From Megawatts to Equity Value: Who Captures the AI Infrastructure Buildout?
Published · Dilikoglu Research
Dilikoglu Research | Sector Commentary | 2 October 2026 Tugay Dilikoğlu | Information cutoff: 2 October 2026
Contract growth creates an opportunity. Delivery, asset returns and financing terms determine how much of it belongs to existing shareholders.
Investment takeaways
The relevant unit of comparison is the economic obligation behind each megawatt: facilities, equipment or a complete cloud service.
Customer advances and secured debt can improve equity returns, but their timing, restrictions and repayment obligations matter as much as their size.
Higher pricing deserves a higher valuation only to the extent that it survives operating costs, reinvestment and the claims of new capital.
AI infrastructure companies can report similar capacity growth while producing very different outcomes per share. The difference begins with what they own and sell, then extends to when they collect cash and who finances the assets. Our view is that the most useful investment comparison follows the entire project from customer commitment to cash available to shareholders. Megawatts establish scale; they do not establish the shareholder's return.
Who earns what?
A data-center landlord primarily earns rent on facilities. A hosting provider adds operating responsibilities around customer equipment. A GPU infrastructure operator supplies the computing assets themselves; an AI cloud platform adds orchestration, software and services. These activities overlap, but their capital obligations differ.
IREN's Microsoft agreement supplies GPU services from its Childress facilities, rather than merely leasing space.1 Nebius combines cloud services with dedicated infrastructure contracts.2 CoreWeave's financing discussed below supports compute deployments.3 Cipher's Barber Lake announcement concerns facility leases.4 Classification should therefore follow the project, rather than assign every activity a single company-wide label.
Business models and equity-value drivers
Business model
Revenue basis
Typical capital obligations
Principal equity-value drivers
Data-center landlord
Contractual rent
Land, power infrastructure, building and cooling
Development yield, lease protection, residual value
Hosting provider
Capacity and service fees
Facility fit-out and operations; equipment ownership varies
Occupancy, cost pass-through, service obligations
GPU infrastructure
Reserved or usage-based compute
Servers, networking and replacement; facilities owned or leased
Pricing, utilization, hardware life, financing
AI cloud platform
Compute plus software and services
Hardware where owned, platform development and customer support
Retention, service economics, utilization, capital burden
Dilikoglu Research analytical classification; responsibilities depend on individual contracts.
A contract is the beginning of the cash-flow schedule
Secured power is an input to development. It becomes revenue-producing capacity only after construction, equipment installation, commissioning and the relevant customer acceptance. Billable capacity and physical utilization can differ under reserved-capacity agreements. Gross facility power and critical IT load also have different denominators: cooling and other overhead prevent a one-for-one comparison.
Total contract value aggregates payments across time. Backlog describes future business under a company's stated definition. Neither is annual revenue. ARR is an annualized operating measure whose definition must be checked; recognized revenue follows actual service delivery and the applicable accounting rules. Cash receipts follow payment terms, potentially arriving before or after revenue.
Nebius's second-quarter 2026 shareholder letter reported annual contract value of $20-25 million per MW on four landmark agreements. Management said most associated capacity would arrive late in 2026, with revenue contributing primarily in 2027.5 Those are disclosed contract observations and a forward delivery expectation, respectively, not evidence that the full run-rate had already been earned.
Financing changes the distribution of returns
Prepayments can reduce construction funding needs, but also reduce later collections. IREN's December 2025 quarterly filing describes Microsoft advances of 20% per tranche, credited against service fees after the twenty-fourth month of the relevant service term.1 They are economically valuable funding, not additional revenue on top of the contract.
A cash bridge must count customer advances once. Starting from accrual earnings, it incorporates cash costs, tax, interest and working-capital changes, including net deferred revenue. If advances already enter operating cash flow, adding them again as separate funding overstates liquidity. Refund rights, restricted cash and draw conditions require separate attention.
From customer commitment to equity cash flow
Customer commitment
Construction, commissioning and acceptance
Billable service: revenue and collections
Operating cash after costs, tax, interest and working capital
Cash available to equity after investment and debt service
Funding requirement: Capex, debt repayments and liquidity needs less operating cash and usable opening cash.
Residual funding: New debt or common equity. New shares dilute existing ownership.
Conceptual DR cash bridge. Customer advances are counted once within operating cash; a funding gap requires financing rather than a distribution.
Borrowing can reduce immediate equity issuance, but repayment and collateral restrictions remain. A facility's headline commitment is not unrestricted cash. Where equity fills the remaining gap, the issue price determines how many new shares participate in future returns. Issuance can still benefit existing holders when the value created exceeds the economic cost of that capital.
Match asset economics to the financing horizon
CoreWeave's $2.6 billion facility, announced on 10 August 2026, has approximately five-year maturity against underlying customer contracts averaging approximately three years. Its disclosed pricing is SOFR plus 5.50%.3 The structure expands financing access for shorter contracts; it also makes renewal or re-leasing economics relevant before debt maturity. Lender participation supports financeability, not a guarantee of common-equity returns.
The investment test is the return on the full capital committed over the asset's economic life. EBITDA excludes replacement spending and financing costs; a high margin alone cannot establish that return. GPU depreciation schedules are accounting estimates, not proof of commercial longevity. Renewal pricing, maintenance, replacement costs and downtime determine how much cash survives. Buildings may outlast servers, but their residual value still depends on location, adaptability and future demand.
Cipher's 25 September announcement illustrates the same discipline for property economics. It extends Barber Lake's contracted duration from ten to twenty years through an amended lease and a separate follow-on commitment. Cipher also bears the first $359.3 million above the initial budget; the tenant reimburses half of further excess costs over the twenty-year term through additional rent.4 Longer income visibility is valuable, but construction exposure and reimbursement timing belong in the same assessment.
There is a strong counterargument to emphasizing capital constraints: scarce capacity, durable customer demand and repeatable asset-backed finance can make expansion highly attractive. IREN's August 27 update reported $3.6 billion of Microsoft-related GPU financing at a 6.0% weighted average interest rate excluding fees; management said financing and prepayments cover 96% of associated GPU capex.6 This supports the case for financing-led growth. It does not imply that all project costs are funded or all future deployments receive identical terms.
CoreWeave's 17 September disclosure of approximately $40 million annualized revenue per MW relates to contracts lasting around three to six months.7 Applying that rate throughout a multiyear fleet forecast would assume repeated renewals at similar economics. Differences in hardware, contract duration, service scope and power measurement must be reconciled before comparing it with other disclosures.
A common analytical framework need not produce identical assumptions. Existing agreements retain their economics; higher prospective pricing applies to new business or explicitly modeled renewals, not unchanged signed agreements. Undisclosed contract terms do not establish that capacity is unsold. Software should not receive a second valuation if its earnings are already capitalized within cloud revenue.
Enterprise value must ultimately reconcile to common equity after debt and other claims, with restricted liquidity, non-core assets and dilution treated consistently. Separate infrastructure value also requires care: assets already supporting capitalized operating cash flows cannot simply be added again. A revenue multiple should be challenged against cash returns after reinvestment, rather than used to bypass that analysis.
What would change our view?
Our assessment would strengthen with repeated on-time acceptance, collections matching contractual schedules, and additional financing that maintains workable collateral coverage and liquidity headroom. Renewals at attractive prices after maintenance and replacement spending would support longer economic lives. Evidence that software improves retention or margins without proportionate capital requirements would also matter.
It would weaken with widening delivery delays, customer refunds, cost overruns exceeding contractual recovery, or debt maturities increasingly dependent on unproven renewals. Repeated equity issuance without corresponding growth in value per share would challenge the investment case even if aggregate revenue continued rising.
The evidence should be assessed together. Attractive contract pricing cannot compensate indefinitely for uneconomic capital requirements; conversely, heavy investment can create substantial equity value when returns exceed its full cost. We favor underwriting that demonstrates both delivery and value retained per share, with financing assessed on executed terms rather than assumed availability.
Sources and evidence notes
All sources accessed 2 October 2026. Company statements are attributed disclosures, not independent verification of future performance. The business-model classification and investment conclusions are Dilikoglu Research analysis; no new numerical forecast or price target is introduced.
Produced independently by Tugay Dilikoğlu, separately from his professional responsibilities. For informational purposes only; not personalized investment advice or an offer to transact. Opinions reflect Dilikoglu Research's assessment at the stated cutoff and may change. This sector commentary does not initiate or change any company rating or target. Refer to the separately published company reports for their dated recommendations, assumptions and report-specific conflict disclosures. Public disclosures do not provide complete project-level cash waterfalls or renewal schedules; no assurance of future financing, performance or investment returns is implied.
Sign up through Beehiiv and confirm your email address to join. Beehiiv manages your subscription and unsubscribe options. Questions? research@dilikogluresearch.com.
Dilikoglu Research | Sector Commentary | 2 October 2026
Tugay Dilikoğlu | Information cutoff: 2 October 2026
Contract growth creates an opportunity. Delivery, asset returns and financing terms determine how much of it belongs to existing shareholders.
Investment takeaways
AI infrastructure companies can report similar capacity growth while producing very different outcomes per share. The difference begins with what they own and sell, then extends to when they collect cash and who finances the assets. Our view is that the most useful investment comparison follows the entire project from customer commitment to cash available to shareholders. Megawatts establish scale; they do not establish the shareholder's return.
Who earns what?
A data-center landlord primarily earns rent on facilities. A hosting provider adds operating responsibilities around customer equipment. A GPU infrastructure operator supplies the computing assets themselves; an AI cloud platform adds orchestration, software and services. These activities overlap, but their capital obligations differ.
IREN's Microsoft agreement supplies GPU services from its Childress facilities, rather than merely leasing space.1 Nebius combines cloud services with dedicated infrastructure contracts.2 CoreWeave's financing discussed below supports compute deployments.3 Cipher's Barber Lake announcement concerns facility leases.4 Classification should therefore follow the project, rather than assign every activity a single company-wide label.
Dilikoglu Research analytical classification; responsibilities depend on individual contracts.
A contract is the beginning of the cash-flow schedule
Secured power is an input to development. It becomes revenue-producing capacity only after construction, equipment installation, commissioning and the relevant customer acceptance. Billable capacity and physical utilization can differ under reserved-capacity agreements. Gross facility power and critical IT load also have different denominators: cooling and other overhead prevent a one-for-one comparison.
Total contract value aggregates payments across time. Backlog describes future business under a company's stated definition. Neither is annual revenue. ARR is an annualized operating measure whose definition must be checked; recognized revenue follows actual service delivery and the applicable accounting rules. Cash receipts follow payment terms, potentially arriving before or after revenue.
Nebius's second-quarter 2026 shareholder letter reported annual contract value of $20-25 million per MW on four landmark agreements. Management said most associated capacity would arrive late in 2026, with revenue contributing primarily in 2027.5 Those are disclosed contract observations and a forward delivery expectation, respectively, not evidence that the full run-rate had already been earned.
Financing changes the distribution of returns
Prepayments can reduce construction funding needs, but also reduce later collections. IREN's December 2025 quarterly filing describes Microsoft advances of 20% per tranche, credited against service fees after the twenty-fourth month of the relevant service term.1 They are economically valuable funding, not additional revenue on top of the contract.
A cash bridge must count customer advances once. Starting from accrual earnings, it incorporates cash costs, tax, interest and working-capital changes, including net deferred revenue. If advances already enter operating cash flow, adding them again as separate funding overstates liquidity. Refund rights, restricted cash and draw conditions require separate attention.
Funding requirement: Capex, debt repayments and liquidity needs less operating cash and usable opening cash.
Residual funding: New debt or common equity. New shares dilute existing ownership.
Conceptual DR cash bridge. Customer advances are counted once within operating cash; a funding gap requires financing rather than a distribution.
Borrowing can reduce immediate equity issuance, but repayment and collateral restrictions remain. A facility's headline commitment is not unrestricted cash. Where equity fills the remaining gap, the issue price determines how many new shares participate in future returns. Issuance can still benefit existing holders when the value created exceeds the economic cost of that capital.
Match asset economics to the financing horizon
CoreWeave's $2.6 billion facility, announced on 10 August 2026, has approximately five-year maturity against underlying customer contracts averaging approximately three years. Its disclosed pricing is SOFR plus 5.50%.3 The structure expands financing access for shorter contracts; it also makes renewal or re-leasing economics relevant before debt maturity. Lender participation supports financeability, not a guarantee of common-equity returns.
The investment test is the return on the full capital committed over the asset's economic life. EBITDA excludes replacement spending and financing costs; a high margin alone cannot establish that return. GPU depreciation schedules are accounting estimates, not proof of commercial longevity. Renewal pricing, maintenance, replacement costs and downtime determine how much cash survives. Buildings may outlast servers, but their residual value still depends on location, adaptability and future demand.
Cipher's 25 September announcement illustrates the same discipline for property economics. It extends Barber Lake's contracted duration from ten to twenty years through an amended lease and a separate follow-on commitment. Cipher also bears the first $359.3 million above the initial budget; the tenant reimburses half of further excess costs over the twenty-year term through additional rent.4 Longer income visibility is valuable, but construction exposure and reimbursement timing belong in the same assessment.
There is a strong counterargument to emphasizing capital constraints: scarce capacity, durable customer demand and repeatable asset-backed finance can make expansion highly attractive. IREN's August 27 update reported $3.6 billion of Microsoft-related GPU financing at a 6.0% weighted average interest rate excluding fees; management said financing and prepayments cover 96% of associated GPU capex.6 This supports the case for financing-led growth. It does not imply that all project costs are funded or all future deployments receive identical terms.
Comparable assumptions require comparable economics
CoreWeave's 17 September disclosure of approximately $40 million annualized revenue per MW relates to contracts lasting around three to six months.7 Applying that rate throughout a multiyear fleet forecast would assume repeated renewals at similar economics. Differences in hardware, contract duration, service scope and power measurement must be reconciled before comparing it with other disclosures.
A common analytical framework need not produce identical assumptions. Existing agreements retain their economics; higher prospective pricing applies to new business or explicitly modeled renewals, not unchanged signed agreements. Undisclosed contract terms do not establish that capacity is unsold. Software should not receive a second valuation if its earnings are already capitalized within cloud revenue.
Enterprise value must ultimately reconcile to common equity after debt and other claims, with restricted liquidity, non-core assets and dilution treated consistently. Separate infrastructure value also requires care: assets already supporting capitalized operating cash flows cannot simply be added again. A revenue multiple should be challenged against cash returns after reinvestment, rather than used to bypass that analysis.
What would change our view?
Our assessment would strengthen with repeated on-time acceptance, collections matching contractual schedules, and additional financing that maintains workable collateral coverage and liquidity headroom. Renewals at attractive prices after maintenance and replacement spending would support longer economic lives. Evidence that software improves retention or margins without proportionate capital requirements would also matter.
It would weaken with widening delivery delays, customer refunds, cost overruns exceeding contractual recovery, or debt maturities increasingly dependent on unproven renewals. Repeated equity issuance without corresponding growth in value per share would challenge the investment case even if aggregate revenue continued rising.
The evidence should be assessed together. Attractive contract pricing cannot compensate indefinitely for uneconomic capital requirements; conversely, heavy investment can create substantial equity value when returns exceed its full cost. We favor underwriting that demonstrates both delivery and value retained per share, with financing assessed on executed terms rather than assumed availability.
Sources and evidence notes
All sources accessed 2 October 2026. Company statements are attributed disclosures, not independent verification of future performance. The business-model classification and investment conclusions are Dilikoglu Research analysis; no new numerical forecast or price target is introduced.
Research disclosure
Produced independently by Tugay Dilikoğlu, separately from his professional responsibilities. For informational purposes only; not personalized investment advice or an offer to transact. Opinions reflect Dilikoglu Research's assessment at the stated cutoff and may change. This sector commentary does not initiate or change any company rating or target. Refer to the separately published company reports for their dated recommendations, assumptions and report-specific conflict disclosures. Public disclosures do not provide complete project-level cash waterfalls or renewal schedules; no assurance of future financing, performance or investment returns is implied.