AVAX One’s next chapter is being built at the intersection of energy, AI infrastructure, and digital assets.
The company, which started in Vancouver and later expanded into Bitcoin mining, is now positioning itself as a power-first digital infrastructure developer focused on AI and high-performance computing. Its first AI/HPC project is underway in Alberta: a 10-megawatt, Tier 3-ready powered-land site developed with Calgary-based BlueFlare Energy Solutions, with end-client deployment readiness targeted for Q1 2027.
For CEO Jolie Kahn, the opportunity is not simply about building another data centre. It is about solving what she sees as the real constraint in AI infrastructure: access to dedicated, reliable, cost-efficient power.
Through a behind-the-meter energy model, AVAX One aims to remove grid dependency from the critical path, using on-site natural gas generation, battery storage, grid backup, and diesel as a final redundancy layer. Kahn argues that this approach gives the company an advantage as AI and HPC workloads move beyond hyperscale campuses and into a distributed “missing middle” of enterprise inference, regulated industries, edge AI, and sovereign compute.
Alberta is central to that strategy. AVAX One already operates Bitcoin mining infrastructure in the province and sees the market as a rare combination of available energy, industrial sites, cool climate, supportive policy, and proximity to North American customers.
Kahn also brings a unique perspective to the AI infrastructure race. Before leading AVAX One, she served as general counsel at MARA Holdings, giving her firsthand experience with the power procurement, permitting, capital structure, and regulatory challenges that come with large-scale digital asset infrastructure.
In this interview with DataCentre.ca, Kahn discusses why Alberta is becoming a strategic market for AI and HPC infrastructure, how behind-the-meter power changes the economics of data centre development, where hyperscalers are leaving gaps, and why Canada may be more important to the global AI buildout than many investors realize.
AVAX One started in Vancouver and is now building AI and HPC data centre infrastructure in Alberta. What made Alberta the right place to scale this next phase of the company?
JK: What made Alberta right comes down to two things: where we already operate, and what the province offers a power-first developer. By way of context, AVAX One’s legacy business started in Vancouver. The company has followed a deliberate progression through its history: the company began as an agricultural technology business, expanded into Bitcoin mining, launched strategic digital asset holdings, and has now moved decisively into modular AI and high-performance computing data center development.
We view the move into AI and HPC as a deliberate evolution into a power-first digital infrastructure company, with each phase an expansion of the same mandate: regulated, institutional-grade exposure to the next generation of digital finance and physical compute infrastructure.
Alberta is where that foundation already exists for us. We have operated Bitcoin mining in the province for some time, so we are not arriving cold. We are extending capabilities and relationships we have already built. Beyond our own footprint, Alberta offers a rare combination for AI infrastructure: abundant low-cost natural gas, power-rich industrial sites with generation already in place, a cool climate that lowers cooling costs, and a policy environment oriented toward operators that bring their own power. The province’s 2025 Utilities Statutes Amendment Act prioritizes projects that supply their own generation, and its data-centre levy framework favors behind-the-meter supply over grid-dependent load.
We already have a first AI/HPC project underway there: a 10-megawatt, Tier 3-ready powered-land site in Alberta, developed with Calgary-based BlueFlare Energy Solutions, on track for end-client deployment readiness by Q1 2027. It is built as a replicable microgrid in roughly 10-megawatt increments, with a broader Western Canada pipeline under evaluation at sizes from 5 to 50-plus megawatts.
More broadly, that is a model we intend to replicate. We are interested in energy-advantaged territories across North America that offer the same combination we found here: available power, supportive policy, and room to build away from constrained urban grids.
You’ve described AVAX One’s approach as “behind the meter” through a contracted energy partnership with BlueFlare. Why is removing grid dependency so important for AI and HPC infrastructure?
JK: For AI and high-performance computing, the binding constraint is time-to-power, and control over that power, rather than the chips themselves. Utility approvals for large grid-connected AI and HPC loads typically run 24 to 60 months, and the interconnection queues behind them now stretch for years. That is longer than the workloads our customers are deploying can wait.
Behind-the-meter generation removes the interconnection queue from the critical path entirely. Through our partnership with BlueFlare, power is dedicated, on-site, and contractually controlled at approximately four cents per kilowatt-hour. Because we own the power purchase agreement and the on-site generation rather than leasing capacity, we control both cost and availability. We build to a true Tier 3 architecture: natural gas as primary supply, the grid as secondary, battery storage for ride-through, and diesel as a final backup. AI workloads do not tolerate downtime or unpredictable power prices, and owning the power is how we take both risks off the table.
There is also a policy dimension that increasingly favors this model in Alberta. The province’s data centre levy falls on grid-connected load and eases as a project supplies its own power, and the system operator has moved to limit new large-load grid connections. Large grid-connected data centres are increasingly expected to pay their own way rather than shift costs onto households. A behind-the-meter operator sits outside that debate by construction. We draw no residential grid power, bid into no capacity market, and socialize no costs onto the communities around us. What looks at first like a reliability decision is, increasingly, a regulatory and community-license advantage as well.
A lot of the AI infrastructure conversation focuses on hyperscalers. Where do you see the gap that companies like AVAX One are positioned to fill?
JK: We describe it as the missing middle. The public conversation centers on gigawatt-scale campuses built for frontier model training. That is the right infrastructure for training. It is the wrong infrastructure for the workloads now scaling fastest: enterprise inference, edge AI, and regulated industries.
Those workloads generally live in the 1-to-50-megawatt range, and they require Tier 3 reliability, dedicated power, and regional placement. The reasons are structural. Latency budgets in applications such as surgical robotics and real-time fraud detection are measured in single-digit milliseconds, while a round trip to a distant hyperscale campus can add 60 to 100 milliseconds before any computation happens. Data-residency requirements in healthcare, finance, and government put workloads behind regulatory walls that centralized hyperscale geography cannot cross. Hyperscale economics do not bend to that footprint, and conventional colocation cannot deliver dedicated capacity at the right price or timeline.
Cloud computing followed the same path. The hyperscalers defined the category first, and then a distributed, regional layer emerged to serve everything the centralized model was not built to handle. AI infrastructure is entering that same chapter, and the distributed layer is exactly where we operate. We do not need to dominate that market to succeed in it. The segment is fragmented by its nature, the unit economics are attractive even at a modest share, and it rewards regional operators that have secured the right power and built the right operating stack.
How does AVAX One’s existing bitcoin mining infrastructure in Alberta give the company an advantage as it expands into AI and HPC workloads?
JK: It gives us a multi-year head start on an aspect of AI infrastructure that is challenging. The hard part has always been securing power, navigating permitting, energizing a site, and running it reliably around the clock. That is precisely what we have done in Bitcoin mining in Alberta and Ohio, where we operate roughly 300 petahash per second of capacity today. The execution, the permitting, the energy procurement, and the uptime discipline carry directly into AI and HPC. We are extending what we have already built in the same energy-advantaged markets.
Mining also gives us operational optionality that a pure data-centre developer does not have. Because we own the power, every secured megawatt can earn from the moment it is energized: capacity not yet contracted to an AI or HPC tenant runs Bitcoin mining in the meantime, and steps down as customer demand comes online.
Mining is a flexible, fast-cycling load that lets us monetize power while we build the customer side, rather than carrying idle megawatts. Few operators in this segment have both the digital-asset operating experience and the power-asset control to manage capacity that way.
There is a growing debate around whether bitcoin mining sites can realistically transition into AI data centres. What are the biggest similarities and differences between those infrastructure models?
JK: There are common elements like site selection, power procurement, cooling, and running compute around the clock. But AI and HPC require far higher rack density, true Tier 3 reliability (99.982% uptime, with N+1 redundancy and concurrent maintainability), low-latency networking, and uptime guarantees that mining simply does not. Retrofitting an existing mining site into an AI facility is bounded by what that site’s original power, cooling, and cabling allow, and many mining sites cannot and should not make that conversion.
Our approach is different, and it sidesteps the conversion debate. We are not transitioning mining sites into AI sites. We build AI and HPC-first facilities, purpose-designed greenfield and optimized for current-generation density and Tier 3 reliability, and we work Bitcoin mining in as a complementary element rather than the starting point. AI and HPC is the primary use case, and the customer we design around.
There are strategic advantages to combining these disciplines. Mining is the flexible load that monetizes the same secured power until and between customer contracts. What actually carries over from mining is secured, cost-efficient power and the operating experience to run it.
From your perspective, where is AI infrastructure capital actually going right now, and what are investors getting wrong about the market?
JK: Much of the capital is flowing into two places: chips and gigawatt-scale campuses. The largest cloud and AI companies are committing on the order of $600 billion in capital expenditure this year by some reports, the majority of it to AI infrastructure, and a growing share of that is going into chips and memory.
GPUs are in high demand and supply is short, creating a constraint the market is feeling. But there is another bottleneck where we see an opportunity to create value. You can buy all the silicon you want, but you cannot energize it if you are on a many years-long interconnection queue waitlist. Across North America, gigawatts of planned capacity are already being delayed or cancelled for lack of power. Capital is underpricing energy and time-to-power needs in favor of compute that cannot yet be plugged in.
The second blind spot follows from the first. Because large anchor leases are the most efficient capital to chase, operators are pulled upmarket toward hyperscale, which leaves the 1-to-50-megawatt segment fragmented and underserved. For a focused operator, that fragmentation is an advantage. What ultimately matters is cost-efficient power, delivered reliably. Once we solve for that, the same advantage holds at sizes well below hyperscale, and that is the segment we are built for.
Canada has abundant energy, land, cooling advantages, and proximity to U.S. markets. Why do you think the country is still underappreciated in the global AI infrastructure race?
JK: It does seem to us that the scale of the opportunity in Canada is not appreciated enough right now. Anyone paying close attention to the AI buildout already understands Canada’s role in it.
Abundant, low-cost natural gas, power-rich industrial sites with generation already in place, a cool climate that lowers cooling costs, stable rule of law.
Much of this capacity sits well away from constrained urban grids, which solves a variety of emerging issues on the policy front. Alberta is targeting a multi-decade data centre buildout, with Invest Alberta estimating a C$75 to C$100 billion economic opportunity.
There is also a demand-side case that gets overlooked. Regulated industries in Canada, including banks, healthcare systems, and federal and provincial governments, face data-residency requirements that effectively exclude US hyperscale infrastructure. That is sovereign workload demand that has to be served inside the country, on operator-independent infrastructure, and it is exactly the kind of compute we are building. Canada is well positioned for AI infrastructure and also has its own regulated and sovereign workloads that domestic infrastructure is best placed to serve.
As the former general counsel at MARA Holdings, what lessons from large-scale digital asset infrastructure are most relevant to building AI infrastructure today?
JK: At scale, power procurement and capital structure matter as much as the hardware. Energy contracts and the regulatory groundwork are important areas of expertise. My Bitcoin mining experience has taught us how to secure power, hold a power purchase agreement, and manage energy use as a flexible asset, which are some of the same disciplines AI and HPC infrastructure depends on.
Another lesson is that two infrastructure waves are converging, and most of the market is watching only one of them. The first is AI compute. The second is the onchain economy: the migration of payments, settlement, and asset ownership onto blockchain rails, where transactions clear faster and at a fraction of the cost of the system they are replacing.
That second wave is easy to underestimate because its link to AI is not obvious. But as AI agents begin to transact on their own, they will need programmable, machine-native rails for payments, identity, and asset transfer, and onchain infrastructure is the layer of the financial system built natively to provide them.
That makes the onchain economy one of the most significant future workloads for the very compute we are building, and it is one that much of the market is still overlooking. Having operated in digital assets natively is precisely the kind of foundation that prepares us to build for what comes next.