Everything About AI
NextFin News - A green hydrogen developer in Nova Scotia is about to make what is being billed as the largest private-sector investment in the province's history — and it is not for hydrogen. EverWind Fuels will announce plans to build a data center at its Point Tupper industrial site and lease the facility to an artificial-intelligence developer, a pivot that turns a clean-fuels project into a bet on the one resource AI companies are fighting hardest to secure: electricity.
The announcement, scheduled for the company's industrial site near Port Hawkesbury on Cape Breton Island, will be joined by Nova Scotia Premier Tim Houston, federal parliamentary secretary Kody Blois, and Membertou Chief Terry Paul, according to a company notice. The identity of the AI tenant and the facility's capacity were not disclosed. But the direction is unambiguous: the company that spent years permitting wind farms and a green ammonia plant now sees its most valuable asset as the power those turbines will generate.
That is the real story here. For three years the AI buildout was framed as a race for chips. It has become a race for watts.
The Deal and the Backdrop
EverWind controls an industrial hub at Point Tupper built on the footprint of the former Gulf Oil Refinery, acquired from NuStar Energy in 2022. Its stated plan is a green hydrogen and ammonia facility fed by more than 650 megawatts of onshore wind generation across Nova Scotia — the Bear Lake, Kmtnuk, Upper Afton and Windy Ridge projects in phase one, with Setapuktuk, Ocean Lake and Harbour Hills to follow in Guysborough and St. Mary's. Construction on the wind portfolio is planned to begin in 2026, with commercial operations targeted for 2028. The company says the 650 megawatts can power roughly 200,000 homes annually.
The ammonia plant itself is sized at 200,000 tonnes a year in its first phase, potentially operational by 2029, with a second phase adding another 800,000 tonnes. An Indigenous consortium led by Membertou First Nation holds a majority stake. In March, Nuveen Energy Infrastructure Credit committed US$175 million (CAD$240 million) toward the platform — one of the largest clean-energy financings in Atlantic Canadian history — and the company has described the total capital invested in Nova Scotia as exceeding $2 billion.
Into that architecture, a data center now fits. The company said it is exploring opportunities to host a data center at its existing industrial site, and that the development would complement its hydrogen and ammonia plans by strengthening the energy and infrastructure platform at Point Tupper while attracting new industries that can create good jobs and contribute to Nova Scotia's economic growth. Claire Parsons, a company spokesperson, put the logic plainly:
A potential data centre development would complement those plans by strengthening the energy and infrastructure platform at Point Tupper, while attracting new industries that can create good jobs and contribute to Nova Scotia's economic growth.
The company has also said it remains committed to green hydrogen and ammonia production.
The timing tracks a broader shift in Ottawa and across the provinces. In June, Premier Houston said data centers would not be entertained until Nova Scotia secured more energy. This month, he reversed course, citing the federal government's new responsible data center framework, major investment in an AI data center in Saskatchewan, and global political volatility. "The conversation around us has changed very quickly," he said. "Data centres are increasingly critical infrastructure supporting our economy, banking, health care, research and government services. Canada should not be dependent on other countries to store and process critical Canadian data. We should build that capacity safely here at home."
Houston set five conditions for any "major" data center: meeting or exceeding federal standards; closed-loop cooling and strict water, noise and emissions rules; no negative impact on other customers' electricity rates; significant local jobs and business benefits; and a direct financial return to Nova Scotians above normal taxes. EverWind says it will meet federal and provincial requirements.
Why a Hydrogen Company Wants an AI Tenant
On the surface, a green ammonia exporter and an AI data center have nothing in common. One ships a chemical feedstock to Europe; the other processes tokens for a tech tenant. Beneath the surface, they compete for the same input and solve the same financing problem.
Green hydrogen is an electricity product. Its economics live or die on the cost and utilization of the power behind it. Wind is intermittent, and electrolyzers that sit idle between gusts destroy returns. A data center, by contrast, is the ideal industrial load: it runs continuously, it signs long-duration contracts, and it pays for capacity whether or not the wind is blowing. Leasing a data center against a wind portfolio converts an intermittent commodity business into a contracted utility-style cash flow.
There is also a sequencing advantage. Ammonia offtake agreements with European buyers depend on policy support schemes that can shift with elections and budgets. An AI tenant is a domestic counterparty whose demand is less exposed to European subsidy risk. For a developer trying to reach a final investment decision on a multibillion-dollar platform, a creditworthy anchor tenant de-risks the bankability of the whole energy portfolio — wind, transmission, and fuel plant alike.
The physical site helps. Point Tupper already has a deepwater terminal, industrial zoning, and grid interconnection work underway, including the Strait Crossing transmission line that will tie the phase-two wind farms directly to the fuel plant. A nearby 150-megawatt generating station, opened in 1973, is planned to switch from coal to natural gas from 2029. For a hyperscale tenant, permitting a new site from scratch can take years; stepping onto a permitted industrial hub with power already in development can cut that timeline in half.
The lease structure matters too. Across the AI data-center market, contracts are stretching to 15 to 20 years as power becomes the asset, with take-or-pay terms that let developers finance construction against contracted rent. Research firm MarketsandMarkets projects the global AI data-center market will grow from US$471.59 billion in 2026 to US$2.02 trillion by 2032, a 27.5% compound annual rate; Fortune Business Insights sizes the facilities segment alone at US$21.27 billion in 2026 rising to US$133.51 billion by 2034. In that market, a developer with secured power and a permitted site holds the leverage, not the tenant.
EverWind's move is therefore not a pivot away from hydrogen. It is a hedge around it.
The Structural Shift: Power Is the New Bottleneck
This is the structural read, and it is the one the market is still pricing incompletely. The AI infrastructure bottleneck has migrated from semiconductors to electricity. Chip supply has loosened; grid interconnection queues have not. In North America, data-center developers routinely wait years for transmission access, and states and cities from New York to Oregon, Iowa, Florida and Tennessee have paused or restricted new hyperscale projects over grid strain and water use. The constraint is no longer compute — it is the electrons that compute runs on.
That changes what "infrastructure" means. A wind portfolio with a permitted industrial site and a substation is now a data-center development platform. The value accrues to whoever controls the interconnection and the electrons, not necessarily to the company that racks the servers. EverWind's model — build the power, own the site, lease the shell to an AI operator — is a template other resource-rich regions will copy: Labrador, Saskatchewan, Manitoba, the Nordic countries, the American Plains states. The competitive advantage is geographic and regulatory, and it does not depreciate the way chip inventories do.
There is a second-order implication most coverage misses. If power is the scarce input, then the cheapest, cleanest electrons will command a premium, and AI training clusters will migrate toward them. That favors regions with stranded or underpriced renewable resources — exactly what Atlantic Canada has. The EverWind deal is an early signal that AI capital is willing to go where the power is, rather than expecting the power to come to the data centers.
But this is not purely structural. There is a cyclical leg to the trade. AI data-center demand is front-loaded by a capital-expenditure supercycle that cannot continue at the current pace indefinitely. Financing costs, chip cadence, and model-efficiency gains will all moderate the slope of demand. The companies that survive the down-leg of that cycle will be the ones with contracted, low-cost power and long-duration tenants — again, the EverWind structure. The cyclical wave will wash out the developers who leased power at spot rates; the structural shift will leave the power owners standing.
The Counter-Thesis: The Grid Is Small, and Politics Are Volatile
The strongest case against this deal is not commercial; it is political and physical. Nova Scotia's grid is small, still heavily coal-dependent, and rate-sensitive. Adding a hyperscale load risks raising electricity costs for households and other businesses — the exact outcome Premier Houston's five conditions are designed to prevent. Manitoba's premier recently rejected a large-scale data-center proposal near Winnipeg, saying it would not create enough jobs to be economically worthwhile. Environmental groups in the Maritimes have already mobilized against data centers, and the province's coal-to-gas transition at Point Tupper means the facility will not be carbon-free when the wind is not blowing.
There is also execution risk. EverWind's wind farms are not yet built; commercial operations are targeted for 2028. A data center that opens before its dedicated generation is online will draw from a coal-heavy grid, undercutting the "green AI" premium that likely attracted the tenant in the first place. And the tenant's identity remains undisclosed, which leaves open the question of whether the counterparty has the balance sheet to honor a 15- to 20-year take-or-pay lease through a cyclical downturn.
These are real risks, but they are manageable rather than fatal. The premier's conditions, if enforced, force the project to internalize its grid costs rather than socialize them. The Indigenous majority stake gives the project political durability in the host region. And the phased build — wind first, fuel plant and data center alongside — means the load can be matched to generation over time rather than all at once.
The falsifying signal is specific: if Nova Scotia Power's next rate filing shows residential or commercial rate increases attributable to data-center load, or if the province rejects or delays the project under its own five conditions, the structural thesis weakens materially. A second signal would be the tenant walking away or lease terms shortening well below the 15-year market standard — that would indicate power is not as scarce as the structure assumes.
What to Watch
Short term (the announcement and permitting): the disclosed investment figure, the tenant's identity, and whether the project clears the province's five conditions without rate relief that shifts costs to other customers. The presence of the premier and a federal parliamentary secretary at the launch suggests Ottawa and Halifax are aligned — for now.
Medium term (2027-2029, construction and commissioning): whether the 650 megawatts of wind come online on the 2028 schedule, whether the ammonia offtake agreements hold alongside the data-center lease, and whether the Point Tupper generating station's coal-to-gas conversion proceeds as planned. Any slippage on the wind build directly threatens the "green" claim the tenant is buying.
Long term (the 2030s, the structural bet): whether AI load continues to migrate toward cheap renewable baseload, and whether EverWind's power-plus-site model gets replicated across Atlantic Canada and the Nordic periphery. If it does, the company will have discovered a more durable franchise than green ammonia alone.
The base case is that the data center proceeds, the wind portfolio energizes on schedule, and EverWind emerges as a hybrid power-and-digital-infrastructure landlord with contracted cash flows from two different end markets. The upside case is that this becomes the template for converting stranded renewable resources across Canada into AI-grade power, multiplying the platform's value. The downside case is that grid politics force rate concessions or delays that strip the economics from both the data center and the ammonia plant.
For three years, the AI buildout was a story about who could buy the most GPUs. EverWind's bet is that the next three years will be decided by who owns the electrons. If power is the asset, a wind developer with a permitted site and a long-term tenant is not a hydrogen company that got distracted — it is the business model that wins the next phase.
Explore more exclusive insights at nextfin.ai.
