IREN, TeraWulf, and Applied Digital Slide as Investors Test the Data Center Trade

IREN, TeraWulf, and Applied Digital reportedly lost roughly 30% within a month, despite announcing contracts and infrastructure milestones that strengthened their growth stories. The divergence, amplified through Google News coverage, signals a sharp change in how investors judge AI data center stocks.

The market is no longer rewarding every megawatt, lease, or customer announcement at face value. Investors now want evidence that contracted capacity will become operating infrastructure, recognized revenue, and durable cash flow without excessive dilution or debt.

That creates the central conflict. Demand for AI computing capacity remains strong, but building the facilities requires years of construction, financing, equipment procurement, and customer coordination. CoreWeave and other specialized infrastructure providers face the same scrutiny, but IREN, TeraWulf, and Applied Digital combine that challenge with operating histories tied partly to cryptocurrency mining.

What Changed Behind the Data Center Stock Selloff

The selloff appears less like a rejection of AI demand and more like a reassessment of execution risk.

The reported one-month declines followed a period when data center stocks had climbed on expanding infrastructure pipelines, major customer commitments, and limited access to utility-scale power. That enthusiasm gave investors little room for delayed construction, expensive financing, weaker revenue recognition, or shifting risk tolerance.

A decline shared by several related stocks also matters. Company-specific disappointments usually produce different outcomes, while a synchronized drop points toward a broader change in how the market values the group.

IREN announced new customer contracts during the decline. TeraWulf disclosed a long-term Anthropic lease. Applied Digital delivered another building and later reported substantial quarterly revenue growth. Positive announcements failed to restore the earlier momentum consistently.

That reaction suggests investors had already priced in strong demand. New contracts therefore had to answer more demanding questions about financing, delivery dates, customer concentration, and cash returns.

The original headline’s roughly 30% figure should be treated as a reported snapshot rather than a permanent measurement. Returns change with the selected start time, intraday price, and publication timestamp. The more durable fact is that all three suffered material drawdowns during a month filled with seemingly favorable operational news.

Google News aggregation can make that tension more visible because readers see upbeat contract announcements beside falling-share headlines. However, aggregation does not establish the cause of a market move. Stock prices also respond to interest rates, positioning, profit-taking, short interest, and changes across the broader technology sector.

The timing is especially striking because none of these companies had abandoned its AI infrastructure strategy. Each was still describing expanding demand and a larger pipeline. The market simply began applying a heavier discount to promises that depend on future construction.

That discount is rational when a company must spend before it can earn. A signed lease can improve revenue visibility, but it does not eliminate permitting, procurement, interconnection, financing, or commissioning risk.

Investors are therefore separating three milestones that promotional narratives often blend together: securing power, constructing a facility, and operating customer equipment. Only the third milestone produces recurring service or rental revenue at the expected scale.

This distinction sets up the real stakes. These companies are not being tested on whether AI requires more computing capacity. They are being tested on whether their capital structures and operating teams can deliver it profitably.

Why IREN, TeraWulf, and Applied Digital Face the Same Pressure

All three companies are trying to convert scarce power access into AI infrastructure revenue, but each must cross a capital-intensive gap first.

IREN offers the most vertically integrated version of the strategy. It plans to provide both facilities and GPU-based cloud capacity, which gives it more potential revenue per megawatt. It also exposes the company to hardware costs, utilization risk, customer service obligations, and rapid GPU obsolescence.

In July, IREN said it had signed new multi-year contracts representing $2.8 billion in total contract value. The company raised its year-end AI Cloud annualized run-rate revenue target above $4 billion and said roughly 85% was contracted.

Annualized run-rate revenue, or ARR, estimates a full year of revenue from capacity operating at a specified point. It is not the same as revenue recognized under generally accepted accounting principles.

IREN explicitly states that its target depends on GPU models, pricing, utilization, commissioning, testing, and customer acceptance. Its contract assumptions therefore contain several operational steps that must occur before the target becomes reported revenue.

The company also said recent customer prepayments represented approximately 45% of related GPU capital expenditure. That reduces IREN’s immediate funding burden, but it does not remove deployment and performance obligations.

TeraWulf is emphasizing long-duration infrastructure leasing. Its July 6 agreement with Anthropic covers a planned campus in Hawesville, Kentucky, with approximately 401 megawatts of critical IT load.

Critical IT load measures the electricity available directly to computing equipment, excluding much of the supporting facility overhead. It provides a more useful indicator of potential computing density than a campus’s total power connection.

TeraWulf said the 20-year lease should generate approximately $19 billion in contracted revenue during its initial term. Initial capacity is expected during the second half of 2027, with the full campus targeted for early 2028.

Those dates explain why a large contract can coexist with a falling stock. The Anthropic lease creates visibility, but significant construction must occur before operations begin.

TeraWulf also agreed to sell its 50.1% interest in the Abernathy joint venture to an investor group led by Fluidstack. The company said the transaction would monetize an investment of approximately $450 million and free capital for wholly owned projects.

Applied Digital follows another model. It develops large campuses and signs long-term leases with hyperscale customers, while retaining a cryptocurrency hosting business that generates current operating income.

Its fiscal fourth-quarter results showed how quickly delivered infrastructure can change the revenue profile. Applied Digital reported quarterly revenue of $258.7 million, up 407% from the comparable period.

However, $152.4 million of its $203 million in HPC Hosting revenue came from tenant fit-out services. These services cover work needed to prepare facilities for customers and do not necessarily carry the recurring economics of base rent.

Base rent contributed $44.1 million during the quarter. This split matters because investors should not apply the same valuation to one-time fit-out activity and long-term rental revenue.

The companies therefore share one pressure source: markets are demanding proof that announced capacity can produce recurring earnings. Their forced response is to provide clearer commissioning schedules, funding plans, and cash-flow bridges during the next several quarters.

The Real Contest Is Contracted Capacity Versus Operating Cash Flow

The primary opponent is not one company against another. It is contracted capacity against operating cash flow.

Data center developers often present their pipelines through megawatts, potential revenue, customer agreements, and targeted completion dates. Those figures help describe future scale, but they do not reveal how much capital remains at risk.

Operating cash flow provides a tougher test. It reflects revenue collected from delivered services after ordinary operating expenses, though investors must still account for capital expenditures and financing costs.

IREN’s new contracts demonstrate demand for its planned GPU capacity. Its customer list includes Microsoft, NVIDIA, Perplexity, Figure AI, Together AI, and several specialized AI developers, according to the company.

The breadth of that list reduces dependence on a single type of workload. Yet IREN’s growth target still requires a rapid expansion from approximately three megawatts of self-built AI Cloud capacity to 480 megawatts planned during 2026.

That is an enormous operational transition. The company must procure systems, finish data halls, connect networking, test clusters, and obtain customer acceptance across multiple deployments.

IREN’s strategic relationship with NVIDIA reinforces its infrastructure credentials. The companies intend to support as much as five gigawatts of NVIDIA-aligned capacity over time, with the Sweetwater campus in Texas expected to serve as a flagship site.

The NVIDIA partnership also includes a conditional right for NVIDIA to purchase up to 30 million IREN shares. That connection is significant, but the five-gigawatt figure remains a long-term ambition rather than operating capacity.

TeraWulf faces a similar timing gap. Its Anthropic agreement establishes a customer and a lease term, but the Kentucky campus will not begin initial service until the second half of 2027 under the current schedule.

The company must fund and build before rent begins at scale. It must also coordinate construction with Anthropic’s technical requirements and maintain access to the required power infrastructure.

Applied Digital has moved further into the operating phase at Polaris Forge 1. Its first 100-megawatt building entered service in October 2025, while another 75 megawatts became live by June 30, 2026.

The company reported approximately 1,410 megawatts of contracted critical IT load across five campuses as of May 31. It estimated approximately $36 billion in contracted revenue over initial lease terms.

Those numbers provide unusual forward visibility. Still, most of the contracted load had not yet entered service, leaving Applied Digital exposed to construction schedules, capital costs, and customer concentration.

Its latest results illustrate the market’s dilemma. The company delivered strong headline growth and adjusted profitability, but it also reported a quarterly net loss attributable to common shareholders of $110.6 million.

For the fiscal year, Applied Digital produced $611.3 million in revenue and a $249.2 million net loss from continuing operations attributable to common shareholders. It ended May with $4.2 billion in cash, cash equivalents, and restricted cash against $5 billion in debt.

The company attributed much of its fourth-quarter expense growth to stock-based compensation and business expansion. Investors must decide which expenses are temporary consequences of scaling and which will remain embedded in the model.

The quarterly results support both interpretations. Applied Digital has operating facilities and growing rent, but its balance sheet and losses show why construction success alone does not settle the valuation debate.

This is the reversal driving the selloff. Contracts once served as sufficient evidence of future growth. Now they are the starting point for a more demanding examination of capital efficiency.

What the Bull Case Still Gets Right

The drawdown does not erase the scarcity value of grid-connected sites, experienced construction teams, or signed customers.

AI developers require dense clusters of accelerators for model training and inference. These systems need electricity, cooling, networking, physical security, and reliable operations at a scale that cannot appear instantly.

Grid interconnections have become particularly valuable because utilities must evaluate how large computing loads affect local transmission and generation. A company with usable power can move faster than a developer starting with undeveloped land.

IREN’s position combines power, facilities, and cloud operations. That structure lets it sell computing capacity directly instead of earning only rent from a tenant.

The model offers more upside if GPU utilization and pricing remain strong. It also lets IREN work with customers that need managed infrastructure rather than an empty powered building.

Customer prepayments offer another advantage. They shift part of the hardware funding requirement to customers and provide evidence that buyers are committing resources before delivery.

IREN reported approximately $7.6 billion in cash and cash equivalents as of June 30, including restricted cash connected to GPU financing. That liquidity supports its expansion, although restricted funds cannot always be used freely.

TeraWulf’s bull case rests on long-duration leases and power development expertise. A 20-year agreement with Anthropic can generate predictable revenue if the company finishes the campus on schedule and meets performance requirements.

The Abernathy sale also shows that development rights can have value before a facility reaches full operation. Recycling capital from a joint venture into wholly owned sites could improve long-term economics if TeraWulf executes successfully.

Applied Digital provides the clearest evidence that the transition from mining infrastructure to AI hosting can produce real rental revenue. Its first Polaris Forge building was fully operational during the reported quarter, generating base rent and tenant recoveries.

The company’s HPC Hosting segment reported $26.2 million in operating profit for the quarter. That figure does not represent companywide profitability, but it demonstrates that delivered capacity can support positive segment economics.

Applied Digital also operates 286 megawatts of cryptocurrency hosting capacity across North Dakota. Management said those facilities were fully utilized at the end of May.

This legacy activity offers current cash generation while the AI campuses develop. It also carries exposure to cryptocurrency mining economics and customer demand, which can make the overall financial story harder to value.

The broader competitive environment supports the demand thesis. Microsoft has signed major capacity agreements with IREN and other specialized providers because AI computing requirements exceed what hyperscalers can satisfy through internal construction alone.

A previously disclosed Microsoft agreement gave IREN a five-year cloud services commitment and included customer prepayment. The Microsoft capacity deal demonstrated that specialized operators can become strategic suppliers rather than speculative landlords.

These companies also occupy different parts of the infrastructure stack. IREN sells cloud capacity, TeraWulf is leaning toward long-term leasing, and Applied Digital combines development, ownership, hosting, and tenant fit-out work.

That diversity matters because the data center trade is not one uniform bet. A slowdown in GPU rental pricing would hurt a cloud operator differently from a landlord with a long-duration lease.

The bullish conclusion remains credible: AI computing demand is real, power is constrained, and customers are signing substantial agreements. The unresolved issue is how much of that value will reach common shareholders after construction and financing costs.

Google News Headlines Cannot Resolve the Funding Risk

Neither bullish contracts nor bearish headlines can answer the hardest question: who funds the build, and on what terms?

Google News can surface every lease, partnership, earnings release, and daily stock move. It cannot standardize the financial obligations hidden behind different business models.

An AI cloud operator may purchase GPUs directly, finance them through secured arrangements, or use customer prepayments. A landlord may rely on project debt, equity partners, or tenant-supported financing.

Each structure allocates risk differently. Debt raises fixed obligations, while issuing shares can dilute existing investors. Customer financing lowers capital needs but can strengthen the customer’s contractual protections.

IREN’s target depends partly on commissioning large amounts of GPU capacity before year-end. Any delay can push recognized revenue into a later reporting period while interest, staffing, and construction expenses continue.

Its ARR metric also assumes utilization and pricing. If market rental rates weaken, annualized economics can differ from the headline target even when the systems are technically available.

Hardware cycles add another risk. GPUs can remain useful for years, but new generations may deliver better performance or efficiency. Operators must recover hardware costs before customers shift toward newer systems.

TeraWulf faces less direct GPU obsolescence under a property-focused lease model. Its larger exposure lies in construction costs, delivery schedules, power equipment, financing, and counterparty performance.

The Anthropic lease is expected to receive investment-grade credit support, according to TeraWulf. Investors still need details about guarantees, termination rights, construction milestones, and remedies for delays.

The concentration issue cuts both ways. A large tenant can support project financing and make a campus economically viable. It can also gain bargaining power and make the developer dependent on one customer’s plans.

Applied Digital’s balance sheet makes funding risk especially visible. Its cash position is substantial, but so is its debt, and multiple campuses remain under construction.

The company has used secured notes, credit facilities, and other financing arrangements to support expansion. Those tools can match long-lived assets with long-term funding, but they increase sensitivity to interest costs and completion schedules.

Its revenue composition also deserves scrutiny. Tenant fit-out revenue drove much of the latest quarterly increase, while recurring base rent remained a smaller component.

That does not make fit-out revenue unimportant. It represents real work tied to customer deployment. However, investors should track whether base rent grows as scheduled after each building enters service.

Accounting presentation can complicate comparisons further. Applied Digital consolidates ChronoScale because it owns approximately 96%, while excluding ChronoScale from certain non-GAAP measures used to describe core operations.

Readers should therefore compare GAAP revenue, losses, segment results, and cash flows rather than relying on a single adjusted metric. Differences in business scope can make peer comparisons misleading.

Short interest and market positioning may intensify daily moves, but they do not provide a complete explanation for a month-long sector decline. Claims that trading activity alone caused the selloff remain difficult to verify.

Macro conditions can also change valuation multiples without altering company operations. Higher required returns reduce the present value of cash expected several years from now, which particularly affects long-duration construction stories.

The cautious view is not that these projects will fail. It is that current disclosures leave a wide range of possible outcomes for cost, timing, and shareholder returns.

Investors should resist treating contracted revenue as guaranteed profit. Leases and cloud agreements create obligations on both sides, and delivery remains essential.

They should also avoid treating a falling share price as evidence that demand has disappeared. The companies continued signing customers and delivering capacity during the decline.

The market is conducting a funding and execution test. That test will be resolved through operational data, not headlines alone.

Three Signals Will Show Whether More Pain Is Coming

The next move depends on commissioning, recurring revenue, and financing discipline, in that order.

The first signal is IREN’s progress toward its 2026 AI Cloud target. Investors should watch commissioned megawatts, accepted GPU clusters, utilization, recognized cloud revenue, and customer prepayments.

Reaching physical delivery milestones on time would strengthen the case that IREN can turn contracts into operating scale. A schedule slip would weaken the current ARR narrative because fewer systems would contribute revenue by year-end.

The distinction between installed and accepted capacity will be crucial. Hardware can sit in a data hall before networking, software, testing, and customer approval are complete.

IREN should also show whether its expanding customer base produces diversified revenue. Multiple customers can reduce concentration, but only if deployments become material and contract economics remain attractive.

The second signal is TeraWulf’s August 5 quarterly update and its subsequent financing disclosures. The company has scheduled its second-quarter results call for that date.

Investors need a clear funding bridge for the Justified Data campus, including construction phases, expected capital sources, and the impact of the Abernathy transaction. They also need evidence that existing facilities are progressing as planned.

Specific construction milestones will carry more weight than another broad demand statement. Procurement, site preparation, utility work, and building delivery can show whether the 2027 service target remains credible.

If TeraWulf secures funding without unexpected dilution and maintains its schedule, the selloff will look more like a valuation reset. If financing costs rise or completion dates move, the risk discount will remain justified.

The third signal is Applied Digital’s conversion of contracted load into recurring base rent. Its fourth-quarter results showed that operating facilities can generate meaningful revenue, but fit-out services still dominated HPC Hosting sales.

The next reports should reveal whether additional delivered capacity lifts rental revenue and segment operating profit. Investors should also monitor debt, restricted cash, interest expense, and construction spending.

Applied Digital reported that another 75 megawatts at Polaris Forge 1 became live after the fiscal year ended. That delivery should provide a measurable test in later results.

A growing share of base rent would reinforce management’s long-term infrastructure thesis. Continued dependence on fit-out work, without matching rental growth, would weaken it.

These signals matter more than any isolated Google News headline because they test the full conversion chain. Power must become a completed building, the building must host accepted equipment, and accepted equipment must generate cash.

The three companies do not need perfect execution to recover. They need enough evidence to narrow the range of possible outcomes that investors currently apply to their valuations.

More pain remains possible because expectations are still high and development timelines are long. Yet a synchronized decline does not prove that IREN, TeraWulf, or Applied Digital has lost its competitive position.

Readers should follow the next earnings releases with three questions in mind. Did capacity enter service on schedule, did recurring revenue rise, and did financing preserve value for shareholders?

If the answers improve together, the decline will have marked a reset in expectations rather than the end of the AI data center trade. If one link keeps failing, the market’s skepticism will have further to run.