural shift.FY Outlook
The Finland halt and the nuclear PPA point in the same direction: AI infrastructure capacity is increasingly gated by permits and power, not chips. Investors should expect more scrutiny of environmental impact, water usage and grid connections as data centre scale increases. Operators that can demonstrate a clear path through permitting and a credible long-term energy strategy will be better positioned than those relying on historical assumptions about approval speed. The next signals to watch are the outcome of the Finnish environmental assessments, the terms and delivery schedule of Google's nuclear agreement, and whether other EU member states adopt similar stop-work approaches. Each will inform how much jurisdictional risk premium is warranted in AI infrastructure valuations.Sources and References
- BBC News (bbc.co.uk)
- The Guardian (theguardian.com)
- The Verge (theverge.com)
Why It Matters
The Finland halt and Google's nuclear PPA show that AI infrastructure capacity is now gated by permitting and power procurement, not chip supply. Investors who continue to model only semiconductor risk will misprice jurisdictional and energy exposure.The reporting and evidence for this briefing were checked against bbc.co.uk (bbc.co.uk) and theguardian.com (theguardian.com) and theverge.com (theverge.com).
