I grew up in Vaasa and have spent nearly twenty years running companies in Mexico. I watch both economies out of habit, and for most of the last decade the Finnish news has been a study in managed decline — a small open economy with an aging population, a structurally weak export base, and monetary policy set in Frankfurt for a eurozone average that does not describe it.

Then in September a single announcement landed that is worth more than a year of Finnish GDP growth, and it is worth understanding why, because the reasoning transfers to every market competing for the same capital.

What was actually announced

On 9 September 2026 Google committed at least EUR 13 billion — roughly USD 15 billion — to AI and data center infrastructure in Finland over 2027 and 2028. Alongside it, Fortum and Google signed a 22-year power purchase agreement covering up to 50% of the output of the Loviisa nuclear plant, with deliveries beginning in 2028 and the full 50% share running from 2030 to 2049.

Ruth Porat, Alphabet’s President and Chief Investment Officer, called it Google’s first nuclear energy deal outside the United States, and described the model in four words to reporters in Helsinki: “We call this BYOP, bring your own power.”

The two companies also signed a memorandum of understanding covering new generation, flexible capacity and portfolio management services, and Google has contracted a 94 MW battery storage system next to its data center in Kajaani.

Set that against the macro

The Bank of Finland’s June 2026 forecast has GDP growing 0.7% in 2026 and 1.2% in 2027. The European Commission’s spring 2026 forecast is close: 0.8% and 1.4%. Unemployment is forecast at 10.4% by the Bank of Finland and 10.1% by the Commission. General government debt is 91.9% of GDP on the Bank of Finland’s numbers and 91.2% on the Commission’s.

On labor cost, tax treatment and growth outlook — the three variables that dominate most site-selection conversations — Finland does not obviously win. So the decision was made on something else.

The binding constraint moved

For AI infrastructure at this scale, the scarce input is no longer capital, and it is no longer engineers. It is firm, low-carbon power that can be contracted for twenty years, in a place cold enough that cooling does not eat the efficiency gain.

Loviisa supplies around 10% of Finland’s electricity from two units that entered commercial operation in 1977 and 1981. That is the asset Google contracted against — not a promise of future generation, but a running plant with a licence and a track record.

One correction worth making, because the shorthand is wrong

The story is often told as Google rescuing a plant that was about to close. It is not that. Loviisa’s previous operating licences ran to 31 December 2027 and 31 December 2030, but the Finnish Government granted new licences on 16 February 2023 permitting operation to the end of 2050. The plant’s life was settled three years before Google arrived.

What the contract unlocks is the money. Fortum’s lifetime-extension programme for Loviisa is around EUR 1 billion across 2023 to 2050, and by Fortum’s own account roughly EUR 700 million of it — about 80% of the projects — was still awaiting investment decisions. Twenty-two years of contracted revenue is what makes those decisions financeable. That is the mechanism, and it is a more interesting one than a rescue.

The criticism is specific, and it is the useful part

Finland does not have unlimited headroom, and the objection being raised domestically is not about the money.

The grid. Fingrid has restricted connection capacity for new industrial-scale demand facilities in Uusimaa, Southwest Finland and Häme over 2025 to 2027, caused by consumption growing faster than expected in those regions while local generation was retired. By Fingrid’s own data, data centers now account for roughly half of all new connection enquiries.

Permitting. Opposition parties have called for a national permitting framework for data centers. Antti Kaikkonen of the Centre Party put the objection plainly — that nobody is presently looking after the overall picture — and Niina Malm of the Social Democratic Party framed it as a question of internal security rather than economics. Finland’s next parliamentary election is due in April 2027, and electricity prices are a live political subject.

Price. Nobody outside the two parties knows what the contract price is. Juha Kinnunen at Inderes has estimated it lands around EUR 75 to 80 per MWh, against Finnish 2030 futures near EUR 50 and Fortum’s current hedges at EUR 40 to 41. Kinnunen’s reading is that this sets a higher reference level for other long-term PPAs, while noting that Google is an exceptionally large and creditworthy counterparty — so the premium may not transfer intact to an ordinary buyer.

Both things can be true. Long-term contracted demand is what finances new generation, and a price set by the most creditworthy buyer in the market is not the price available to everyone else. Whether the MoU actually delivers added capacity alongside added load is an execution question, and it will be answered in 2028, not in a press release.

What this changes if you allocate capital

Energy stopped being a line item in the operating budget and became a variable in the location decision. That is the whole story, and it is not confined to data centers — it applies to any process where power is a material share of unit cost and the investment horizon is measured in decades.

Finland’s weak headline macro did not disqualify it, because the variable that decided the outcome was never in the macro data. Three things were, and they are worth naming separately because a market can have one and not the others:

• Firm power — generation that is there when you need it, contractable on a twenty-year horizon, and low-carbon enough to survive the buyer’s own reporting obligations.
• A grid that can connect it — transmission capacity at the specific point on the map where the investment would sit, not somewhere in the national aggregate.
• Permitting that can say yes on a twenty-year view — and, just as importantly, that can say no quickly, because an eighteen-month maybe is a rejection with extra steps.

Finland has the first, is short of the second in exactly the regions where demand wants to sit, and is openly arguing about the third. That is a better position than most markets, and it is still not a comfortable one.

The question this puts to Mexico

Mexico spends most of its investment-attraction conversation on labor cost, proximity and trade treatment. Those are real advantages and they are not going away. But they are the variables that mattered in the last cycle.

If the binding constraint on the next wave of industrial investment is firm power, grid capacity and permitting certainty, then the honest question for anyone marketing a Mexican location is not whether the workforce or the tax treatment is competitive. It is which of those three the site is actually short of, and whether that can be fixed inside the investor’s decision window.

The same question applies to a Nordic company evaluating Mexico. Nearshoring analyses I have been asked to review frequently model labor, logistics and duty with real precision, and treat electricity as a per-kWh assumption pulled from a tariff table. For an energy-intensive process that is the single largest unmodeled risk in the file.

I have written about the other side of this corridor — what changed in the trade framework and why Europe became arithmetically more interesting to Mexican companies — in USMCA wasn’t renewed — and why Europe now matters more.

What I would do with this

Separate energy from the operating assumptions. If power is more than a few percent of unit cost, it does not belong in a tariff-table line. Model it as a contracted price with a term, a counterparty and a renewal risk, the way you would model debt.

Ask the connection question before the site question. The binding limit is rarely national generation. It is transmission capacity at one point on the map, and the answer is held by the grid operator, not the economic development agency.

Put a date on the permitting answer. Not whether permission is obtainable — nearly everything is, eventually — but how long the process takes, who can stop it, and what an election does to that timetable.

Treat a headline PPA price as a data point, not a benchmark. The price a hyperscaler contracts is a function of its credit and its volume. Yours will be different, and the difference is quantifiable before you commit to a location.

Where I fit

I am KHT-qualified in Finland — the Finnish chartered accountant qualification, exam passed in 2005, membership currently inactive after more than twenty years abroad — with Big 4 training at Ernst & Young in Helsinki, San Jose, London and Paris, and nearly two decades running operations in Mexico across construction, mining and retail. I work as a fractional CFO and advisor, and a large part of that work is with Nordic companies entering or operating in Mexico, and with Mexican companies that need to be legible to a European counterparty.

If you are building the investment case for a Mexican location and the energy assumption is a single number in a spreadsheet, that is usually worth two weeks of work before it becomes a twenty-year commitment. The first conversation is free.

[email protected]

← Back to Insights