Atlantic Corridor: Initial Research

A Note on This Release

This document is initial research, released 2026-06-19. It is the public edition of work that has, until now, existed only inside the restricted tier of the Atlantic Corridor programme — a study BlackGrid Labs opened in June 2025 under an internal charter (BGL Corridor Memorandum AC-M-01, internal, 2025-06) and has run continuously since. Five subsequent Corridor documents remain restricted. What follows is the evidentiary base: the part of the programme that can be published without disclosing counterparties, candidate sites, or commercial terms.

The programme’s method should be stated before its findings, because the method is the product. Between July 2025 and February 2026 the lab commissioned seven research sweeps — Atlantic grid structure (BGL Research Sweep AC-S1, 2025-07), the tidal record (AC-S2, 2025-08), offshore wind leasing (AC-S3, 2025-09), cold-climate cooling (AC-S4, 2025-10), subsea fibre (AC-S5, 2025-12), the sovereign compute market (AC-S6, 2026-01), and workforce and federal programs (AC-S7, 2026-02). Each sweep produced a sourced corpus; the corpora were fused into a single governed structure and reasoned over deterministically by D.A.E., the Derivation Algorithmic Engine at the core of the lab’s M.A.D. stack. Every quantitative claim in this document ties back through a derivation record to a named source in one of those corpora. Where two sources disagreed — and on tidal capacity, capital cost per megawatt, and grid carbon intensity they disagreed often — the disagreement was logged in a contradiction register rather than averaged away (BGL Contradiction Register AC-CR-2, internal, 2025-09; AC-CR-5, internal, 2026-01). The registers are part of the sealed record; the resolved figures are what appear below.

Three categories of material have been withheld. First, the siting register: the lab has scored 190 candidate sites across 23 axes in a sealed deterministic run (BGL Run Ledger AC-0640, internal, sealed), and neither the site identities nor the axis weights are published here, for reasons discussed in the register’s own section. Second, commercial soundings: conversations with onshore wind developers, provincial officials, and prospective anchor tenants are summarized only at the level of posture, never identity (BGL Power Sounding PS-AC-2, internal, anonymized, 2026-04; BGL Memorandum of Record MR-AC-2, internal, 2026-04). Third, measured network telemetry: the latency figures in this document are modeled, clearly labelled as such, and gated behind a vendor-measurement verification step before the lab will rely on them (BGL Corridor Memorandum AC-M-09, internal, 2026-02).

A word on figures. The lab’s editorial policy for public research requires that every external claim carry its attribution inline, that internal artifacts be identified as internal, and that figures be re-verified against primary sources before reliance (BGL Editorial Note ED-AC-1, internal, 2026-06). Electricity tariffs change on regulatory schedules; data-centre economics move quarterly; submarine cable systems are announced and cancelled. The numbers here were current when the sweeps closed. They are presented so a skeptical reader can re-run the argument, not so the argument can be taken on faith — the same standard the lab applies to its own engine, whose runs are bit-identical for identical inputs and whose findings carry their lineage with them (D.A.E. Determinism Attestation DA-2026-Q1, internal).

The decision to publish was itself deliberated. The restricted brief that preceded this document closed with an instruction not to forward, excerpt, or cite. The release determination (BGL Corridor Memorandum AC-M-11, internal, 2026-06) concluded that the evidentiary base — grid structure, price data, the tidal record, cooling climatology, fibre geography, market signals — is assembled entirely from public sources and gains nothing from secrecy, while the lab’s specific conclusions about where and with whom gain everything from it. The line between those two things is the line between this document and the five that follow it.

The Thesis

The thesis can be stated in nine words: cold power, cold compute — the export product is not electricity. Atlantic Canada is conventionally analyzed as an energy exporter, or more often as an energy problem: four small grids at the end of the continental system, one of them still burning coal for roughly 47% of its generation in 2023 (CER; NSUARB), all of them paying industrial power prices between 1.9 and 2.3 times Montréal’s (Hydro-Québec, 2025). The Corridor programme’s finding is that this framing looks at the right region through the wrong commodity. The region’s exportable asset is not the electron. It is the computation the electron performs, executed where the cooling is nearly free, landed onto fibre that reaches Europe faster than any route from Ashburn, Toronto, or Montréal, inside a jurisdiction whose data-protection status the European Commission has formally recognized (European Commission, 2024).

The argument stands on four legs, and it is worth being precise about what each leg is and is not. The first leg is thermodynamic: Halifax spends 8,668 hours a year — 98.9% of the year — below 27°C, the upper bound of the ASHRAE-recommended envelope for IT intake air, and requires mechanical cooling for roughly 92 hours a year (climate.OneBuilding.org TMYx 2011–2025; ASHRAE TC 9.9; BGL Engineering Note EN-2025-19, internal, 2025-11). That supports annualized power usage effectiveness in the 1.1–1.2 range against 1.4–1.6 for warm-climate builds — roughly 5–10% of total campus energy, every year, from geography alone (Uptime Institute, 2024; Google; BGL Research Sweep AC-S4, 2025-10). The second leg is contractual: the bankable path to clean power in the region is dedicated onshore wind under power purchase agreement, not grid supply — Nova Scotia has roughly 620 MW operating with a further gigawatt targeted by 2030, and the first direct-to-buyer structures have already appeared (Government of Nova Scotia; CBC; BGL Power Sounding PS-AC-2, internal, 2026-04). The third leg is geographic: the EXA Express system — 4,600 km, six fibre pairs, roughly 53 Tb/s of design capacity — lands at Halifax and anchors modeled round-trip times of ~40–46 ms to Dublin and ~48–52 ms to London, some 15–30 ms inside anything achievable from the major North American metros (EXA Infrastructure; TeleGeography; BGL Run Ledger AC-0601, internal, sealed). The fourth leg is jurisdictional: Canada holds a standing EU adequacy decision, renewed January 2024, that the United States functionally does not, and Canadian-owned operators sit outside the reach of the US CLOUD Act that follows American hyperscalers onto Canadian soil (European Commission, 2024; CLOUD Act, 2018; BLG, 2026).

Equally important is what the thesis does not rest on, because the region’s promoters have historically rested on exactly these things. It does not rest on tidal power: the Bay of Fundy’s Minas Passage holds an extractable resource of roughly 2,000–2,500 MW — more than Nova Scotia’s entire demand — and after fifteen years of attempts the contracted pipeline stands at approximately 16.5 MW, none of it yet delivering sustained power (FORCE; OERA; NRCan; BGL Corridor Memorandum AC-M-04, internal, 2025-09). It does not rest on grid supply in Nova Scotia, where large-industrial power costs roughly 13.43¢/kWh all-in and carries about 660 g CO2e/kWh — some 6.6 times the national grid average — making default utility service simultaneously a cost and an ESG liability (Hydro-Québec, 2025; CER, 2022 data). And it does not rest on any capital-cost advantage from cold: conventional hyperscale construction runs USD 9–11.3 million per megawatt and AI-dense builds exceed USD 20 million, with cold-climate proxies like Oslo at 12.4 $/W against 9.8 $/W in cheap US metros — cold is an operating advantage, not a construction one (CBRE, 2025; Epoch AI; BGL Contradiction Register AC-CR-5, internal, 2026-01).

The inversion at the heart of the thesis deserves one more sentence of plain statement. Everywhere in the world, the binding constraint on compute is now power — interconnection queues in the United States exceed 2,060 GW, connection waits run four to seven years, and transformer lead times exceed 160 weeks (LBNL, 2024; Data Center Knowledge). Atlantic Canada cannot win that game on grid headroom; Nova Scotia’s entire system is roughly 2.5 GW of capacity against a 2.2 GW peak, so even a 100 MW campus is several percent of the province (CER; NERC LTRA). The Corridor thesis is that this weakness is the design constraint, not the disqualifier: a campus built as a dedicated-generation project — wind under PPA, storage, grid firming only — is precisely the co-developed firm-power structure the market’s largest buyers are now siting around, and the value of that power is multiplied, not merely transmitted, when it leaves the region as computation rather than as electrons across interties that do not exist (BGL Revenue Model RM-AC-3, internal, 2026-04). The sections that follow take each element of that claim apart in order: the grids, the megaproject record, the tidal saga, the wind pipeline, the cooling physics, the fibre shore, the export logic, the demand signal, the workforce, and finally the deterministic register the lab built to decide where — if anywhere — the thesis should touch ground.

Four Grids at the End of the Line

Under Section 92A of the Constitution Act, 1867, electricity is provincial jurisdiction, and Atlantic Canada expresses that fact as four small, structurally distinct systems: Nova Scotia, New Brunswick, Prince Edward Island, and Newfoundland and Labrador (Constitution Act, s.92A; CER). None operates a wholesale market. All four are vertically integrated — a private monopoly in Nova Scotia, Crown corporations in New Brunswick and Newfoundland and Labrador, and in Prince Edward Island a Fortis-owned distributor that imports nearly everything it sells. Understanding the Corridor thesis requires understanding each of them, because the thesis draws power posture, firming, and political geography from all four.

Nova Scotia is the largest Maritime load and the hardest case. Nova Scotia Power, a subsidiary of Emera, operates roughly 2.5 GW of capacity against a winter peak near 2.2 GW, and in 2023 still generated about 47% of its electricity from coal at Lingan, Point Aconi, Trenton, and Point Tupper (CER; NSUARB; Emera). The consequence is the worst large-grid carbon intensity in the country — roughly 660 g CO2e/kWh in 2022 against a national average near 100 g (CER, 2022 data) — and industrial power among the most expensive in North America: Hydro-Québec’s 2025 survey puts Halifax large-power all-in costs at 13.43¢/kWh, versus 5.83¢ in Montréal, with the utility’s Large General tariff at 10.44¢/kWh before riders (Hydro-Québec, 2025; Nova Scotia Power tariffs). Federal regulation requires coal off the system by 2030 (ECCC; Canada Gazette SOR/2024-263), and the province has legislated an 80% renewable target for the same year — which is why Nova Scotia is simultaneously the region’s most stressed grid and its most aggressive wind procurer. The lab’s tariff-normalization run, which restated all four provinces’ industrial rates on a common all-in basis, confirmed the survey’s ordering within a few percent (BGL Run Ledger AC-0512, internal, sealed).

New Brunswick is the hub. NB Power’s fleet is the most diverse in the region — the 660 MW Point Lepreau CANDU station, the 668 MW Mactaquac hydro plant now committed to life extension, the Belledune coal unit facing the same 2030 deadline, and oil-fired standby at Coleson Cove (NB Power; CER). Its intensity, roughly 330 g CO2e/kWh in 2022, is half Nova Scotia’s (CER). More important than its fleet is its position: New Brunswick holds the region’s only meaningful interconnections — on the order of a gigawatt toward Québec, a gigawatt toward New England, and roughly 300 MW toward Nova Scotia — making it the transit province for every Maritime import, export, and firming transaction (CER; NERC LTRA; NPCC). Its N-4 large-industrial tariff, 7.85¢/kWh energy plus $19.98/kW demand, lands Moncton at 11.13¢/kWh all-in in the 2025 survey (NB Power tariffs; Hydro-Québec, 2025). For a campus that firms dedicated wind against the grid, New Brunswick’s mix and its interties make it the natural firming counterparty — a point the restricted siting register weighs explicitly (BGL Run Ledger AC-0640, internal, sealed).

Prince Edward Island is the miniature that proves the pattern. The Island generates a large share of its own energy from wind — on a generation basis its intensity was roughly 2 g CO2e/kWh in 2022 — but imports the majority of its electricity from New Brunswick over two 180 MW submarine cables completed in 2017, because wind without firming is not supply (CER; Maritime Electric). Charlottetown’s 11.27¢/kWh all-in large-power cost sits between Moncton and Halifax (Hydro-Québec, 2025). The municipal utility in Summerside, which pairs its own wind with storage and demand programs, is a small working demonstration of the wind-plus-firming structure the Corridor thesis scales (Summerside Electric). PEI matters to the thesis less as a siting candidate than as a governance exhibit: a jurisdiction that decarbonized its generation identity while remaining wholly dependent on a neighbour’s wires.

Newfoundland and Labrador is the region’s energy giant and its cautionary institution, treated at length in the next section. Its island grid runs near 17 g CO2e/kWh — hydro-dominated, with the oil-fired Holyrood plant in staged retirement as Muskrat Falls energy displaces it (CER; NL Hydro). Its industrial rates are the region’s lowest, but its exportable surplus is spoken for by contract and by the province’s own rate-mitigation needs, and its interconnection to the Maritimes is a single 500 MW subsea link (Emera; NL Hydro). Labrador, meanwhile, hosts 5,428 MW at Churchill Falls whose disposition after the 1969 contract is the subject of the December 2024 memorandum of understanding with Hydro-Québec — a document that will shape eastern Canadian power flows for decades and that the Corridor programme tracks as an exogenous variable, not a dependency (Government of NL; Hydro-Québec; CBC).

Two structural observations close the survey. First, the four systems are regulated separately — NSUARB, the New Brunswick EUB, IRAC, and the NL PUB — with no regional operator, no common tariff logic, and no market interface; every cross-border arrangement is bilateral and contractual (CER; BGL Research Sweep AC-S1, 2025-07). Second, the region’s aggregate is small: roughly 2.6 million people (StatCan), a combined peak on the order of 6 GW, and interties measured in hundreds of megawatts. Analyses that treat Atlantic Canada as a power-export platform run into both facts immediately. The Corridor thesis is built to route around them: dedicated generation avoids the tariff problem, provincial siting avoids the regional-coordination problem, and exporting computation avoids the intertie problem entirely (BGL Corridor Memorandum AC-M-01, internal, 2025-06).

Muskrat Falls and the Cost of Moving Electrons

No Atlantic energy analysis is credible without confronting Muskrat Falls, because Muskrat Falls is the controlled experiment the region already ran on the question the Corridor thesis answers differently: what does it cost to turn remote Atlantic energy into delivered value? The project’s answer — an 824 MW hydro station on the Lower Churchill River, sanctioned in December 2012 at roughly $7.4 billion and completed at approximately $13 billion including financing — is among the most expensive lessons in Canadian energy history on a per-capita basis, borne by a province of half a million people (Commission of Inquiry Respecting the Muskrat Falls Project, 2020; CBC; NL Hydro).

The delivery infrastructure is the part the Corridor programme studies most closely. Getting Muskrat’s energy to consumers required the Labrador-Island Link — roughly 1,100 km of HVDC from Muskrat Falls to Soldiers Pond on the Avalon Peninsula, including the Strait of Belle Isle marine crossing — and then the Maritime Link, a further 500 MW HVDC system with about 170 km of subsea cable across the Cabot Strait to Nova Scotia, built by Emera at roughly $1.6 billion and in service in 2018 (NL Hydro; Emera). The links worked, eventually; the Labrador-Island Link’s software-plagued commissioning ran years behind the generation station itself, and full reliable operation was not declared until 2023 (CBC; NL Hydro). In exchange for financing the Maritime Link, Nova Scotia receives the Nova Scotia Block — roughly 20% of Muskrat’s energy over 35 years — which is now a material clean-energy input to a coal-heavy grid (Emera; NSUARB). The all-in arithmetic remains sobering: on the order of $15 billion of generation and transmission capital, across two provinces and two subsea crossings, to move under a gigawatt of firm clean power to markets measured in single-digit terawatt-hours.

The institutional aftermath matters as much as the engineering. Commissioner LeBlanc’s inquiry, reporting in 2020 under the title Muskrat Falls: A Misguided Project, found that the project proceeded on suppressed risk assessments, optimism-biased cost estimates, and governance that insulated the proponent from challenge (Commission of Inquiry, 2020). The rate consequences were severe enough that Ottawa restructured the project’s finances in July 2021 with a $5.2 billion package to prevent electricity rates from absorbing the overrun (Government of Canada; CBC). The lab’s reading, recorded early in the programme, is that Muskrat is not an argument against Atlantic energy development — it is an argument about method: single-option analysis, unfalsifiable benefit claims, and the absence of any mechanism by which a skeptical party could re-run the proponent’s numbers (BGL Research Sweep AC-S1, 2025-07; BGL Corridor Memorandum AC-M-01, internal, 2025-06). Those are precisely the failure modes a deterministic evidence discipline exists to remove, and the Corridor programme’s verification-gate structure — no capital movement without named, independently checkable gates — is written in Muskrat’s shadow (BGL Verification Gate Schedule VG-AC-1, internal, 2026-05).

The forward story sharpens rather than softens the lesson. In December 2024, Newfoundland and Labrador and Hydro-Québec signed a memorandum of understanding that would replace the 1969 Churchill Falls contract — under which Québec has purchased most of the plant’s 5,428 MW at fractions of a cent per kilowatt-hour — with payments rising toward a billion dollars a year and joint development of new Labrador capacity, including the long-deferred 2,250 MW Gull Island project (Government of NL; Hydro-Québec; CBC). If consummated, the arrangement points Labrador’s next tranche of energy west and south through Québec’s system, not east through new Atlantic wires. The regional intertie ambitions tell the same story: the Atlantic Loop — the proposed Québec-to-Maritimes transmission reinforcement — collapsed in October 2023 over cost allocation, surviving only as the narrower CIB-financed 345 kV Wasoqonatl intertie between New Brunswick and Nova Scotia (CBC; Canada Infrastructure Bank).

For the Corridor thesis, the Muskrat record yields three planning constants. First, subsea and long-distance transmission in this region costs billions per gigawatt and a decade per attempt, and its political risk compounds its engineering risk — so any thesis that requires new bulk wires to market is a thesis about the 2040s (BGL Research Sweep AC-S1, 2025-07). Second, clean firm supply that already exists in the region — the Nova Scotia Block, Lepreau’s output, Muskrat’s surplus in wet years — is contractually and politically committed; a compute campus cannot plan around acquiring it and should plan around not needing it (BGL Power Sounding PS-AC-2, internal, anonymized, 2026-04). Third, and decisively: over the exact period Muskrat spent $13 billion to move 824 MW of value to market over wires, the fibre systems landing on the same coast moved terabits per second of value to two continents at a capital cost orders of magnitude lower per delivered dollar. The region has already built the cheap export corridor. It is not electrical (EXA Infrastructure; TeleGeography; BGL Corridor Memorandum AC-M-11, internal, 2026-06).

The Tidal Saga

If the Corridor programme has a founding discipline, it was learned from the Bay of Fundy. The bay holds the highest tides on Earth, and the Minas Passage — the five-kilometre throat through which the Minas Basin fills and empties twice daily — carries a kinetic resource assessed at several thousand megawatts, of which roughly 2,000–2,500 MW is considered extractable: more than Nova Scotia’s entire peak demand, renewable, and, uniquely among renewables, predictable to the minute for centuries ahead (NRCan; FORCE; OERA). On paper it is the finest clean-energy resource in eastern North America. The record of attempts to harvest it is the reason this document treats every energy claim in the region with suspicion until it survives a contradiction register.

The saga begins earlier than most accounts note. The Annapolis Royal generating station — a 20 MW tidal barrage commissioned in 1984 as North America’s only tidal power plant — ran for thirty-five years before closing in 2019, its turbine implicated in fish mortality and its basin ecology permanently altered; it never scaled and never spawned a successor (Nova Scotia Power; CBC). The modern in-stream era opened with the creation of the Fundy Ocean Research Centre for Energy, a purpose-built test facility in the Minas Passage with five subsea berths and tens of megawatts of installed transmission capacity — infrastructure that remains, two decades of effort later, the most valuable physical asset the sector has produced (FORCE; NRCan).

The corporate record is brutal in its consistency. OpenHydro, the Irish developer backed by Naval Energies, deployed a two-megawatt open-centre turbine at FORCE and went into liquidation in 2018, leaving roughly C$5.9 million in unpaid regional debts and the turbine itself abandoned on the seabed of the Minas Passage, where it remains (CBC; Global News). An earlier deployment of the same design had been destroyed in days by the Passage’s currents — a resource so strong it shredded the machine built to harvest it. Sustainable Marine Energy then achieved what no one else had: Canada’s first floating tidal power delivered to the grid, from its PLAT-I platform in Grand Passage. In 2023 the company ceased Canadian operations and entered insolvency anyway, publicly attributing the decision to an impassable federal permitting process under the Fisheries Act — its chief executive stating the company could not obtain authorization to deploy at FORCE despite years of monitoring data (CBC; DFO). The pattern is not one failure but a fifteen-year sequence of them, each with a different proximate cause — engineering, finance, regulation — and a single common outcome: the bay has never sustained even five megawatts of continuous delivery (BGL Research Sweep AC-S2, 2025-08).

The regulatory anatomy deserves precision, because it is frequently caricatured. Nova Scotia built one of the world’s more thoughtful marine-energy frameworks — the Marine Renewable-electricity Act of 2015 created demonstration permits and feed-in structures specifically for the sector (Government of Nova Scotia). The binding constraint sits federally: in-stream turbines require Fisheries Act authorization from the Department of Fisheries and Oceans, which must assess harm to fish populations — including striped bass and inner Bay of Fundy Atlantic salmon — in an environment so energetic that the monitoring instruments themselves frequently fail, making the evidentiary standard difficult to satisfy in principle (DFO; OERA). The result is a genuine impasse rather than simple obstruction: the proponent cannot prove absence of harm without deploying, and cannot deploy without proving absence of harm. The lab’s contradiction register on tidal capacity claims logged a dozen materially different public figures for the same resource and pipeline, a dispersion itself diagnostic of a sector where promotional and engineering numbers had decoupled (BGL Contradiction Register AC-CR-2, internal, 2025-09).

The current chapter is modestly real again. In 2025, new berth awards at FORCE went to Eauclaire Tidal and Orbital Marine Power under fifteen-year power purchase agreements — new awards totalling roughly 12.5 MW that bring the site’s contracted pipeline to approximately 16.5 MW (FORCE; Government of Nova Scotia). Orbital’s floating O2 platform has operated commercially in Scotland’s Fall of Warness, which makes this the first FORCE cohort arriving with a proven at-sea machine rather than a prototype. A site visit by programme staff to the Minas Passage shoreline in March 2026 recorded the state of the facility directly (BGL Site Visit Log SV-017, internal, 2026-03).

The programme’s posture follows from the record, and it is deliberately unromantic. Tidal is perfectly predictable physics attached to brutally unpredictable engineering and regulation, and a siting analysis must treat it as exactly that: long-dated optionality, worth locating near, worth zero in any bankable power plan before the mid-2030s (BGL Corridor Memorandum AC-M-04, internal, 2025-09). In the sealed siting register, proximity to the Fundy tidal infrastructure appears as a positive axis with a deliberately small weight — an option premium, not a supply assumption — and the register’s method suppresses the axis entirely for any scenario dated before 2034 (BGL Run Ledger AC-0640, internal, sealed; BGL Site Axis Codebook, rev. 4, internal, 2026-03). If the Eauclaire and Orbital deployments deliver sustained power through a full winter, that weight will be revisited. Not before.

Offshore Wind and the Onshore Bridge

Where tidal supplies the cautionary tale, wind supplies the plan — provided the two timescales it runs on are kept rigorously separate. Offshore wind is Atlantic Canada’s 2030s story, and it is genuinely large; onshore wind is the 2020s story, and it is the only power supply the Corridor thesis allows itself to bank on. Conflating them is the single most common analytical error in regional energy commentary, and the programme’s wind sweep was structured specifically to keep the ledgers apart (BGL Research Sweep AC-S3, 2025-09).

The offshore framework is real and recent. The 2024 amendments to the federal-provincial Accord Acts — Bill C-49, which received royal assent in October 2024 — converted the offshore petroleum boards into offshore energy regulators, giving Nova Scotia and Newfoundland and Labrador a licensing regime for seabed wind for the first time (Parliament of Canada; Canada–Nova Scotia Offshore Energy Regulator). A federal-provincial regional assessment of offshore wind development, completed in 2023, mapped the candidate areas and the conflicts — fisheries above all — before any licence was issued (IAAC, 2023). On that foundation Nova Scotia has set a policy of licensing up to 5 GW of offshore wind by 2030, and the first call for bids, covering 2.5 GW, qualified seven bidders in June 2026 (Government of Nova Scotia; CBC). Beyond the licensing programme sits the province’s far larger ambition — the “Wind West” concept, a vision of tens of gigawatts of Atlantic offshore wind serving national demand, with figures near $60 billion attached in public discussion (CBC; Government of Nova Scotia).

The programme’s reading of the offshore file is respectful and cold. Licensed is not built: no Canadian offshore turbine exists, no marshalling port has been constructed, the supply chain would be imported into competition with American and European pipelines, and the fisheries conflicts documented in the regional assessment — in a region where the lobster fishery is the economic anchor of hundreds of communities — are unresolved (IAAC, 2023; DFO). Comparable jurisdictions run seven to ten years from licence to first power under mature regulators. The sweep’s conclusion, carried into the sealed brief and preserved here: first Nova Scotian offshore turbines are realistically a mid-2030s event, and offshore wind is therefore an expansion path for a campus’s second decade — a reason the region’s power story gets stronger with time — not a source of 2020s electrons (BGL Research Sweep AC-S3, 2025-09; BGL Corridor Memorandum AC-M-07, internal, 2026-01).

The onshore ledger is where the bankable megawatts live. Nova Scotia has roughly 620 MW of onshore wind operating and a further gigawatt targeted by 2030 through provincial procurement — the Green Choice Program and rate-base solicitations — with the 168 MW Goose Harbour Lake project energizing in 2026 as the current procurement’s flagship (Government of Nova Scotia; CBC). More structurally significant than any single farm is a contracting innovation: the Mersey River project’s direct-to-buyer sales structure, the first in the province to sell wind output directly to a corporate offtaker rather than into the utility’s rate base — the exact contractual chassis a compute campus PPA requires, now demonstrated under Nova Scotian regulation (CBC; BGL Power Sounding PS-AC-2, internal, anonymized, 2026-04). New Brunswick’s onshore fleet stands near 355 MW with roughly 452 MW more contracted for 2027–28 delivery, much of it in partnership with First Nations co-owners (NB Power; CER). Across both provinces, the lab’s soundings with developers found consistent appetite for campus-scale offtake: a creditworthy 100–300 MW buyer with a twenty-year horizon is the counterparty every Maritime wind developer’s financing model wants and the region’s small utilities cannot be (BGL Power Sounding PS-AC-2, internal, anonymized, 2026-04).

The arithmetic that matters for the thesis is capacity-factor arithmetic. A 100 MW campus at high utilization consumes approximately 876 GWh a year (BGL Revenue Model RM-AC-3, internal, 2026-04). Maritime onshore wind at a 40–45% capacity factor yields roughly 3.5–3.9 GWh per installed megawatt-year, so covering that campus on an annual-energy basis requires on the order of 230–250 MW of dedicated nameplate — comfortably within a two-project portfolio of current Maritime scale — before firming (BGL Research Sweep AC-S3, 2025-09). Firming is the honest residual: wind’s hourly intermittency against a 24/7 load means storage plus grid backstop, and the grid backstop is exactly where the thesis touches the expensive, carbon-heavy tariffs documented earlier. The programme’s modelling treats the firming fraction — not the headline PPA price — as the number that decides the campus’s effective cost and effective carbon intensity, and the first named verification gate of the entire programme is a wind-PPA term sheet whose firming terms survive scrutiny (BGL Verification Gate Schedule VG-AC-1, internal, 2026-05; BGL Corridor Memorandum AC-M-07, internal, 2026-01). Nothing in the offshore story is needed for that gate to pass. Everything in the offshore story makes the second decade better if it does.

The Cold Calculus

The strongest leg of the thesis is the one that requires no counterparty, no regulator, and no construction: the climate. Data-centre cooling economics turn on a single question — for how many hours a year can outside air, or water cooled by outside air, absorb the facility’s heat without mechanical refrigeration — and Atlantic Canada’s answer is close to the best on the continent. Using typical-meteorological-year files built from 2011–2025 observations, Halifax sits below 27°C — the upper bound of the ASHRAE-recommended intake envelope for IT equipment — for 8,668 hours a year, 98.9% of all hours, with an annual maximum of 30.0°C; mechanical cooling is needed for roughly 92 hours a year (climate.OneBuilding.org TMYx; ECCC Canadian Climate Normals; ASHRAE TC 9.9). Moncton reads 8,569 hours below 27°C, 97.8% of the year (climate.OneBuilding.org TMYx). For evaporative and water-side systems the wet-bulb statistic governs, and both sites hold wet-bulb temperatures below 15°C for roughly 83% of all hours (BGL Engineering Note EN-2025-22, internal, 2025-11). The lab did not take the sweep’s figures on trust: the economizer-hour numbers were recomputed independently from the raw hourly files before entering the governed corpus, and the recomputation is the version cited here (BGL Engineering Note EN-2025-19, internal, 2025-11; BGL Run Ledger AC-0575, internal, sealed).

Translated into engineering economics, those hours support annualized power usage effectiveness in the 1.1–1.2 range for a competently designed facility — against roughly 1.4–1.6 for warm-climate builds and a global survey average that has been stuck near 1.55 for years (Uptime Institute, 2024). The world’s benchmark operators show what the ceiling looks like: Google reports a fleet-wide trailing PUE of 1.09, Meta’s Luleå campus in subarctic Sweden runs near 1.07, and Icelandic and Nordic operators publish 1.05–1.2 (Google; Meta; atNorth). The delta between 1.15 and 1.5 is roughly 5–10% of total campus energy — every year, compounding, at whatever the campus’s power price is — plus a capital-side bonus that is easy to miss: the worst-case chiller plant, sized for the hottest hour of the design year, shrinks dramatically when the design maximum is 30°C rather than 40°C (ASHRAE Standard 90.4; BGL Research Sweep AC-S4, 2025-10). At the 876 GWh annual consumption of a reference 100 MW campus, a seven-point PUE saving is on the order of 50–60 GWh a year that never has to be generated, contracted, or firmed (BGL Revenue Model RM-AC-3, internal, 2026-04).

The honest ledger requires the caveats, and the programme’s engineering notes spend more pages on them than on the headline. First, maritime air is corrosive: salt aerosol places coastal sites in elevated ISO 9223 corrosivity categories, which pushes the design toward indirect or filtered economization — heat exchange without introducing outside air to the data hall — and toward marine-grade materials for everything exposed (ISO 9223; BGL Engineering Note EN-2026-03, internal, 2026-02). Indirect economization surrenders a point or two of PUE against direct free cooling; the modeled 1.1–1.2 band already assumes it. Second, fog and humidity duration in Halifax specifically complicate direct evaporative designs — another argument for the indirect topology. Third, the industry the campus would serve is densifying toward liquid: AI-class racks at 80–130 kW are cooled direct-to-chip, which changes where the heat is captured but not where it must be rejected — and heat rejection is precisely where cold climate pays, since dry coolers with cold approach temperatures replace chiller plants at the loop’s far end (LBNL, 2024; Open Compute Project; BGL Cold-Intake Bench CI-77, internal, 2026-05). Fourth, the Bay of Fundy adds a genuine water-side option — cold seawater as a heat sink, with intake ecology and biofouling as the engineering price — which the programme carries as an unweighted design alternative rather than a plan (BGL Research Sweep AC-S4, 2025-10).

What cold does not do is cheapen construction, and the programme’s contradiction register was built partly to kill this specific piece of promotional arithmetic. Conventional hyperscale capacity costs roughly USD 9–11.3 million per megawatt to build in 2025–26; AI-dense capacity exceeds USD 20 million per megawatt; and the cold-climate proxy the sweep examined most closely — Oslo — prices at 12.4 $/W against roughly 9.8 $/W in the cheapest US metros, because labour, logistics, and site development in small northern markets offset the cooling plant’s savings (CBRE, 2025; Epoch AI; BGL Contradiction Register AC-CR-5, internal, 2026-01). A reference 1 GW AI campus runs to roughly USD 38 billion up front with all-in total cost of ownership near USD 8.5 million per megawatt-year (Epoch AI). Atlantic Canada should expect the Oslo pattern, not the Texas one. The thesis therefore books cold as what it is: a permanent operating-cost and carbon advantage of 5–10% of total energy, a smaller resilience-critical chiller plant, and nothing more (BGL Corridor Memorandum AC-M-11, internal, 2026-06).

One more climatic fact belongs in this section because it cuts against the region: storms. Post-tropical storm Fiona in September 2022 was the most damaging storm in the region’s recorded history, with weeks-long distribution outages across Nova Scotia and Prince Edward Island (ECCC; Canadian Hurricane Centre). For a campus, Fiona is a design case rather than a disqualifier — transmission-level interconnection, on-site firming, and hardened construction are already in the reference design for other reasons — but the programme treats storm exposure as a scored siting axis with Fiona’s wind and surge fields as the benchmark overlay, and several otherwise attractive coastal candidates in the sealed register lose material rank to it (BGL Engineering Note EN-2026-07, internal, 2026-04; BGL Run Ledger AC-0640, internal, sealed). Cold that arrives with hurricanes is still cold; the register simply prices the hurricanes.

The Latency Shore

Geography gave Atlantic Canada a second endowment, and unlike the tides this one is already fully industrialized: it is the closest point on the North American mainland to Europe, and the transatlantic fibre system knows it. The great-circle mathematics are fixed physics — light in fibre propagates at roughly 4.9 microseconds per kilometre, so every thousand kilometres of route saved is roughly ten milliseconds of round trip — and Halifax sits some 800–1,000 km closer to Ireland than New York is, before a single routing decision is made (TeleGeography; BGL Research Sweep AC-S5, 2025-12).

The incumbent asset is the EXA Express system — built as Hibernia Express in 2015 expressly to be the fastest transatlantic route, and still among the fastest in service: roughly 4,600 km, six fibre pairs, on the order of 53 Tb/s of design capacity, with New York–London round-trip times of 58.55–58.95 ms marketed to the latency-sensitive trading industry that financed it (EXA Infrastructure; TeleGeography). The system’s western landfall is on the Nova Scotian shore near Halifax — which means the segment every New York financial packet traverses to reach London passes the Corridor’s front door. A facility interconnecting at the Halifax landing rather than backhauling to it removes the entire New-York-to-Nova-Scotia leg from the path to Europe. The programme’s latency model, built on published cable routes and standard propagation constants, puts Halifax–Dublin at roughly 40–46 ms round trip and Halifax–London at roughly 48–52 ms — approximately 15–30 ms inside what is achievable from Ashburn, Toronto, or Montréal, whose packets must first cross their own continent-side backhaul before entering the same cables (BGL Run Ledger AC-0601, internal, sealed; TeleGeography). These are modeled figures. The programme’s own rules forbid reliance on them until vendor-measured round-trip times replace them, and that measurement is a named verification gate, not a formality (BGL Corridor Memorandum AC-M-09, internal, 2026-02; BGL Verification Gate Schedule VG-AC-1, internal, 2026-05).

The corridor is thickening rather than thinning. Amazon’s Fastnet system — announced as a dedicated transatlantic cable from Maryland to County Cork, targeting 2028 service — confirms that hyperscale operators are building fresh North-America-to-Ireland capacity on exactly this great circle (AWS). Google’s Grace Hopper and the Amitié consortium’s Boston–Bordeaux system renewed the US-side corridor in 2022–23 (Google; TeleGeography). And the long-proposed Leif Erikson project — a direct Canada–Ireland cable — remains the sector’s recurring sketch of the route the geography wants: a purpose-built Atlantic-Canada-to-Europe system with no US landing at all, a design whose jurisdictional value the sovereignty section makes plain (TeleGeography; BGL Research Sweep AC-S5, 2025-12). The programme treats all announced systems as available infrastructure, not deliverables; the sealed brief’s instruction on this point — this is not a proposal to build a transatlantic communications link — survives into the public edition unchanged.

Two sobering observations keep the section honest. First, latency is a differentiator only for the workloads that feel it: interactive inference serving European users, financial and market infrastructure, synchronous replication for sovereignty-bound European data — a real and growing set, but not the whole market; batch training does not care where the checkpoint lives (BGL Research Sweep AC-S6, 2026-01). Second, a landing point is not an interconnection ecosystem. Halifax’s internet exchange is small; the region’s carrier diversity is thin next to Ashburn’s or Montréal’s; and the national research network’s regional segments, while present, are not a commercial fabric (Halifax Internet Exchange; CANARIE). The thesis is that a campus of sufficient scale creates the ecosystem it needs — carriers follow anchor tenants — but the programme logged this as an honest chicken-and-egg risk, informed by an interview with a subsea maintenance contractor on the realities of remote-landing operations (BGL Interview Log IL-051, internal, anonymized, 2026-01; BGL Risk Register RR-AC-1, internal, 2026-05). The cartographic work supporting the sealed register — substation adjacency on one layer, inferred fibre conduit on another — exists precisely to find the small set of sites where the power geography and the fibre geography intersect (BGL Cartographic Layer CL-9, internal, 2026-02; CL-11, internal, 2026-03).

The Export Product Is Not Electricity

The sections above assemble the parts; this one states the machine. Atlantic Canada’s energy debate has been conducted for fifty years in a single vocabulary — generation, transmission, export — and every major disappointment in the regional record, from the Fundy barrage studies to Muskrat’s economics to the Atlantic Loop’s collapse, traces to the same structural fact: the region’s energy is remote from load, and wires to load cost billions and decades (Commission of Inquiry, 2020; CBC; BGL Research Sweep AC-S1, 2025-07). The Corridor thesis dissolves the problem instead of solving it. If the energy cannot economically travel to the demand, move the demand to the energy — and export the output of the demand over infrastructure that already exists, at near-zero marginal transport cost, to the highest-value buyers on Earth.

The precedent is a century old and the region’s neighbours are built on it. Québec’s and British Columbia’s aluminum industries exist because smelters were sited at remote hydro rather than hydro being wired to smelters; the electricity was exported as metal (NRCan; StatCan). Iceland ran the same play twice — first aluminum, then data centres — precisely because its grid is islanded and its power could never reach a foreign socket (atNorth; IEA). Computation is the most extreme case of the embodied-energy export pattern yet devised: the product moves at light speed, the transport network is already amortized, and the value density per megawatt-hour is unmatched by any physical commodity (BGL Corridor Memorandum AC-M-11, internal, 2026-06).

The arithmetic deserves to be stated in one place. A 100 MW campus consumes roughly 876 GWh a year (BGL Revenue Model RM-AC-3, internal, 2026-04). Sold as electrons at a healthy Maritime export price near 5¢/kWh, that energy is worth roughly $44 million a year — if interties existed to carry it, which they do not: Nova Scotia’s external ties are the 500 MW Maritime Link and roughly 300 MW toward New Brunswick, both spoken for (Emera; CER; NERC LTRA). Sold as wholesale colocation, the same energy footprint earns on the order of USD 1.1 million per megawatt-year — roughly $110 million annually at 100 MW — against North American wholesale benchmarks near $196/kW-month and Frankfurt at $235–265 (Vantage; CBRE, 2025). Sold as GPU capacity, the multiplier climbs again: market rates run $2–3 per H100-hour, near $5 per B200-hour, and $10–18 per GB200-hour, with take-or-pay structures like the $21 billion CoreWeave–Meta commitment demonstrating what anchored demand pays for committed capacity (CBRE; Data Center Knowledge; BGL Research Sweep AC-S6, 2026-01). The same megawatt-hour is worth roughly 2.5× as colocation and an order of magnitude more as accelerated computation than it is as an exported electron — and the electron export requires transmission that would take a decade to build, while the computation export requires fibre that has been in the water since 2015 (EXA Infrastructure; BGL Revenue Model RM-AC-3, internal, 2026-04).

Power price remains the swing variable, and the programme models it without sentiment: at a contracted 4.7¢/kWh the reference campus’s energy bill is roughly $41 million per 100 MW-year; at Nova Scotia’s 15¢-class retail-adjacent rates it exceeds $131 million — a $90 million annual swing that is the difference between a defensible business and none (Hydro-Québec, 2025; BGL Revenue Model RM-AC-3, internal, 2026-04). This is why the PPA is the business case, why the first verification gate is a term sheet rather than a rendering, and why the thesis’s recommended structure anchors 40–60% of capacity on take-or-pay with a sovereign or regulated tenant to secure debt, leases a wholesale tranche to European regulated buyers at a sovereignty premium, and retains a GPU-service tranche for spread capture (BGL Revenue Model RM-AC-3, internal, 2026-04; BGL Corridor Memorandum AC-M-07, internal, 2026-01).

The framing also answers the objection the region will raise first: that a campus consumes clean energy the grid needs for decarbonization. The answer is structural, not rhetorical. The thesis is a dedicated-generation project — new wind, contracted additionally, that exists because the campus financed it — with the grid used for firming alone; it competes with provincial decarbonization only at the margin of the firming fraction, and a well-structured campus accelerates the wind build-out the province’s own 2030 targets require by supplying the creditworthy offtake Maritime developers currently lack (Government of Nova Scotia; BGL Power Sounding PS-AC-2, internal, anonymized, 2026-04). Everywhere else on the continent, compute demand is colliding with constrained grids — 2,060 GW of interconnection queue, four-to-seven-year waits, 160-plus-week transformer lead times (LBNL, 2024; Data Center Knowledge). The Corridor structure sidesteps the collision by never asking the grid for what the grid does not have. Cold power, cold compute: the export product is not electricity, and was never going to be.

The Sovereign Demand Signal

A supply thesis is only as good as its buyer, and the Corridor programme spent a full sweep establishing that the buyer exists, is growing, and pays a premium for exactly what Atlantic Canada can offer (BGL Research Sweep AC-S6, 2026-01). The demand signal is sovereignty: the accelerating requirement that certain AI workloads run on infrastructure whose jurisdiction, ownership, and legal exposure the customer can prove. McKinsey estimates 30–40% of global AI infrastructure spending will be sovereignty-shaped by 2030 — on the order of USD 500–600 billion (McKinsey). Nvidia’s disclosures make the demand concrete from the vendor side: sovereign-AI revenue exceeded USD 30 billion in fiscal 2026, roughly tripling year over year to about 14% of the company’s revenue (Nvidia). Governments are building the counterpart supply with public money: Canada’s CAD 2 billion Sovereign AI Compute Strategy splits $700 million for private-sector capacity, $1 billion for public supercomputing, and $300 million for an access fund (ISED); the European Commission’s InvestAI initiative mobilizes roughly €200 billion including €20 billion for four to five AI gigafactories, atop EuroHPC’s roughly €10 billion across nineteen AI factories (European Commission; EuroHPC JU).

The buyer taxonomy matters more than the totals. The programme sorts demand into three tiers. Tier one is governments and defence establishments, which purchase jurisdiction as much as computation. Tier two is regulated industries — European banks, insurers, health systems — that are legally constrained from placing certain workloads on US-controlled public cloud and are the natural tenants of a wholesale tranche. Tier three is model developers needing committed capacity at national scale: Cohere’s $725 million Canadian facility, built by CoreWeave with a $240 million federal contribution, is the domestic template, and the June 2026 Bell–Cohere–Hypertec arrangement — a USD 220 million sovereign stack — is the second data point in the same series (Government of Canada; CBC; BGL Research Sweep AC-S6, 2026-01). Each tier maps onto a tranche of the recommended revenue structure: sovereign take-or-pay anchors the debt, regulated wholesale pays the premium, GPU services capture the spread (BGL Revenue Model RM-AC-3, internal, 2026-04).

Canada’s claim to a European premium rests on three legs, each verifiable. First, adequacy: the European Commission’s decision recognizing Canadian privacy law as adequate under the GDPR was reaffirmed in the Commission’s January 2024 adequacy review — meaning personal data flows from the EU to Canadian commercial operators without supplementary legal machinery, a status the United States has achieved only through successive frameworks that European litigation keeps placing at risk (European Commission, 2024). Second, the physical layer: roughly 85% non-emitting national generation and the cold-climate PUE documented above give Canadian compute a carbon denominator European procurement rules increasingly score (CER; ECCC; Uptime Institute). Third, jurisdictional positioning: Canada sits outside the US CLOUD Act’s compulsory reach while remaining inside the Western security perimeter — a neutral-but-allied posture that European buyers cannot get from an American operator anywhere, including on European soil (CLOUD Act, 2018; BLG, 2026).

The third leg carries the decisive fine print, and the programme’s threat model exists to keep it in view: the CLOUD Act follows ownership, not geography. Data held by a US-headquartered hyperscaler in a Canadian or European data centre remains reachable by US legal process; the sovereignty premium therefore attaches only to Canadian-owned operators, and evaporates the moment the operating entity’s control moves south (BLG, 2026; BGL Threat Model TM-AC-1, internal, 2026-02). That ownership gap is the market wedge — the reason a new Canadian operator can exist at all in a market otherwise owned by four American firms — and it is also a corporate-structure constraint the thesis imposes on any eventual operating company. An interview with EU procurement counsel conducted for the sweep put the point bluntly: European sovereign buyers now read the shareholder register before the spec sheet (BGL Interview Log IL-058, internal, anonymized, 2026-03).

The programme also logs the demand-side risks without flinching. The premium can compress: if InvestAI’s gigafactories land on schedule, European demand may be met in Europe, and the exportable residual shrinks to workloads that specifically want non-EU-but-adequate territory — a real category (redundancy, neutrality, cost) but a narrower one (European Commission; BGL Risk Register RR-AC-1, internal, 2026-05). Adequacy is reviewed periodically, and while Canada has held the status since 2002, a future review is a genuine tail risk to the premium’s legal leg (European Commission, 2024). And sovereign procurement is slow, political, and lumpy — anchor tenancy from a government buyer is bankable when signed and worthless when promised. Which is why anchor-tenant soundings appear in the verification-gate schedule as a condition of capital, not a consequence of it (BGL Verification Gate Schedule VG-AC-1, internal, 2026-05).

Workforce, Institutions, and the Federal Lever

Infrastructure theses fail on people more often than on physics, and the programme’s seventh sweep examined whether a region of 2.6 million can staff, build, and politically sustain a campus of national significance (BGL Research Sweep AC-S7, 2026-02; StatCan). The demographic story is better than its reputation. After decades of outflow, Atlantic Canada has posted its strongest population growth in generations, driven substantially by the Atlantic Immigration Program — the employer-led permanent-residence pathway made permanent in 2022 specifically because the pilot outperformed — and by interprovincial migration that turned positive during the remote-work era and has partially persisted (StatCan; IRCC). Halifax specifically has been among the faster-growing metropolitan areas in the country, a fact its business development agency documents annually (Halifax Partnership).

The institutional base is deeper than the population suggests, because it was built for a different industry that needs the same skills. Dalhousie University anchors ocean science globally — the Ocean Frontier Institute, established with one of the largest federal research awards in Canadian history, spans Dalhousie and Memorial — and the region’s engineering pipeline runs through Dalhousie, the University of New Brunswick, Memorial, and the community college systems that supply the trades (Dalhousie; Memorial; UNB). Canada’s Ocean Supercluster, one of the five federal innovation clusters, is headquartered in the region and has spent years building exactly the marine-operations supply chain — subsea cable handling, coastal construction, remote monitoring — that a shore-adjacent campus and its wind portfolio require (Canada’s Ocean Supercluster). The Centre for Ocean Ventures and Entrepreneurship in Dartmouth concentrates the applied end of that ecosystem on one wharf (COVE). None of this is data-centre labour per se; all of it is adjacent, and the sweep’s labour-radius analysis found that the specialized construction trades — not operations staff — are the binding human constraint, consistent with every large build on the continent (BGL Labour Radius Study LR-AC-1, internal, 2026-02; LBNL, 2024).

The programme’s labour-radius work deserves a paragraph because it feeds the sealed register directly. For each candidate site, the study computed 60-minute commute isochrones and populated them with census labour-force data by occupation class — electricians, millwrights, HVAC and controls technicians, network engineers — then scored sites on both steady-state operations coverage (a 100 MW campus runs on the order of 50–150 permanent staff) and peak-construction coverage, where requirements in the low thousands strain any Maritime labour market and force a camp-or-commute decision that materially affects cost and community relations (BGL Labour Radius Study LR-AC-1, internal, 2026-02; StatCan). Operations coverage passed comfortably at most urban-adjacent candidates; construction coverage is the axis where Atlantic Canada genuinely trails the large metros, and the register scores it honestly rather than waving at immigration (BGL Run Ledger AC-0640, internal, sealed).

The federal lever is the region’s distinctive institutional asset. Atlantic Canada is the only region with its own standing federal development agency — the Atlantic Canada Opportunities Agency, whose Regional Economic Growth through Innovation programming funds exactly the ecosystem-building a campus anchors (ACOA). The national instruments stack on top: the Sovereign AI Compute Strategy’s private-capacity stream is purpose-built for Canadian-owned compute operators (ISED); the Cohere–CoreWeave precedent established that Ottawa will co-fund sovereign capacity at the quarter-billion scale (Government of Canada); and provincial payroll-rebate and investment structures in Nova Scotia have decades of practice landing anchor employers (Invest Nova Scotia). The programme’s posture on public money is disciplined: federal participation is treated as an accelerant and a political stabilizer, never as the load-bearing element of the capital stack, because programme dependence is one of the failure patterns the Muskrat record teaches (Commission of Inquiry, 2020; BGL Memorandum of Record MR-AC-2, internal, 2026-04).

Finally, the political geography. Any Atlantic siting engages the rights and interests of the Mi’kmaq and Wolastoqey nations, whose treaty relationships in the Maritimes are governed by the Peace and Friendship Treaties — which did not cede land — and whose participation in energy development has matured from consultation to co-ownership, as the First Nations equity positions in the current New Brunswick wind procurements demonstrate (Crown-Indigenous Relations; NB Power; KMKNO). The sweep’s working conclusion is that the region’s recent co-ownership models are an asset to the thesis, not an obstacle: a campus whose dedicated wind portfolio carries Indigenous equity from inception aligns the project with the direction provincial procurement is already moving, and consultation status appears in the sealed register as a scored axis with co-development readiness explicitly favoured (BGL Research Sweep AC-S7, 2026-02; BGL Run Ledger AC-0640, internal, sealed). The lab does not publish further detail here, deliberately: partner-specific engagement belongs to the restricted tier until the partners themselves choose otherwise (BGL Memorandum of Record MR-AC-2, internal, 2026-04).

The Siting Register

Everything above is evidence; the register is the decision. Between January and May 2026 the programme assembled and scored a register of 190 candidate sites across the four Atlantic provinces — every parcel-scale location that survived a coarse screen for transmission adjacency, fibre plausibility, and buildable land — against 23 axes, in a sealed deterministic run designated AC-0640 (BGL Run Ledger AC-0640, internal, sealed). The register’s contents are restricted. Its method is not, and publishing the method is the point of this section: the lab believes siting decisions of this consequence should be made by a process the eventual counterparties — provinces, utilities, First Nations, lenders — can re-run rather than merely read.

The 23 axes fall into seven families, disclosed here at the family level. Power: substation distance, transmission voltage class, PPA-eligible wind resource within contracting radius, firming tariff exposure. Network: great-circle and route-modeled latency to Dublin and London, landing-station distance, conduit inference confidence, carrier diversity (BGL Cartographic Layer CL-9, internal, 2026-02; CL-11, internal, 2026-03). Climate: economizer hours, wet-bulb duration, design-maximum temperature (BGL Run Ledger AC-0575, internal, sealed). Hazard: storm wind field and surge exposure benchmarked to Fiona, flood mapping, geotechnical class (BGL Engineering Note EN-2026-07, internal, 2026-04; ECCC). People: operations and construction labour radii (BGL Labour Radius Study LR-AC-1, internal, 2026-02). Governance: municipal permitting posture, provincial policy alignment, consultation and co-development readiness. Optionality: tidal adjacency, offshore-wind lease-area adjacency, expansion land. Each axis is computed from named public data plus the programme’s own layers; no axis is a judgment call entered by hand, because hand-entered judgments are where siting studies hide their conclusions (BGL Site Axis Codebook, rev. 4, internal, 2026-03).

The scoring discipline is where the lab’s architecture earns its presence in this document. The register runs on the same deterministic stack described in the lab’s other public research: inputs are fused into a governed structure with source lineage preserved, the run is bit-identical for identical inputs, and every site’s score decomposes exactly into named axis contributions that trace to named sources — the property the optimization literature calls reproducibility and the reliability establishment keeps finding absent from learned systems (Venzke & Chatzivasileiadis, arXiv:1910.01624; NERC, 2024; D.A.E. Determinism Attestation DA-2026-Q1, internal). Two design rules matter most. First, axis suppression: where the contradiction registers left a figure unresolved — tidal delivery dates, unpublished tariff riders — the affected axis is suppressed for the affected sites and the suppression is logged, rather than a guessed value polluting the ranking; the AC-0640 run logged every such suppression to its derivation record (BGL Derivation Record DR-0640-0117, internal, sealed; BGL Contradiction Register AC-CR-2, internal, 2025-09). Second, fan-out control: axes are audited so that no single data source silently drives multiple axes and multiplies its own weight — a defect the lab’s engine team has documented in other scoring corpora and specifically engineered against here (OrangePeel Engine Note TAL-AC-4, internal, 2026-03).

What the run found can be characterized without being disclosed. The distribution of scores is not smooth: sites cluster into a small elite tier where power, fibre, and labour geographies intersect, a broad middle differentiated mainly by hazard and permitting axes, and a long tail eliminated by one or two disqualifying axes — most often construction-labour radius or storm surge (BGL Run Ledger AC-0640, internal, sealed). The elite tier is smaller than the region’s promoters would hope and larger than its skeptics would predict. Sensitivity analysis — re-running the register under perturbed weights — found the top tier stable under any defensible weighting, which is the register’s most decision-relevant property: the shortlist is not an artifact of the lab’s preferences (BGL Run Ledger AC-0640, internal, sealed).

Why seal it? Three reasons, stated plainly because an unexplained redaction invites the wrong inference. First, land: a published shortlist of parcel-scale sites is a speculation map, and optioning ground before announcing interest is elementary discipline for any siting programme. Second, counterparties: several axes encode the posture of identifiable parties — a utility’s interconnection appetite, a municipality’s permitting record — whose cooperation is better served by private engagement than public scoring. Third, honesty about immaturity: the register ranks candidates on modeled latency and unexecuted PPAs; publishing rankings built on figures the programme itself has gated behind verification would lend them a false solidity (BGL Corridor Memorandum AC-M-09, internal, 2026-02; Windfall Access Note WA-19, internal, 2026-05). The register will be re-run as gates close — measured RTTs, term sheets, provincial engagement — and each re-run produces a new sealed ledger whose derivation chain back to this public evidence base is unbroken. That chain, not any single ranking, is the artifact the lab is actually building (BGL Verification Gate Schedule VG-AC-1, internal, 2026-05).

Objections and Failure Modes

The lab’s research culture requires that a public document state what would make it wrong, and the Corridor risk register maintains the list this section summarizes (BGL Risk Register RR-AC-1, internal, 2026-05). The objections below are the strongest the programme has been able to construct against its own thesis. None is rhetorical; each carries the evidence that keeps it open and the gate that would close it.

Power availability is the binding constraint, and the region’s is small. The thesis’s answer — dedicated generation with grid firming — is a structure, not a signed contract. If Maritime wind procurement continues to be absorbed entirely by provincial decarbonization obligations, a campus PPA competes with the coal phase-out for the same turbines, the same crews, and the same interconnection studies, and the province’s regulator may reasonably prefer ratepayers to tenants (NSUARB; Government of Nova Scotia; ECCC). The firming fraction is the acute version: every hour the campus leans on the Nova Scotia grid it pays 13¢-class prices for 660 g-class carbon, and enough such hours dissolve both the cost case and the ESG case simultaneously (Hydro-Québec, 2025; CER). The closing gate is explicit: a term sheet whose firming terms survive scrutiny, or no thesis (BGL Verification Gate Schedule VG-AC-1, internal, 2026-05).

The latency claim is modeled, not measured. Route-modeled RTTs systematically flatter real networks: actual paths detour, cross-connects add hops, and landing-station interconnection is not the same as presence in the landing station. The programme’s 40–46 ms Halifax–Dublin figure could plausibly degrade by several milliseconds in practice, and the competitive margin against Montréal — whose cheap Hydro-Québec power funds a lot of tolerance for 15 extra milliseconds — is the thesis’s to lose (TeleGeography; Hydro-Québec, 2025; BGL Run Ledger AC-0601, internal, sealed). Vendor-measured RTTs are gate two, and the lab has committed to publishing the comparison between its model and the measurements when they exist — including if the model loses (BGL Corridor Memorandum AC-M-09, internal, 2026-02).

The sovereignty premium may compress or move. Europe is building its own gigafactories with €20 billion of InvestAI money; if that capacity lands on time, the exportable European workload narrows to redundancy, neutrality, and price arbitrage (European Commission; EuroHPC JU). Canada’s adequacy status — the premium’s legal leg — is periodically reviewed (European Commission, 2024). And the premium attaches only to Canadian ownership, a constraint that binds precisely at the moment of maximum temptation: when a US hyperscaler offers to buy the operating company (BLG, 2026; BGL Threat Model TM-AC-1, internal, 2026-02). The programme treats ownership durability as a structural design requirement of any operating entity, and treats a sale-to-hyperscaler exit as thesis failure, not thesis success.

Construction reality in a small labour market. The capex ledger already shows cold-climate small markets building at Oslo prices, not Texas prices (CBRE, 2025). The construction labour radius is the register’s harshest axis, and the honest scenario for a first campus includes imported crews, camp logistics, and schedule risk priced accordingly — the Muskrat inquiry’s testimony on remote-site productivity is regional required reading (BGL Labour Radius Study LR-AC-1, internal, 2026-02; Commission of Inquiry, 2020). Transformer and switchgear lead times bind everyone equally, but a small buyer has less procurement leverage than a hyperscaler with a standing order book (Data Center Knowledge; LBNL, 2024).

The region’s own record. The most uncomfortable objection is inductive: Atlantic Canada’s energy-adjacent megaproject record — tidal’s fifteen barren years, Muskrat’s $13 billion, the Atlantic Loop’s collapse — is a base rate, and base rates do not care about thesis quality (Commission of Inquiry, 2020; CBC; BGL Research Sweep AC-S2, 2025-08). The programme’s response is not confidence but structure: every prior failure in the record proceeded past unresolved contradictions on someone’s authority, and the entire apparatus of this programme — contradiction registers, axis suppression, named verification gates, sealed but re-runnable rankings — exists so that this project cannot proceed past its own (BGL Contradiction Register AC-CR-2, internal, 2025-09; BGL Verification Gate Schedule VG-AC-1, internal, 2026-05). If the gates do not close, the capital does not move. That sentence is the risk management.

What would falsify the thesis outright. For the record, four findings would end the programme rather than delay it: a measured Halifax–Dublin RTT above roughly 52 ms, erasing the latency margin; demonstrated unavailability of additional Maritime wind at campus scale on any bankable terms this decade; loss or material conditioning of Canada’s adequacy status; or a firming regime whose realized carbon intensity puts the campus above the warm-climate alternatives it claims to beat (BGL Risk Register RR-AC-1, internal, 2026-05). The lab publishes these conditions so that outside readers can hold it to them.

Status of the Corridor Programme

This document is the first Corridor release and the programme’s public foundation: initial research, released 2026-06-19. Five subsequent Corridor documents remain restricted. Their subjects can be named even where their contents cannot: the siting register and its derivation chain; the power-structure work — PPA, firming, and storage architecture; the network verification file as measured figures replace modeled ones; the counterparty and governance record; and the capital-structure analysis for a Canadian-owned operating entity (Windfall Access Note WA-19, internal, 2026-05). Restricted documents are distributed through the lab’s Windfall portal to verified partners; access requests and requests for the underlying research corpus — seven commissioned sweeps with full source lists — go to research@blackgridlabs.com.

The programme’s working conclusion is unchanged from the sealed brief, and it is repeated here verbatim so the public and restricted records cannot drift: the campus case stands on four legs — free cooling, co-developed wind power, measured European latency, and Canadian-owned sovereign jurisdiction. It does not stand on tidal energy, on Nova Scotia grid supply, or on any cost advantage from cold alone. Each leg has a named verification gate before capital moves (BGL Verification Gate Schedule VG-AC-1, internal, 2026-05). The gates in force at release: wind-PPA term-sheet viability in Nova Scotia or New Brunswick; vendor-measured Halifax–Dublin round-trip time; provincial engagement on dedicated generation; and anchor-tenant soundings (BGL Corridor Memorandum AC-M-11, internal, 2026-06).

The lab’s posture toward the region deserves a closing sentence without numbers in it. Atlantic Canada has spent half a century being told its energy story by proponents with something to sell, and the region’s institutional memory — the abandoned turbine on the floor of the Minas Passage, the inquiry volumes on Muskrat Falls — has earned its skepticism (Commission of Inquiry, 2020; CBC). This programme’s wager is methodological: that a thesis assembled in public from checkable sources, scored by a process its counterparties can re-run, and bound to falsification conditions it publishes against itself, is the kind of proposal the region’s history has made it right to demand. Cold power. Cold compute. The export product is not electricity — and the proof, like everything else this lab produces, carries its derivation with it.