They All Got Butchered. Why?

CoreWeave, Nebius and IREN got hammered. Did the fundamentals really change – or is the market questioning the economics of building AI infrastructure? And where may the long-term opportunity lie?

Three companies. One enormous AI infrastructure race. A brutal market sell-off.

CoreWeave, Nebius and IREN are building businesses around the same structural trend: growing demand for computing power to train and run artificial intelligence.

Yet their shares have come under significant pressure as investors question debt, capital expenditure, financing costs and how quickly infrastructure investments can generate sustainable cash flow.

Reuters reported in October that short sellers had accumulated approximately $6 billion in unrealised gains on bearish positions in AI cloud and data-centre companies. That illustrates the deterioration in market sentiment, but it does not prove that AI demand is collapsing.

Our central thesis: the market is increasingly questioning the economics of supplying AI infrastructure — not simply the long-term demand for AI itself.

The distinction matters. Demand can grow while infrastructure stocks fall if investors believe the cost of meeting that demand is too high or the returns will take too long to materialise.

Our research focuses on three different investment cases:

• CoreWeave (CRWV): a major AI cloud provider with a substantial contracted backlog and significant financing requirements.
• Nebius (NBIS): a rapidly expanding AI cloud platform with exceptional revenue growth and demanding growth expectations.
• IREN (IREN): an infrastructure business transitioning from Bitcoin mining towards AI cloud services, with potential upside if its expansion translates into profitable revenue.

Among the three, IREN is our preferred candidate for deeper long-term investigation. Its power and data-centre infrastructure, growing AI Cloud business and experience operating energy-intensive computing facilities create a differentiated investment case.

That does not mean IREN is necessarily the cheapest stock, the safest business or the most likely winner. Its transition, financing needs and execution risks remain substantial.

The central question for 2027: which company can turn AI demand into sustainable cash flow — and create the most value for shareholders along the way?

01 — Why the sell-off matters

The first phase of the AI boom rewarded companies that could secure GPUs, build data centres and sign major customer contracts.

The next phase is more demanding. Investors increasingly want to know whether those contracts will generate sufficient returns to justify the capital required to deliver them.

The capital expenditure problem

AI infrastructure requires more than GPUs. Companies must fund data-centre construction, electricity infrastructure, cooling, networking, hardware and ongoing operations. Much of this investment occurs before the associated capacity begins generating revenue.

That creates a financing gap: companies spend enormous amounts today in anticipation of revenue over several future years. If borrowing costs rise, construction is delayed or customers change their deployment schedules, expected returns can deteriorate quickly.

Revenue is not the same as cash flow

A large backlog or contracted annual recurring revenue can provide visibility into future demand. Neither guarantees profit. Investors need to understand:

• How much capital is required to deliver contracted capacity?
• When will new capacity become operational?
• How much revenue will each deployment generate?
• What remains after operating costs, interest and further investment?
• Will the company need to issue additional shares to finance expansion?

These questions explain why companies can report rapid growth while their share prices fall sharply.

The financing signal

CoreWeave's financial disclosures explicitly warn that the business expects significant infrastructure investment to continue and that future requirements may necessitate additional debt or equity financing. Equity issuance would dilute existing shareholders (source: CoreWeave's Q2 2026 SEC filings).

This is the central tension across the sector: infrastructure may be valuable, but building it requires capital before the full economic benefit arrives.

Grabit Labs view: the market is increasingly distinguishing between demand for AI and the financial returns available to the companies supplying that demand.

02 — CoreWeave (CRWV): growth at a high financial cost

The business

CoreWeave operates specialised cloud infrastructure for demanding AI workloads. Its customers need access to high-performance GPUs without necessarily building all the required infrastructure themselves. CoreWeave's opportunity is to provide that capacity at scale and convert long-term customer commitments into recurring revenue.

What the numbers show

Source: CoreWeave's Q2 2026 results filed with the SEC.

The backlog is substantial, but it is not cash in the bank. It represents amounts expected to be recognised in future periods, subject to delivery and service requirements. CoreWeave must still build and finance the infrastructure required to fulfil those commitments.

Its filings also describe significant accumulated losses, negative cash flows from investing activities and the possibility of further debt or equity financing.

The bull case

If AI computing demand continues expanding, CoreWeave could benefit from strong utilisation, customer commitments and favourable pricing. Its backlog offers visibility into future business. A recovery in financing conditions and better cash generation could also help investor confidence.

The bear case

If debt becomes more expensive, customer deployments are delayed or the company must invest more than expected to fulfil its contracts, returns could deteriorate. The risk is not necessarily a lack of demand. It is that financing and infrastructure costs consume too much of the economic benefit generated by that demand.

What matters in 2027

• Free cash flow relative to capital expenditure.
• Debt financing and liquidity.
• Conversion of backlog into operational capacity and recognised revenue.
• Customer concentration and contract economics.
• Whether expansion can be financed without materially worsening shareholder economics.

Our assessment: CoreWeave offers direct exposure to AI infrastructure demand, but its financing burden makes cash generation a critical test. Strong demand alone will not settle the investment case.

03 — Nebius (NBIS): exceptional growth, high expectations

The business

Nebius is building an AI cloud platform for customers that need large amounts of computing capacity. Its opportunity extends beyond supplying hardware: the company is developing the infrastructure, software and operational capabilities needed to serve AI customers at scale. That creates the potential for a substantial cloud business, but also requires continued investment.

What the numbers show

Nebius reported second-quarter 2026 revenue of approximately $582.3 million, up 454% year over year. Adjusted EBITDA was approximately $236.2 million (source: Nebius Q2 2026 results).

The revenue growth is impressive, but adjusted EBITDA is not free cash flow. Investors must also account for capital expenditure, depreciation, financing and the cost of bringing new capacity online.

The bull case

If demand continues growing and Nebius successfully deploys new capacity, revenue could scale substantially. Customer prepayments and financing arrangements may help the company manage capital requirements.

The bear case

The risk is that expansion costs rise faster than expected or new capacity takes longer to become productive. If the company needs substantial additional capital before generating the anticipated cash flow, investors may demand a lower valuation. Nebius also faces a high bar because its market valuation reflects considerable future growth.

What matters in 2027

• Revenue growth relative to capital expenditure.
• GPU utilisation and deployment schedules.
• Operating margins and free cash flow.
• Financing terms and potential shareholder dilution.
• Whether growth justifies the valuation investors are paying today.

Our assessment: Nebius has demonstrated exceptional growth, but its investment case depends on translating that growth into sustainable economic returns. The question is not simply whether the company can grow — it is how much value that growth creates for shareholders.

04 — IREN (IREN): the infrastructure transition

Why IREN deserves special attention

IREN has a different starting point from a conventional AI cloud company. Its historical business has included Bitcoin mining, which requires substantial electricity access, physical infrastructure, cooling and large-scale computing operations.

The company is transitioning towards AI cloud services and high-performance computing. This creates a potentially attractive combination: existing infrastructure and operating experience alongside exposure to a growing market for AI computing. But the transition also introduces execution risk.

What the numbers show

IREN's fiscal year 2026 ended on June 30, 2026. Its full-year results reported:

Source: IREN's FY2026 results, released August 27, 2026.

AI Cloud revenue grew approximately eightfold from FY2025, but Bitcoin mining still generated most of the company's full-year revenue.

The net loss was affected by $638.8 million in non-cash impairments, primarily related to decommissioning mining hardware as sites were converted to support AI Cloud growth. Those charges should be distinguished from recurring operating expenses, but they demonstrate that changing the business model has real economic consequences.

Contracted AI demand

IREN reported approximately $4 billion in contracted annual recurring revenue associated with 2026 capacity, with approximately $1 billion in ARR operating at the time of its August update. The company also reported delivery of its first 50 MW liquid-cooled deployment at Childress to Microsoft, with additional deployments planned. IREN targeted approximately 0.8 GW of cumulative delivered IT capacity during 2027.

These figures demonstrate the scale of the ambition. They do not guarantee that every deployment will be completed on schedule or that all contracted revenue will translate into free cash flow. IREN's disclosures explain that its ARR target depends on assumptions about capacity, GPU models, contracting, utilisation and pricing, and that revenue ramps as GPUs are commissioned, tested and accepted by customers.

The power and infrastructure advantage

AI data centres require reliable power, cooling, networking and physical sites capable of supporting high-density computing. Securing those resources can be difficult and time-consuming. IREN's existing infrastructure and experience operating energy-intensive facilities may help it compete for AI workloads.

However, power access alone does not guarantee attractive returns. IREN must also finance equipment, commission facilities, deliver customer contracts and operate the business profitably.

The Bitcoin connection

Its experience with energy-intensive computing could help support the transition towards AI. But we should not assume that Bitcoin mining and AI Cloud will remain two equally important earnings engines in 2027. IREN has stated that it expects to transition substantially away from mining by the end of 2026.

The key question: can IREN replace its historical mining economics with a larger, more valuable AI infrastructure business?

If the transition succeeds, the company could develop a different earnings profile. If AI deployments are delayed, it could lose some historical revenue before the new business reaches the expected scale.

The bull case

IREN successfully deploys new capacity, converts contracted demand into recognised revenue and demonstrates attractive returns on its investment. In this scenario, investors may begin valuing IREN more as an AI infrastructure company and less through its historical mining business.

The bear case

Deployment schedules slip, capital expenditure rises or financing becomes more expensive. The company needs additional capital before its AI infrastructure generates the anticipated returns, dilution increases and the transition away from mining creates a gap in revenue. In this scenario, AI demand can remain strong while IREN's shares continue to struggle.

What matters in 2027

• Delivery of planned capacity.
• Conversion of contracted ARR into recognised revenue.
• Cash generation relative to capital expenditure.
• Financing costs and potential dilution.
• Customer concentration and contract economics.
• Returns on invested capital.
• The pace and economics of the transition away from Bitcoin mining.

Our assessment: IREN is a particularly interesting long-term research candidate because it combines infrastructure, power access, AI expansion and a transition from Bitcoin mining. But its potential depends on execution and financing, not simply on the growth of AI.

05 — Valuation: which company offers the most interesting setup?

A major share-price decline does not automatically make a stock cheap. The right question is whether the current valuation is attractive relative to the company's future revenue, cash flow, debt and execution risk.

The following market snapshot uses October 23, 2026 closing data and trailing revenue through June 2026.

Source: Second Order AI, using a Financial Modeling Prep daily snapshot dated October 23, 2026. Multiples are approximate and depend on the treatment of cash, debt and revenue periods.

CoreWeave: a lower historical revenue multiple

CoreWeave screens cheaper than Nebius and IREN on this comparison. However, the gap between its market capitalisation and enterprise value highlights the importance of financing. A lower revenue multiple does not automatically mean a better investment when the business requires substantial capital and incurs significant financing costs.

Nebius: high growth expectations

Nebius has the highest multiple. That does not mean the stock cannot rise. It means investors are paying for substantial future growth, so the company must execute well to justify the valuation.

IREN: a potential transition opportunity

IREN's multiple sits between the other two. Its smaller enterprise value and transition towards AI Cloud make it worth investigating, particularly if future AI revenue and cash flow grow faster than the market expects.

But the available figures do not establish that IREN is the cheapest stock. A separate valuation model published by Investing.com on October 19 assessed CoreWeave as having more modelled upside than IREN or Nebius at that time — one model, on earlier prices, not a universal measure of fair value.

Our view: IREN may offer an interesting risk-and-reward setup for a long-term investor, but a defensible claim that it is the best-valued company requires a forward-looking model of revenue, capital expenditure, financing and free cash flow.

06 — Why we are particularly interested in IREN

1. Existing infrastructure

IREN has experience operating energy-intensive facilities and developing large-scale computing infrastructure. That may support its transition towards AI, although the suitability and economics of individual sites must be evaluated.

2. AI Cloud growth

The AI Cloud business has grown rapidly from a relatively small base. If IREN can continue expanding and convert that growth into sustainable cash flow, its earnings profile could change significantly.

3. The transition itself may create an opportunity

Business transitions can be difficult for markets to value. Investors may focus on historical earnings, current capital expenditure or short-term losses without fully accounting for the potential value of future AI capacity. The reverse is also possible: investors may overestimate how quickly the transition can be completed or how profitable it will become.

4. A different path into AI infrastructure

IREN's history in Bitcoin mining differentiates it from a pure-play AI cloud provider. Its operating experience may be useful, but mining should not be treated as a guaranteed second growth engine throughout 2027. IREN's future value increasingly depends on the economics of its new business.

Our long-term thesis: IREN deserves close attention because its infrastructure transition could create value that is not fully reflected in its current valuation. That thesis needs confirmation through capacity delivery, financing discipline and improving cash generation.

07 — 2027: three possible scenarios

The next phase will not be determined by demand alone. Financing, execution, power availability and the economics of deployed capacity will determine which companies turn growth into shareholder value.

Bull case — AI infrastructure becomes a cash-flow engine

AI adoption keeps expanding and customers pay for the infrastructure being built today. CoreWeave converts backlog into revenue and improves cash generation. Nebius scales its platform while controlling costs. IREN delivers new capacity and demonstrates attractive returns. The recent sell-off may eventually be viewed as a period when investors became overly focused on near-term financing risk.

What would confirm it: on-time delivery, strong customer demand, improving margins and financing that does not require excessive dilution.

Base case — growth continues, but the market becomes more selective

AI demand remains strong, but the cost of supplying it stays high. Revenue grows across the sector, yet capital expenditure, depreciation and financing costs limit free cash flow. Some projects are delayed. A few companies establish durable positions, while others struggle to justify their valuations.

What would confirm it: continued revenue growth alongside uneven cash generation, high capital requirements and substantial volatility around earnings reports.

Bear case — the economics fail to meet expectations

AI infrastructure spending slows or customers delay deployments. Capacity becomes more competitive, pricing weakens and financing becomes more expensive. Companies with large commitments may need to raise capital or cut expansion plans. The problem would not be that AI disappears — but that providers invested too much, too early, or at costs that cannot be recovered quickly enough.

What would confirm it: repeated deployment delays, weaker pricing, rising financing costs, deteriorating cash flow and significant changes to customer commitments.

These are conditional scenarios, not price targets or probability-weighted forecasts.

08 — What would change our view?

A credible investment thesis must include the conditions that could prove it wrong.

We would become more constructive on IREN if:

• AI capacity is delivered on schedule.
• Contracted demand converts into recognised revenue.
• Cash generation improves relative to capital expenditure.
• Financing remains manageable without excessive dilution.
• Returns on invested capital support the company's valuation.

We would become more cautious if:

• Capacity delivery repeatedly slips.
• Financing requirements materially exceed expectations.
• Customer commitments are delayed or reduced.
• AI infrastructure pricing weakens.
• The transition away from mining creates a revenue gap before AI earnings scale.

These milestones are more informative than short-term price movements alone.

09 — The bigger picture

The strongest technology company, the strongest business and the best-performing stock can be three different things. A company can have excellent technology but poor financial discipline. A business can grow rapidly but deliver weak shareholder returns because it needs too much capital. Another can have a less impressive growth story but offer better returns if its valuation is attractive enough.

That is why our analysis does not stop at AI demand. We want to understand who can build the infrastructure, finance it responsibly and turn it into sustainable cash flow.

Grabit Labs view

We believe the recent AI infrastructure sell-off deserves a closer look from long-term investors. It does not automatically mean the AI opportunity is over. It may instead mark a shift in what investors demand from the companies supplying that infrastructure.

Growth alone may no longer be enough. The market increasingly needs evidence of financing discipline, reliable execution and sustainable cash generation.

Among CoreWeave, Nebius and IREN, we see IREN as a particularly interesting long-term research candidate because of its infrastructure base and transition towards AI Cloud. But the thesis must be tested against valuation, financing requirements and execution milestones.

The AI infrastructure race is not just about who builds the most capacity. It is about who captures the most value from it.

This report is for informational and educational purposes only. It is not investment advice or a recommendation to buy or sell securities. Contracted ARR, backlog, adjusted EBITDA and planned capacity are not equivalent to recognised revenue, free cash flow or shareholder returns. Valuations reflect market prices on the date stated and will change.