TeraWulf vs IREN: does the sector-wide mining-to-AI pivot re-rate on contracts

TeraWulf and IREN each announced a landmark AI-cloud contract this summer, and the market treated the two as the same event. TeraWulf dated Anthropic on a 20-year lease worth about $19 billion; two weeks later IREN said its 2026 AI-cloud capacity was now contracted at $4 billion of annualized run-rate revenue. Both stocks trade as AI-infrastructure names now, not bitcoin miners, after a 2026 sell-side wave — Morgan Stanley initiated the sector in February calling miners critical energy infrastructure for the AI buildout. But the two deals are not alike, and the difference is the one number that decides whether this re-rating holds: the gap between contracted revenue and delivered, customer-accepted capacity.

Contracting is not delivering. Neither company books a dollar until capacity is actually turned over — IREN's own language is that revenue ramps "upon commissioning, testing, and customer acceptance." So a contracted number is a ceiling on future revenue, not today's revenue, and the distance between the two is execution risk with a price tag.

TeraWulf's $19 billion is years away from the phone in your hand

The headline is the most aggressive in the group, and decomposition is brutal on it. The Anthropic lease covers about 401 MW of critical IT capacity at the Justified Data Campus in Hawesville, Kentucky. As of TeraWulf's second-quarter report, zero megawatts of those 401 are online.

That single flat-zero is easy to miss because the company also reports 102 MW of revenue-generating AI capacity. Those are different buildings. The 102 MW sits at the Lake Mariner campus in New York; the $19 billion contract sits at Justified, where the campus is still in development and initial delivery is scheduled for the second half of 2027, with full 401 MW not due until early 2028. The implied straight-line of the deal — roughly $950 million a year on paper before extensions that could push it past $30 billion — is not revenue the company will recognize in the next three quarters; the first of it likely lands well over a year from now, against capex still being spent.

So the TeraWulf re-rating is a bet on cash flows 12 to 18 months out that today rest on a signed lease, an investment-grade tenant, and a build that has not turned a contracted watt to revenue. Its genuinely operating AI business — the 102 MW at Lake Mariner feeding $31.9 million of quarterly HPC lease revenue — is real but small relative to the valuation the contract headlines pulled forward.

IREN's gap is larger but right now

IREN's numbers are starker on their face: $4 billion of contracted ARR for 2026 capacity against only $1 billion operating as of late August. That is a $3 billion gap it says it plans to close by December 31 — essentially a quarter's worth of commissioning and acceptance.

The difference is that this gap is current. Delivery is already happening, not scheduled. The first of four 50 MW liquid-cooled Horizon deployments at Childress had been handed to Microsoft by August; Horizon 2 was in commissioning; Horizons 3-4 were in late-stage construction targeting fourth-quarter delivery. IREN's self-built AI Cloud capacity is meant to go from roughly 3 MW to 480 MW across 2026.

The financing structure also de-risks the gap in a way TeraWulf's does not. Customers prepay 45-55% of estimated GPU capex before service starts, and dedicated GPU financings cover roughly 90% of the associated capex — a $3.6 billion facility for the investment-grade Microsoft contract at a 6% weighted average interest rate, plus a $2.8 billion facility for non-investment-grade customers at 9%. Much of the build is funded by the tenant and debt markets rather than a pure equity call, which narrows the downside if the build slips. It does not eliminate it: IREN still swung to a $702.6 million net loss for fiscal 2026, mostly non-cash impairment of retired mining hardware, and it carries no bitcoin treasury, leaving dilution as the funding backstop if delivery costs outrun the committed capital.

The re-rating already happened — the question is whether it prices delivery

None of this is hypothetical about whether the sector re-rates; it already has. Morgan Stanley's February initiation pushed TeraWulf up 13% in a day; Applied Digital has run hundreds of percent on the same pivot; and IREN now falls and rises with the AI-neocloud complex, not with bitcoin — it dropped about 6.5% on its earnings day largely tracking CoreWeave down double digits in a broad AI-infrastructure selloff.

The only framework that cleanly separates the two trades is VanEck's June model for valuing miners as AI infrastructure. It prices operators on multiple of gross energized power, but it explicitly warns that the market premium is migrating from "companies with signed contracts" to "those demonstrating the ability to build on time and on budget," that the group has delivered only about a quarter of its leased capacity, and that missing construction milestones invites "structural de-ratings." Its operative metric is the delivery-to-leased ratio.

That is the answer to the question the two stocks together pose. The premium accrues not to the biggest contracted number but to the operator who closes the gap into recognized, customer-accepted operating revenue. IREN's test is immediate and binary — does it convert the $3 billion by New Year's Eve, and does it do so at the contracted price per megawatt. TeraWulf's test is a year and a half of funded construction against a campus with zero revenue yet. The sector re-rates on headline contracts only because the financial machinery that underlies those contracts — prepayments, investment-grade tenants, debt that now funds 90% of capex — lends them credibility. If a complex that lives on that credibility holds its multiple while delivery slips — if IREN's fourth quarter closes short or Justified slides again — then the re-rating was priced with no delivery discount, and the contracts were never the point after all.

The single line to watch is not the ARR target. It is the gap itself: contracted ARR versus delivered ARR, quarter over quarter, and whether the market marks the difference.