The "Bitcoin Miners" That Now Rent to AI: What Today's Dip Really Says

By one measure, today was a quiet, boring day in a noisy part of the market. TeraWulf fell roughly 4%, and Applied Digital and IREN pulled back a few percent, all on the day's coverage being described as profit taking after a strong year. TeraWulf had been up about 44% year to date before this hiccup. There were no new company catalysts, and the slip came at a time when the group had just recovered from a much worse stretch.

That label — profit taking — is accurate as far as it goes. It also hides the part a shareholder actually needs to understand. These are not really Bitcoin miners anymore. They have turned into AI landlords, renting out power-advantaged data centers to companies that need computing capacity they can't build fast enough. And nearly every wild swing in their shares now traces to something other than the companies themselves: the mood of the lending market and the tech sector around them.

The economics really did flip

TeraWulf was founded to mine Bitcoin with cheap electricity. Its most recent quarter showed how thoroughly that business has changed. The company did $44.8 million in revenue, and 71% of it — $31.9 million — came from leasing high-performance computing capacity to AI tenants, not from mining. Its prize contract is a 20-year lease to Anthropic for roughly 401 megawatts at a Kentucky site, worth about $19 billion over the initial term — more than TeraWulf's entire market cap when it was signed — and up to roughly $33 billion if Anthropic exercises both five-year extensions.

The catch is in the timing. That revenue mostly doesn't begin arriving until the second half of 2027 and into early 2028. TeraWulf had about 102 megawatts of revenue-generating capacity online at its flagship Lake Mariner site and 336 megawatts under construction. The same pattern holds across the group: IREN raised its year-end AI-cloud run-rate target to more than $4 billion, roughly 85% of it already under contract, on top of a roughly $9.7 billion five-year AI cloud deal with Microsoft. Applied Digital booked $258.7 million in revenue last quarter, up 407% from a year earlier, and holds about 1.4 gigawatts of contracted capacity worth roughly $36 billion in base-term leases.

In other words, all three now sell a promised stream of future rent, most of which lands years from now, and get valued as if that promised stream will arrive on schedule and in full.

Why they whiplash

Here is the mechanics of the volatility. A company priced on revenue that mostly starts in 2027-28, financed with large amounts of debt, does not move on its own earnings — it moves on the cost of capital and on investor sentiment about AI data-center spending. Those are forces the company does not control.

The clearest proof is what happened in July. All three names fell 33% to 38% in a single month, a decline blamed on "capex fatigue" at the hyperscalers, rising credit costs, and a semiconductor rout. The diversified data-center ETF fell only 13% over the same stretch. The underlying contracts did not deteriorate in any of those weeks; the market's mood did. The names carry betas of roughly 4x to 5.7x, so a wobble in the AI theme gets amplified several times over by the time it reaches these stocks. Then the group recovered, TeraWulf into that 44% year-to-date gain, and now the market is taking a few low-key points of profit off the table.

So today's dip is price action, not a change in the business. The question is whether that distinction should make you a buyer.

Who actually backs the rent

For a group priced on promises, the single fact that separates a durable revenue stream from a hope is who signed the lease and whether they can pay. This is where the three diverge from the generic "AI hype" label.

Google doesn't merely rent from TeraWulf through its partner Fluidstack — it backstops $1.8 billion of Fluidstack's lease obligations and took roughly an 8% equity stake in TeraWulf for its trouble. A company does not write checks like that to secure space it thinks might get canceled. IREN's tenant list runs through Microsoft and NVIDIA. Applied Digital's big new leases are take-or-pay deals of up to 15 years with what it describes as a "high investment-grade hyperscaler" — meaning the tenant must pay for the capacity whether it uses it or not.

That is the moat in this corner of the market. It is not the machines; it is the land, the power access, and the creditworthiness of whoever promised to pay the rent. Contrarian calls in this sector live or die on that, because a cracked tenant — not a falling bitcoin price — is what turns a recovery story into a falling knife.

The honest read

Here I will be blunt rather than obliging. I do not see today's few-percent pullback as a setup to buy, because it is not a bottoming signal — it is a mild profit-taking blip after a recovery, and the group already got its real washout back in July. The moment to test the contrarian case was when the contracts were intact and the stocks were down 35%; the returns today are the market's mood, and chasing them gets you the worst of the betas.

The legitimate risk has shifted somewhere harder to see. These stocks are not cheap on what they earn today — all three are loss-making on a GAAP basis, TeraWulf's most recent quarterly loss of roughly $940 million driven almost entirely by a non-cash warrant revaluation, and IREN trades near 24x trailing sales while Applied Digital is near 12x. The market is paying full price for on-time execution of years of construction — Anthropic capacity in 2027-28, more megawatts at Lake Mariner, Applied Digital's 1.4 gigawatts — financed with billions in debt precisely as financing got more expensive.

That is the trade you are arguing against if you chase the bounce, and it resolves on a 2027-28 horizon, not in trading days. The facts that would change the picture are concrete: a tenant breaking a lease, a refinancing that stalls at rates the math cannot absorb, or hyperscalers cutting their capital-spending guidance. Until one of those appears, a 4% move up or down is telling you less about these companies than about the people trading them. The better risk/reward is on a deeper, exhaustion-driven pullback — not on catching a quiet profit-taking day.