Everything About AI
NextFin News - Centalion, the Geneva-based energy commodities trader formerly known as Gunvor Group, is treating the Nordic data center buildout as a structural driver of power demand rather than a passing AI capex cycle, positioning itself to profit from a region where electricity load is forecast to grow faster than new generation can be built. The company sees data centers as a central pillar of its Nordic power push, a bet that hinges on a simple imbalance: Nordic data center electricity consumption is projected to reach 28 terawatt-hours a year by 2030, or 5% of total regional demand, up from 8 TWh and 2% in 2024, according to projections by Statnett, the Norwegian transmission system operator.
The Nordic Power Squeeze: Demand Rising Faster Than Supply
The opportunity for a power trader is not merely that data centers consume electricity. It is that they consume it in large, concentrated, inflexible blocks in a market where the cushion of spare capacity is shrinking fast. Statnett's projections break the 2030 picture down by country: Sweden at 9 TWh a year, Denmark at 8 TWh, Norway at 7 TWh, and Finland at 4 TWh, the Finnish figure roughly doubling from 2024 levels. That concentration matters because data center demand is not spread evenly across the grid; it lands in specific bidding zones and can dominate local load. In 2023, data centers consumed between a third and 42% of all electricity in Amsterdam, London and Frankfurt, and almost 80% in Dublin, according to the think tank Ember.
On the supply side, the margin for error is narrowing. Svenska Kraftnat, Sweden's grid operator, forecasts that the Nordic regional power surplus will fall to 29 TWh by 2030 from around 53 TWh in 2026. In Sweden alone, the surplus is projected to shrink by 65% to 12 TWh, with generating capacity rising only to about 56.1 gigawatts from roughly 50 GW in 2026. When demand growth from one sector outpaces the growth of total system capacity, the result is not just higher average prices; it is wider intraday and seasonal price spreads, more frequent congestion, and a greater premium on the ability to balance, store, and route electrons. That is the terrain a commodities trader is built to navigate.
The Nordic region's attraction to data center developers rests on a durable foundation rather than a temporary subsidy. Nearly all Nordic electricity comes from renewable or low-carbon sources, with Norway deriving more than 95% of its power from hydropower, while Sweden and Finland lean heavily on hydro, wind, and nuclear. Cool ambient temperatures cut cooling costs. Political stability and sophisticated grid and communications infrastructure round out the case. Regional data center capacity is projected to exceed 2.8 GW by 2030, and the region already hosts more than 200 facilities with total installed IT power capacity exceeding 600 MW, with Sweden and Norway each approaching or exceeding 1 GW.
Yet the region's readiness is not accidental. Denmark's grid operator, Energinet, began anticipating a data center boom as early as 2017, years before AI became mainstream, and the Nordic transmission system operators have maintained some of the lowest congestion levels in Europe through deliberate planning. That head start is now paying off as the established markets hit physical limits: Ember reports connection waits of up to 13 years in some European locations, and Dublin has paused new data center projects until 2028 over power concerns. The Nordics, by contrast, are emerging as the release valve for a continent running short of connectable capacity.
Why a Trader, Not Just a Generator, Wins
The conventional read of the AI power story is that generators and equipment suppliers are the clear winners. That is true as far as it goes, but it stops at first-order effects. Centalion's positioning points to a second-order conclusion: when a grid cannot expand as fast as load arrives, the value migrates to the intermediaries who can optimize what already exists. Hyperscalers are already acting on this logic. Google has committed to taking half of the electricity from Finland's Loviisa nuclear plant for 20 years, part of a roughly €13 billion data center investment in the country announced in September 2026. Meta has explored entering electricity trading to secure long-term supply for its AI campuses. These are not just procurement deals; they are acknowledgments that power availability, not server count, is the binding constraint on AI buildout.
The scale of the buildout underscores the point. OpenAI has selected Narvik, a town of about 21,000 people in northern Norway, for its first European data center. Microsoft has said it would invest $3.2 billion to expand AI and cloud infrastructure in Sweden. Google has pledged another €1 billion to expand its Finnish campus, and AI chip developer Groq has announced its first European data center in Finland. Over the next decade, energy demand from data centers in the region is expected to grow four- to fivefold, according to Ember, putting Norway and Sweden on par with the biggest European hubs. Across Europe as a whole, data center power demand is forecast to reach 168 TWh in 2030 and 236 TWh in 2035, an increase of almost 150% in ten years, with demand in Sweden, Norway and Denmark alone expected to triple by 2030.
For a trader, the profit pool sits in the gaps these long-term deals cannot close. A 20-year nuclear offtake locks in baseload, but it does not hedge intraday volatility, cross-border congestion, or the need to balance intermittent wind and hydro against a load that runs flat and hard. As the Nordic surplus shrinks toward 29 TWh, each marginal megawatt-hour becomes more valuable and more volatile. Trading houses monetize exactly that: arbitrage between hydro-rich Norway and thermal or nuclear-heavy neighbors, balancing services, and the spread between physical power and financial hedges. The data center boom, in this reading, is less a volume story than a volatility story.
"Grid availability is actually now topping the list as the biggest factor affecting where these big companies are choosing to put their facilities – above price or anything," said Elisabeth Cremona, a senior analyst at Ember. "These companies really prioritize speed."
The grid bottleneck is visible in real time. T1 Energy, a US data center developer, secured only 50 MW of grid power from Statnett for its Mo i Rana facility in March 2026 while 396 MW remained stuck in the interconnection queue. Industrial Info Resources tracks 121 data center projects in Finland alone, worth nearly US$69 billion in investment. When committed capital vastly exceeds connectable capacity, the scarce asset is not the building or the chips; it is the grid connection, and the right to move power through a constrained network.
Cyclical Wave or Structural Shift?
The central question for investors is whether this is a cyclical AI spending wave that will mean-revert or a structural re-rating of power as a scarce commodity. The evidence points to structural. Three factors separate this from a typical capex cycle. First, the demand driver is a step-change in load intensity: AI training and inference workloads run at high utilization around the clock, unlike the more variable load of traditional cloud or enterprise computing. Second, the supply response is physically constrained. Building new generation and transmission in the Nordics means multi-year permitting, environmental reviews, and grid reinforcement, none of which can be compressed quickly. Svenska Kraftnat's forecast of only modest capacity growth to 56.1 GW by 2030, even as data center demand alone adds roughly 20 TWh of new load, shows the mismatch is baked into the decade.
Third, the demand is global and compounding. The International Energy Agency's base case projects global data center electricity consumption could reach 945 TWh by 2030 and 1,200 TWh by 2035, up from about 415 TWh in 2024, when the sector consumed roughly 1.5% of the world's electricity and had been growing at about a 12% compound annual rate since 2017, more than four times faster than total global electricity consumption. In the United States, the Department of Energy estimates data centers consumed about 4.4% of total US electricity in 2023 and could reach between 6.7% and 12% by 2028. A cyclical wave does not compound at that rate across multiple continents simultaneously.
The long-term shape of the Nordic curve reinforces the structural read. A study commissioned by the Norwegian data center industry and prepared by consultancy THEMA projects Norwegian data center electricity demand rising from 1.6 TWh in 2024 to 18 TWh by 2040, more than a tenfold increase, and 23 TWh by 2050, with growth slowing only after mid-century as the fleet matures. That is not the profile of a spending cycle; it is the profile of a new, permanent load class on the grid.
That said, the cyclical and structural legs should be separated. In the short term, the buildout is sensitive to AI capital expenditure budgets, chip supply, and financing conditions. If hyperscaler spending slows, project announcements will fall and some grid applications will be deferred. But even a slower growth path leaves data center load far above today's level, because the installed base of AI infrastructure draws power continuously once built. The structural call is that power demand from digital infrastructure does not revert to its old trend; the cyclical call is that the pace of new project announcements can fluctuate with the AI investment cycle.
"We are still in the early innings," said Carl Sjölund, a partner at EQT in Stockholm.
The Counter-Case: Grid Constraints Could Kill the Boom
The strongest argument against the bullish reading is that the constraint cuts both ways. If the Nordic grid cannot deliver power, data center developers will go elsewhere, and the projected 28 TWh of demand by 2030 will never materialize. Grid operators and analysts have flagged exactly this risk: Norway's development is limited in the short term by complex connection factors and in the longer term by limited grid capacity and available generation, with the THEMA study warning that physical limitations shape the expected path even as demand grows more than tenfold by 2040. Cushman & Wakefield's 2026 market comparison ranks Helsinki first and Oslo second among EMEA primary markets, but rankings of attractiveness are not the same as connectable capacity.
This counter-thesis attacks the core of the opportunity: if demand is capped by the grid, then the volume growth that underpins the trading thesis is capped too. The answer is that a cap on volume does not cap the value of trading; it may enhance it. Scarcity raises the price of the marginal megawatt-hour and widens spreads between constrained and unconstrained zones, which is precisely what a trader monetizes. Even if total Nordic data center load lands below the 28 TWh forecast, the path to that lower number runs through higher volatility and more congestion, not less. The falsifying signal is specific: if Statnett's 2030 data center demand forecast is revised down by more than 30% while Nordic price spreads and congestion revenues simultaneously narrow, the thesis that scarcity enriches traders would be wrong. A demand cut paired with calm, well-connected markets would mean the boom was deferred, not transformed.
What to Watch
The forward picture splits by time horizon. In the short term, watch interconnection queue data from Statnett and the other Nordic TSOs, plus the pace of final investment decisions on new generation and transmission. A backlog that grows faster than approvals is bullish for volatility and bearish for volume. In the medium term, the key variable is the Nordic power surplus: Svenska Kraftnat expects it to fall to 29 TWh by 2030, and any acceleration of that decline would lift the scarcity premium. In the long term, the structural question is whether AI workload growth holds at the IEA's projected trajectory toward 945 TWh globally by 2030.
Three scenarios frame the outlook. The base case is that data center demand reaches most of the projected 28 TWh by 2030, grid expansion lags, and traders capture value from volatility and congestion. The upside case is that AI load grows faster than forecast while generation additions stall, producing sustained price spikes and a permanent re-rating of Nordic power as a scarce asset. The downside case is that grid constraints force developers to relocate to other regions, demand lands well below forecast, and spreads normalize as the AI capex cycle cools.
The data center boom is turning electricity into the scarce input of the AI age, and the Nordics sit at the center of that shift. For Centalion and other traders, the prize is not the electrons themselves but the disorder their arrival creates in a grid that cannot expand fast enough to meet them.
Explore more exclusive insights at nextfin.ai.
