Hut 8 enters the fall transformed in its own paperwork while still wearing its old skin in the ledger. The second-quarter print closed with 949 megawatts of IT capacity contracted across two AI campuses. Roughly $26.6 billion of base-term value sits behind those leases, and $1.75 billion of expected average annual net operating income follows once the data halls energize. The signature event came on July 20, when a highly rated tenant doubled its Beacon Point footprint to 704 megawatts at a price per megawatt indistinguishable from the first lease. Commercial velocity of that shape does not merely add backlog. It reprices every undeveloped megawatt in the eight-plus gigawatt pipeline the platform has originated behind it.
The mechanism that matters, beyond headline capacity, is capital-market feedback. In the spring the project-level debt markets funded the River Bend campus with $3.25 billion of secured notes at a 6.192 percent coupon. The same markets then handed $4.25 billion to Beacon Point at a rate two tenths of a point tighter, one rating notch higher, and a faster amortization schedule, because the collateral story had visibly matured between the two underwritings. On September 4, ERCOT granted a conditional Base Load classification for Beacon Point in the Batch Zero process, the stricter of the two large-load designations, which preserves the campus capacity in the grid study without a re-queue. Power origination is the binding constraint of the whole AI build-out. The proof compounding across the year changed the grade of the assets being underwritten. It remains conditional, yet the fact of the stronger grade is the point.
Compute from mining and services still supplies nearly all current revenue, and bitcoin production jumped to 935 coins in the quarter against 308 a year earlier on the Vega startup and the re-energized Drumheller fleet. The headline net loss of $177.1 million is dominated by a $138.6 million mark-to-market write-down on held bitcoin, a noncash artifact of an admitted treasury strategy. Meanwhile unrestricted cash at the parent rests near $234 million against an interest bill that begins in late 2026. A collateral-heavy facility also matures the following April with a margin-call structure attached. The share count has also grown roughly ten percent since the spring record date. The carry costs are landing later and larger.
The catalyst path is dated. Initial energization at Beacon Point and the first River Bend data hall arrive in the first half of 2027. Beacon Point halls follow through mid-2028, which is when contracted revenue begins to convert. Between now and then the market reads a quarterly verification contest: construction progress that keeps the P&L noise on its noncash side, pipeline conversion announcements out of the diligence-stage portfolio, and the final determinations on the Batch Zero file.
The relevant competitive set spans the pure-play bitcoin miners converting legacy power contracts into AI campuses, chiefly the large Texas and Nebraska operators, the smaller specialized data center developers bidding for the same interconnection slots, and the REIT-style giants of the digital infrastructure world whose balance sheets dwarf anything the platform carries. Hut 8 occupies the narrow middle. It originates its own power rather than leasing it, it underwrites greenfield builds rather than retrofits, and it has already attracted investment-grade demand to two of its own campuses before a single contracted hall energizes. That combination, power origination in Texas and Louisiana, status as a merchant miner with a treasury, and status as a first-generation AI landlord, is what separates the equity story from both the miner cohort and the data center REIT cohort.
The platform is Miami-headquartered and dual-listed, with capacity under management split between bitcoin sites in Alberta, New York, and Texas, a handful of high performance computing facilities in Canada, and the two new AI campuses now under construction. Roughly 700 megawatts of the operating fleet serves American Bitcoin, the mining platform in which the parent retains a majority position alongside the Trump family investors who co-founded it. The operating assets still earn their keep, yet they no longer define the strategic direction. The defining assets are the two campuses, because they are where the contracted $26.6 billion value sits.
The strategic reset ultimately traces back to March 2025, when the parent contributed substantially its entire bitcoin mining fleet into a newly relaunched entity majority-owned by Hut 8 and co-founded with the Trump retail magnet. The carve-out accomplished something structural. Mining, the segment with the most volatile cash flows and the highest cost of capital, gained its own public currency and a treasury mandate, while the parent realigned around activity with financeable, long-duration economics: originating power, building campuses, and leasing them to counterparties whose credit the project bond markets recognize. Two lines of business now sit side by side. One is a levered claim on bitcoin prices through a controlled, publicly traded miner. The other is a landlord issuing into a private credit market that has just demonstrated a tightening spread appetite for its collateral.
That structural duality is the real business model, and the second-quarter print is the first filing in which both sides appear at once. The mining tie-up supplied a tripled coin haul and a treasury of more than seventeen thousand coins across the consolidated group. The AI campuses supplied the leases, the project notes, and the ERCOT classification. Each side subsidizes the other's optics, and the tension between them, a volatile coin price against a fixed coupon schedule, is the thread the rest of this report follows.
The offered product is not mine output or even data hall capacity. It is pre-contracted grid capacity, delivered on a schedule. That product produces a hard moat for precisely as long as AI demand growth outpaces transmission build-out. A data hall without an armed feed is a shell, and multi-year interconnection queues, particularly inERCOT and the southeastern grids, mean holders of studied and classified megawatts can justly claim scarcity value that parties starting today cannot simply buy their way into. The interesting operational asset is therefore invisible. It is the reserved step in a queue.
The first commercial evidence of that pricing power came from the Beacon Point campus near Corpus Christi. The tenant, unnamed and of high credit standing, executed a second fifteen-year triple-net lease on substantially the same terms as its first. When a tenant doubles down at an unchanged implied yield, it signals satisfaction on the earlier build, an absence of alternative vendors at comparable scale and schedule, and a steady willingness by the sponsor's underwriting committee to set lease terms as late as the post-signing date. A triple-net structure layers the tenant's footprint atop the campus economics, because taxes, insurance, and maintenance flow outside the landlord gross ceiling and effectively protect the NOI stream.
The second mechanism is design-for-tenancy. Phase 1 of Beacon Point was redesigned during development to align with a reference architecture for hyperscale clusters, packing roughly fifty-seven percent more IT capacity into an unchanged land and utility envelope. The first building's scope was revised upward from around 224 megawatts to 352. That is not merely efficient engineering. Megawatt-hour demand per acre moves precisely, and a sponsor able to redesign mid-build without triggering utility re-energization or new approvals exhibits rare regulatory dexterity. The redesigned rubric now exists as a reusable product, which lowers development capital needs for the rest of the pipeline.
The menu of moats also includes a fintech overlay that gets little attention: project finance. Indeed, the clearest competitive moat in the industry is not ownership of power plants but access to capital markets at contracted-economics terms. In the spring the parent's development subsidiary priced notes against specific campus cash flows, deep leverage that residential issuers rarely see for comparable finance, and achieved a coupon compression between the first and second offerings. The fixed-income market effectively certified the developer's methodology, an asset-light word for future repeated issuance potential. Access to low-cost project finance is a moat because it capitalizes future campuses cheaper than equity or secured lending would allow for less-proven peers.
The reported quarter demands a two-lens read. On the operating side, revenue growth ran above eighty percent year over year as energized machines at Vega and a refreshed Drumheller fleet lifted production to 935 bitcoin against 308 last year. Contract cost of revenue grew far slower than revenue, carrying the segment's gross margin to roughly 64 percent from 47 percent and capturing the operating leverage behind ASIC optimization. The operation produces modest but genuinely positive cash economics. Revenue landed at $74.9 million against $41.3 million a year earlier, with compute supplying the overwhelming share of the total. Adjusted EBITDA, now defined to exclude marks on digital assets, printed $10.4 million against $4.2 million in the prior-year quarter. Both lines confirm the fleet upgrades flowed straight into the P&L rather than into stranded capacity.
The ledger tells a louder story. A net loss of $177.1 million sits opposite a year-ago gain, and $138.6 million of it is the fair-value write-down on the consolidated bitcoin reserve after the coin backed off its early-quarter strength. That loss is noncash and valuation-driven rather than operational. Double the fair-value volatility, in arithmetic terms, without a single dollar leaving the treasury. Loss accounting of this type at a holder with more than 17,000 coins on balance recurs every quarter. Investors have to drive a wedge between two P&Ls: operating income, and the mark-to-market ledger that dominates optics and clouds institutional readability.
The balance sheet has turned into a peculiar arrangement of two distinct liquidity blocks. Roughly $6.8 billion of restricted cash, the note proceeds, is already fenced into construction and debt-service reserves at the project subsidiaries and legally cannot serve corporate uses. Only $234 million is available for corporate use. The platform holds a further $1.04 billion fair value across the bitcoin treasury, of which a large fraction, roughly 4,850 coins plus the miners' pledged allocation, is encumbered against facilities. Total debt stands near $7.6 billion, largely nonrecourse to the parent. Clearance between the restricted pile and the parent's operating burn is the entire liquidity story over the next ten quarters.
The capital-shape evidence in the quarter is the maintenance of a cleaner capital structure. At the parent level, a longstanding convertible note, or rather its accreted balance and all remaining recourse obligations, disappeared this quarter via conversion or refinance. The FalconX facility was refinanced from 9 percent to 7 percent collateralized with bitcoin, and the legacy credit lines were retired. The parent now touches no recourse debt. Everything expensive sits at bankruptcy-remote SPVs. The end effect is equity that behaves like a call option on two projects whose failure modes are isolated from one another.
The forward window runs on a dated schedule, and the dates form the near-term catalyst chain. River Bend's first data hall remains scheduled for delivery in the second quarter of 2027, with Beacon Point's initial energization in the first quarter of 2027 and first data hall in the third quarter of that year. Phase 2 halls at Beacon Point run through the second quarter of 2028, and tribunal interest begins next month. Every bridge quarter in between is a filing cadence of mobilization reports: slab pours at the auxiliary yard and main building, equipment arriving on site, substations. Each can be verified or falsified through photography and disclosed milestones, and each is a quarterly referendum on the 2028 NOI ramp.
Execution risk sits decaying around specific named choke points. ERCOT's Batch Zero designation for Beacon Point is conditional, not final. The conditions attach to demonstrated financial commitment, quarterly stability assessments, and the state audit process, and losing the designation would re-queue the campus capacity behind every other large load applicant in Texas. River Bend sits in the MISO queue on a Louisiana site with in-state political support and utility coordination, an easier path in most respects, yet the campus has a shorter interconnection window and depends on eastern grid assessments that have historically slipped. Supply chain exposure is 2027-relevant too, though the long-lead equipment across both campuses is already ordered and the partnership delivery model the sponsor uses has been de-risked.
The most quietly consequential decision in the file is the parent's equity behavior. The record date for the annual meeting caught 112.6 million shares in April. By quarter end, the count had grown to roughly 123 million. ATM issuance for a full quarter plus conversion dilution combined to mint close to ten million shares in a single spring at prices well below the current ten. That is capital cost at the worst possible moment of the cycle, and it happened at the parent's own ATM rather than at the project subsidiaries. The clean way to fund development equity was stranded capital, and the parent chose the avenue that left it unstranded right at the lows.
Between now and 2028 the company guides with a $1.75 billion average annual NOI figure attached to the whole contracted portfolio. Any worsening of the schedule at either campus answers the conversion question directly. The market is showing the lease-bearing campuses a skeptical discount precisely because the revenue converts nothing until energization. Every quarter between now and the first data hall is a pure construction and financing verification exercise, and the share count behavior in that window constitutes the market's sharpest adjudicator of the parent's discipline.
The concentrated risk is single-tenant exposure at Beacon Point. Two seventeen-year leases of comparable size went to the same unnamed counterparty, which concentrates the campus's utility-scale cash flows on one tenant's creditworthiness. Triple net and investment grade DP the downside in normal markets, yet the exposure is real. Full-throttle AI capex retrenchment from even a single highly rated balance sheet would strand contracts with no spot market fallback. In a bandwidth sense the run-rate risk is lower by virtue of the size, though in the landlord model counterparty concentration is precisely what the grade-of-tacket language under-weights.
Second is the carry risk, and it is dated and hard. Interest payments on both note packages begin in November 2026, and the accumulated interest expense is already visible in the statement, with the quarter showing roughly $51.2 million of gross interest outflow before capitalization and the interest income earned on undeployed noteproceeds. The parent-parent gap is bridged by a bitcoin-backed facility with a collateral ratio that,in a meaningful drawdown, triggers margin call behavior before maturing in April 2027. A deep crypto winter would combine a weaker asset with a resolution mechanism the parent cannot instantly satisfy. Add convert dilution in the unlikely event of a bear equities cycle and the parent's cushion does not offer much彷徨 room for self-funding missteps.
Third is regulatory reversal risk, which the Batch Zero outcome embodies. The classification is conditional and, under the applicable rule, failure to satisfy conditions disqualifies the load from the study batch. The state audit process contributes an additional point of political risk. Losing base load status would not destroy the economics of the campus. It would very likely reintroduce years to a build schedule that is already stretching to 2028, re-cycling market concerns about interconnection risk that the second-quarter debt raises had already begun to discount.
Fourth is the transmission of bitcoin treasury risk into equity risk. The parent holds roughly 9,300 coins directly and consolidates the controlled miner's another 8,000 as a majority owner. Fair-value swings accrue to net income each quarter regardless of cash flow. That introduces three layers of correlation between the equity and bitcoin prices, direct treasury exposure, service fee inputs for the mining affiliate, and pledged collateral. The de-correlation case rests on contracted NOI surpassing the mining business's revenue, which the 2028 schedule delivers. The compound downside scenario, should all these risks fire in the same cycle, is a leaking breach of the parent's liquidity strait. In that configuration, the parent's real options are dilutive or asset-sale based. The probability-weighted view is not that the surface jumps straight to zero, but there is no range of outcomes in the scenario tree where the equity avoids a painful re-rating exercise. The share count and dilution cost history adds a second-order behavioral risk. Investors have now seen the ATM window opened at the worst possible time.
The proper valuation frame is a sum-of-the-parts model anchored to project-finance comparables, because the parent now carries three asset classes that trade at radically different multiples: contracted AI leases, a levered mining platform, and a bitcoin treasury. Applying the contracted-cash-flow model first, the contracted IT capacity carries $26.6 billion of base-term contract value, plus $1.75 billion of expected average annual net operating income at stabilization. Triple-net data center REITs with comparable tenant credit trade at capitalization rates roughly in the five and a half to six and a half percent band. Applying a cautious construction-complete discount for delivery risk of roughly two hundred basis points, the stabilizing stream capitalizes between $20 billion at the rich end and $14 billion at the conservative end. Both figures sit before subtracting the $7.6 billion of nonrecourse campus debt. The spread between those two anchors is essentially the market price of construction certainty in this asset class.
The mining and treasury side resolves more cleanly because it is liquid and marked. Consolidating the controlled miner at a market-linked stake in an entity whose treasury sits around eight thousand coins, then adding the parent's own nine thousand coin reserve at fair value, produces roughly $1 billion of hard, markable asset value on the bitcoin side. The service agreements that bundle the mining fleet into the parent campuses produce stable colocation income that justifies a multiple above pure miner trading floors, since hosting contracts are closer to infrastructure cash flows than coin harvesting. The mining stake and treasury together form a real and immediate floor worth roughly a tenth of current market value.
Under this sum-of-the-parts arithmetic, the equity math flexes across three scenarios, and the spread between them mirrors how much delivery uncertainty the market absorbs. In the bear frame, both campuses complete but a late-decade crypto winter compresses the ramp assumptions, the parent consumes its cushion, and dilution runs heavy. Applying that discount and a late renewal haircut, the contracted stream values near $14 billion gross and the coin assets compress toward $700 million. After nonrecourse debt the parent equity lands near $7 billion, roughly forty percent below the current $11.7 billion market value. The bear frame is a re-rating, not a wipe-out. In the base frame, construction holds to schedule and the stabilizing NOI holds near the guided figure. At a six percent capitalization rate the stream carries near $29 billion gross, minus the notes, plus treasury, a parent equity value in the low-to-mid teens of billions, roughly twenty to thirty percent above the current close. In the bull frame, pipeline conversion beyond the two campuses, renewal exercises at Beacon Point, and tightened project-debt spreads push the contracted book toward the $50 billion renewal-heavy figure the company has disclosed. The bull outcome doubles the equity from the current print.
The derived conclusion is that the market currently pays a mid-cycle infrastructure price with half the delivery outcome still in question, and the option skew runs asymmetrically to booked-but-unbuilt value. The scenario math prices delivery slip as the dominant risk, not technology or tenant failure. The equity carries roughly $2 billion of immediately markable bitcoin and a(colocated service annuity that would survive even a total AI-campus failure, so a genuine wipe-out scenario requires simultaneous collapse of the AI lease outcome and the coin price. That asymmetry, narrow but genuine downside floor against an uncapped contracted stream if 2028 goes to plan, is what the valuation argument reduces to.
The quarter established one specific fact that the rest of the year gets measured against: a sophisticated tenant with investment grade credit chose to double its committed footprint at a fully unbuilt campus, on essentially unchanged lease economics. That single data point validates the sponsor's delivery model more decisively than any revenue line in the file, because a tenant of that standing underwrites construction risk with its own covenant, which converts a developer's promise into a counterparty's obligation. Combined with tighter spreads on the second note offering and the conditional Base Load classification at ERCOT, the quarter assembled external, arm's-length validation of the development platform from three separate counterparties: a tenant, a bond desk, and a grid operator.
The strategic machinery now visible is a repeatable origination cycle. Power gets sourced into a queue position, the position gets underwritten by a tenant before the concrete is poured, the campus factors itself in the project bond market against the contracted cash flows, and the parent's construction balance sheet rolls into the next site before the first site produces a dollar of revenue. That flywheel converts the company from a bitcoin miner with a construction habit into something structurally closer to a private credit callable developer, where the recurring revenue is development fees and the assets themselves are rateable at project finance terms. The mining affiliate alongside it acts as both a customer and a call option on the other digital asset cycle, and the two engines now read as deliberately paired rather than accidentally tangled.
The counterargument deserves its own weight here. Everything above assumes that AI compute demand in 2028 resembles the 2026 forward order book, that the unnamed tenant's balance sheet holds, and that ERCOT's conditional classification converts to a final determination without re-queuing. A hyperscaler capex slowdown, a listed-but-unproven macro rotation out of long-duration AI infrastructure, or a Texas political turn against large-load reclassification would each pressure the thesis independently. The equity at an eleven billion market value is already paying for a meaningful chunk of the delivery outcome, and the parent's own ATM behavior during the spring share issuance suggests management itself prices capital defensively when development equity is needed. The contrarian read is that the proper entry is post-energization, after the first data hall produces documented cash flow and the tenant concentration risk resolves one way or the other.
What resolves the thesis over the next twelve months sits in four observable channels. First, the construction verification chain at River Bend, whose first data hall delivery in mid-2027 is the single most falsifiable milestone in the entire story, and whose slip of even one quarter historically re-priced comparable developers by multiples of the underlying delay. Second, the ERCOT final determination pipeline for Beacon Point, where the quarterly stability assessments through 2027 and the state audit process together resolve whether the Base Load classification hardens or degrades. Third, the parent liquidity trajectory, where a bitcoin-collateral facility matures in early 2027 as the first project note payments land. That combination forms a hard, dated test of the capital plan. Fourth, the share count behavior, because the spring dilution pattern established a benchmark the parent either contradicts or confirms at the next development equity need. Each channel produces a documented, public, and dated data point, and the assembled picture across those four channels is what converts the current write-down-heavy income statement into either an infrastructure platform durable across cycles or a levered coin miner with a construction overhang.