I keep coming back to a single number that never appears in the press release. Not C$10 billion. Not the megawatt count. The number I want is Nova Scotia's peak load, and it sits somewhere around 2,200 megawatts. Hold that against a project promising over ten billion Canadian dollars, an AI data center, and a "wind build" in one of Canada's windiest provinces. Something in that arithmetic doesn't close, and the gap is where the real story lives.
The announcement itself is thin enough to read in a breath. EverWind Fuels plans to invest more than C$10 billion in Nova Scotia to build an AI data center powered by wind, boosting the economy, jobs, tax revenue, and renewable energy capacity. That's the whole of it. No interconnection number. No megawatt target. No customer. No financing structure. No discussion of storage, which is the first thing any engineer who has actually built a data center would ask about, and the fact that it's absent tells you more than any of the numbers that are present.
So let me tell you what it looks like from the inside. In 2020 I audited over 150 Uniswap V2 liquidity pools during DeFi Summer, chasing a slippage edge case that touched two million dollars of user funds. That work taught me a discipline I now apply to every energy announcement that lands in my inbox: the spreadsheet always looks beautiful until you ask the protocol what happens when the assumption breaks. In DeFi the assumption is liquidity. In renewable-powered compute, it is dispatchability. And wind is the least dispatchable thing we have.
The engineering gap hiding in plain sight
A modern AI training cluster runs at power densities that would have seemed absurd a decade ago. Rack-level draw has climbed from five to eight kilowatts in traditional halls to thirty, fifty, sometimes beyond a hundred kilowatts for dense GPU configurations. These are rigid, oscillating, high-quality loads. Data center operators sign availability guarantees at 99.99 percent and above, because a training run that dies at hour eleven of a fourteen-hour job is not an inconvenience, it is a seven-figure write-off.
Wind onshore in Nova Scotia runs a capacity factor in the range of 35 to 42 percent. That is an excellent resource by global standards, and it is also fundamentally incompatible with a rigid baseload load if wind is the only thing on the other side of the meter. The physics does not negotiate. When the wind drops, something else must pick up the load within milliseconds, and the press release does not tell us what that something is.
The realistic architecture, if this project is real, is wind plus lithium iron phosphate storage for two to four hours plus deep interconnection with Nova Scotia Power plus probably gas peaking as backstop. Long-duration storage, flow batteries or compressed air, is unlikely here, because data centers do not need four-to-eight-hour shifting as much as they need millisecond frequency response and absolute reliability. A four-hour LFP system at current costs lands somewhere around ¥0.3 to ¥0.6 per kilowatt-hour, cycling six to eight thousand times, responding in milliseconds. That is the sensible pairing. It is also nowhere in the announcement.
The capacity-versus-energy sleight of hand
Here is where the language starts doing work it shouldn't. The release says the project will enhance renewable energy capacity. Capacity is not energy. Energy is not dispatchability. Adding wind megawatts to a provincial grid does not mean a single electron of that wind reaches the data center's power supplies, because electrons follow the path of least resistance across a shared network and the accounting is done in certificates, not copper.
This is the oldest trick in climate communications, and I have watched it migrate from hydrogen roadshows into AI infrastructure pitches over the last two years. The renewable capacity figure is real. The green claim attached to the compute load is a separate proposition entirely, and it requires twenty-four-seven carbon-free energy matching to be defensible. Hyperscalers with published RE100 and 24/7 CFE commitments understand this because their auditors make them understand it. Nova Scotia's renewable certificate market is thin, which means anyone serious about green compute there is either building behind the meter or importing international I-RECs to make the spreadsheet reconcile.
And then there is the Atlantic Loop. This is the variable that quietly determines whether the whole thing makes sense. If the transmission corridor connecting Nova Scotia to Quebec hydro finally gets built, the province gets stable, dispatchable, low-carbon baseload, and the strategic necessity of a local wind-plus-data-center complex drops sharply. If it does not get built, and it has been delayed for years, then coal retirement in 2030 opens a supply gap so wide that something has to fill it. The project's value proposition flips depending on a transmission line the announcement never mentions. That is not a detail. That is the entire thesis.
What EverWind actually is
Here is the part that the crypto wires left out, and it changes how you read everything else. EverWind Fuels is not a data center operator. It is a green hydrogen and green ammonia developer, with its Point Tupper project aimed at exporting to Germany under the H2Global procurement mechanism. Its core business is electrolysis, ammonia synthesis, and shipping molecules across the Atlantic.
Green hydrogen currently costs somewhere in the range of four to eight dollars per kilogram depending on power price and electrolyzer economics, against gray hydrogen at roughly half that. The gap is bridged by subsidies and offtake agreements, and European demand for green ammonia has been materializing more slowly than the headlines promised. I have spent enough time in Berlin institutional circles to know how this story usually ends, and it usually ends with a developer looking for a second narrative.
So why does a hydrogen exporter suddenly want to build an AI data center? There are three plausible answers, and they all point the same direction. One: the wind farm needs a local load to justify its interconnection, and a data center is a very photogenic anchor. Two: hydrogen export risk needs diversifying. Three: AI-era power, land, and policy incentives are simply too attractive to ignore. All three explanations converge on a single conclusion, which is that the hydrogen economics have not yet closed. The data center is not the business. The data center is the argument that unlocks the business.
The bottleneck migrated and nobody updated the map
What makes this announcement interesting, despite being nearly empty of verifiable content, is the macro signal buried in it. For three years the constraint on AI was chips. Now the constraint is power, and the geography of compute is being redrawn around places with cheap electrons, cold air, and available interconnection. Nova Scotia has decent wind and genuinely excellent cooling weather for low PUE operation. What it does not have is spare grid capacity, and its peak load is roughly a fifth of what a serious multi-gigawatt build would require.
That means a real project at this scale triggers transmission upgrades, substation expansions, and distribution reinforcement that almost certainly sit outside the ten-billion-dollar headline. A capital figure that excludes grid works is not a capital figure. It is a marketing number. The same is true of the transformer supply chain, where lead times stretched to two to three years during the 2023 to 2024 crunch, squeezed simultaneously by data center buildouts and renewable construction. You can announce anything you like. The transformers arrive when they arrive.
And underneath all of it sits the question no press release ever answers. Who runs this data center, and who is buying the compute? There is no named hyperscaler, no offtake agreement, no indication of whether the workload is training or inference. The industry has spent the last two years announcing capacity faster than it has signed customers, and building first while hunting for tenants later has an ugly track record in every infrastructure cycle I have watched since the ICO boom.
I want to be fair here, because there is a version of this story that is true and boring and good. Nova Scotia needs to replace coal by 2030. Wind is its best resource. Data centers are its best new load. A developer that already owns wind development capability and needs a local customer is doing exactly the rational thing. The boring version requires an interconnection queue number, a named offtaker, a disclosed financing stack, and a storage configuration with megawatt-hours attached. None of those four appear. Until they do, the ten billion should be read as a letter of intent to have a conversation, not a commitment to pour concrete.
Mining for truth in the noise of a ten-billion-dollar headline
My audit instinct says test the claim, not the tone. When I found that slippage bug in 2020, nobody wanted to hear that the edge case existed, because the happy path was making everyone money. The same discomfort applies here. The uncomfortable question is not whether Nova Scotia wants this. Of course it does. The uncomfortable question is whether wind can be the primary supply for a rigid load, and the answer is no unless storage and interconnection carry the weight, and neither appears in the announcement.
What the developer gets from this announcement is attention, a seat at the policy table, and a renewable-capacity story that plays well in both Ottawa and Berlin. What the province gets is a headline in a year when it badly needs one. What the reader should take away is narrower and more useful: the AI buildout has crossed the threshold where power availability, not silicon, determines which projects exist, and any announcement that does not lead with interconnection and offtake is a press release, not a project.
Watch for four signals and ignore everything else. An interconnection queue number. A named compute customer with a signed power purchase agreement. A financing structure naming the investment tax credit or a strategic investor. A storage configuration with hours and megawatts attached. Any one of those makes this real in a way no dollar figure can. All four make it inevitable.
In the meantime, the most honest reading is that this is a green hydrogen company using an AI data center to make its wind farm legible to lenders, and that the AI data center is using a wind farm to make its power supply legible to regulators. Both parties benefit from the story. Neither has shown their work. The electrons, when they arrive, will not care what the press release said, and the grid operators who have to keep the lights on at 3 a.m. on a still February night in Halifax will be the ones who find out first whether any of this was real.