DMG Blockchain: The AI Premium Hangs on a Letter of Intent

DMG Blockchain is a small Canadian bitcoin miner whose market value has stopped looking like one. The stock trades at a market cap near US$80 million, yet the underlying business just posted a quarterly revenue figure of C$6.4 million, down 45% from a year earlier, and a net loss of C$3.9 million. The difference between that shrinking mining operation and the price investors pay for the shares is not explained by the coin business. It is an AI premium — and it is bigger, and more fragile, than it looks.

The mining business is getting smaller

DMG runs a 65-megawatt site at Christina Lake, British Columbia that it owns down to its own substation. In the June 2026 quarter it mined 61.9 bitcoin at a fleet efficiency of 21.9 joules per terahash, with 1.47 exahash of capacity. The trend is downward: revenue fell 13% from the March quarter, bitcoin mined fell 10%, and the hashrate fell 14%, as low coin prices and retiring inefficient machines shrink what the facility produces.

The balance sheet tells the same story with a twist. DMG holds C$31.6 million in digital currency and about C$2.8 million in cash, against current loans payable that have grown to C$19.7 million from C$10.9 million a year earlier. In other words, the one hard-asset cushion that used to back a miner, its bitcoin, is now the same pile the company would have to spend, pledge, or dilute against to build anything. There is no meaningful P&L profit here to fund a transition.

The pivot, and the one contract that would matter

Late in 2025 DMG abandoned the pure-mining path, withdrawing its 3 EH/s hashrate target to convert Christina Lake into a "50-megawatt critical IT load liquid-cooled AI data center". In June 2026 it reached the first tangible step: a letter of intent to provide 50 megawatts of AI colocation to a single, unnamed tenant at the site, with an initial 12-year term and three five-year renewals, the first phase targeted for delivery by the end of 2026.

Read that description carefully, because its structure is doing a lot of work. The LOI is non-binding. The tenant is unidentified, under a nondisclosure agreement, and is expected to provide an "investment-grade backstop" for review — a credit test, not a signed check. DMG says there is no guarantee a definitive agreement will be reached. And the company intends to finance the conversion primarily with debt. Every load-bearing step between "we want to build this" and "we are collecting colocation revenue" is still ahead, and each one is conditional.

The colocation math, if it closes, would be transformative for a company this size: 50 megawatts of hosted critical load is an order of magnitude larger than everything DMG currently earns. But the site has only 75 megawatts of total power today, 15 of it firm and 60 non-firm (curtailable), and roughly 25 megawatts of that would be consumed by cooling and non-compute infrastructure before the tenant's load begins. The capacity exists on paper; the money to convert it, and the tenant's commitment to pay for that power, do not yet.

What DMG actually owns that is scarce

Look one layer down, and there is a defensible reason a smaller player might matter. The binding constraints on AI data centers in 2026 are not graphics cards — they are power interconnection, and for government work, security accreditation. That second one is a genuine chokepoint: a Sensitive Compartmented Information Facility, or SCIF, is a room certified by the U.S. government against electronic eavesdropping, and it cannot be rushed. Governments across the West are pressing for "sovereign AI" capacity they can trust not to run foreign code.

DMG has positioned itself on exactly that scarce, slow-to-build node. In April 2026 it received two megawatts of SCIF-rated prefabricated data center units, and in June it signed its first small colocation contract for them — about C$670,000 over slightly more than two years — with an option to add SCIF services. CEO Sheldon Bennett has said he expects customers to value SCIF-rated capacity for secure sovereign and government compute.

That is the most interesting part of the story, and also the clearest measure of how early it is. C$670,000 over two years is not a business; it is a proof that the units can host paying AI workload. The credential they pitch, SCIF accreditation, is genuine and slow to replicate. But DMG has demonstrated 2 megawatts of it, is pitching it underneath a 50-megawatt LOI whose tenant is unnamed, and has not shown that the sovereign-compute buyers it hopes for will choose a British Columbia miner over hyperscale or government-run alternatives.

The premium, and what would validate it

This is where structure and price part company. Put the numbers together on a consistent basis: against a market value near US$80 million, DMG holds roughly C$31.6 million in digital currency and a shrinking, cash-burning mining operation. Most of the market capitalization is already an AI option — the market is paying for the 50-megawatt conversion before the definitive agreement, the financing, or the first megawatt of tenant load exists.

The thesis is not therefore wrong. It is unproven at its three load-bearing links: a non-binding letter of intent must become a signed contract, a build DMG cannot fund from C$2.8 million of cash must be financed without drowning shareholders in dilution, and the SCIF/sovereign niche must actually pay colocation rates that cover an asset-heavy, debt-funded conversion. Any one of those failing collapses most of the premium, because the underlying miner is worth only what its falling coin output and bitcoins carry.

Watch the specific conversions, not the headlines. A definitive agreement naming a tenant with real credit, committed construction financing, and a board-approved budget would turn the AI premium from an option into an asset — and would change the stock's value faster than any single quarter of mining results. Until one of those links closes, DMG's market price is a bet on a letter of intent. That is a bigger premium than the current business justifies, and a smaller one than the 50-megawatt dream would be worth.