J.P. Morgan estimates that hyperscaler capital expenditure will reach $697 billion in 2026. Morgan Stanley estimates global data center investment through 2028 at approximately $2.9 trillion. The capital scale is unprecedented. The accounting complexity is equally unprecedented and, unlike the capital, has not been resolved.
On April 6, 2026, FASB Chair Rich Jones added a research project to FASB's agenda to monitor current accounting trends related to data infrastructure investments, including data centers, the financing behind them, and their use in a company's business and related contractual arrangements. The CFO Brew and Accounting Today confirmed the project was added specifically because accounting treatment for data center projects "is drawing scrutiny." The project is in a research phase: FASB staff will present information to the board to determine whether to add items to the technical agenda. No timeline was given.
What this means for CFOs making AI data center accounting decisions in Q3 2026: there is no authoritative GAAP guidance on the specific multi-standard issues that AI data center arrangements create. Every policy decision your company makes before FASB issues guidance will be retroactively evaluated when guidance eventually arrives.
A typical AI data center arrangement simultaneously triggers five GAAP standards, each of which requires separate analysis, and the interaction among the five has not been addressed by any FASB pronouncement, EITF consensus, or SEC staff guidance:
ASC 810, Consolidation: does the data center special purpose vehicle require consolidation as a variable interest entity?
ASC 842, Leases: is the capacity arrangement a lease subject to right-of-use asset recognition or a service contract?
ASC 815, Derivatives and Hedging: is the power purchase agreement that supplies electricity to the data center a derivative?
ASC 360, Property, Plant and Equipment: what development costs are capitalisable to the data center asset?
ASC 740, Income Taxes: does the OBBBA's Section 168(n) qualified production property election apply to an AI data center facility?
This blog covers each standard, the specific accounting question it raises in the data center context, the current state of GAAP on that question, and what accounting policy decision your CFO and controller must document before Q3 close.
Why AI Data Centers Create a Multi-Standard Technical Accounting Problem With No Authoritative GAAP Answer
A traditional data center owned and operated by a company for its own IT infrastructure is accounted for under ASC 360 as property, plant and equipment, with straightforward capitalisation of construction costs and depreciation over the asset's useful life. The accounting is unremarkable.
An AI data center arrangement in 2026 is different in three structural features that each trigger a separate accounting analysis.
First, the ownership and financing structure. Many AI data center projects are financed through special purpose vehicles, sale-leaseback arrangements, project finance structures, or joint ventures with hyperscalers, utilities, or financial investors. The SPV or joint venture structure triggers the ASC 810 VIE consolidation analysis before the arrangement can be reflected in the company's financial statements.
Second, the capacity arrangement. Rather than owning a data center outright, many AI companies and enterprise customers enter into capacity reservation agreements with hyperscalers or third-party data center operators. These arrangements may be structured as service contracts, in which the company purchases computing capacity as a variable service, or as leases, in which the company has the right to direct the use of an identified asset for a period of time. The distinction triggers the ASC 842 lease vs service contract analysis.
Third, the power supply arrangement. AI data centers consume enormous quantities of electricity. The June 2026 Hut 8 Beacon Point data center financing, led by J.P. Morgan on a $4.25 billion senior secured note offering, involved a 15-year triple-net lease structure. Many AI data center power supply arrangements involve long-term power purchase agreements with electricity generators. A PPA that locks in a price for electricity delivery over multiple years may be a derivative under ASC 815, depending on whether it meets the definition of a derivative and whether any scope exception applies.
At the March 10, 2026 FASAC meeting, FASAC members flagged that data center investments driven by AI development create multi-footnote disclosure challenges where investors struggle to connect information disclosed across different footnotes. The same arrangement appears in the consolidation footnote, the lease footnote, the derivatives footnote, and the capex section of MD&A, without a single integrated disclosure that explains how the pieces relate to each other.
The KPMG report on data centre accounting confirmed: choosing whether to classify a data center as property, plant and equipment or as investment property "can have a huge impact" on the financial statements. The same is true for every other accounting classification decision in this multi-standard landscape.
What Did FASAC Flag in March 2026 About AI Data Center Accounting?
FASB's Financial Accounting Standards Advisory Council met on March 10, 2026. FASAC is the external advisory body that provides input to FASB on its standard-setting agenda. At the March 2026 meeting, FASAC members specifically flagged data center investments driven by AI development as creating multi-footnote disclosure challenges.
The specific concern FASAC members raised: investors receive information about a company's AI data center investment in fragments. The consolidation analysis appears in the VIE footnote. The lease obligations appear in the lease footnote. The power purchase agreement appears in the derivatives and hedging footnote (if it is a derivative) or in the commitments and contingencies footnote (if it is a service contract). The capitalised construction costs appear in the PP&E footnote. The tax implications appear in the income tax footnote.
None of these footnotes refers to the others. An investor trying to understand the total financial commitment and financial statement impact of a company's AI data center investment must read five separate footnotes, identify the elements of the same underlying arrangement across all five, and construct their own integrated analysis. Most investors cannot do this. Most sell-side analysts choose not to.
The FASAC feedback was directional: FASB should consider whether disclosure improvements are needed to help investors connect the information across footnotes. The March 2026 FASAC meeting preceded the April 6 research project addition by approximately four weeks, and the research project's scope confirms that FASB acted on the FASAC feedback.
What Is FASB's April 2026 Research Project on Data Infrastructure and When Will It Produce Guidance?
On April 6, 2026, FASB Chair Rich Jones added a research project to the research agenda to monitor current accounting trends related to data infrastructure investments and non-traditional lending, including private credit. The project was confirmed across Accounting Today, CFO Dive, CFO Brew, and the FASB's own press release.
The project's scope: the board defines data infrastructure investments as data centers, the financing behind them, their use in a company's business, and related contractual arrangements. This scope is intentionally broad and covers the full range of data centre accounting questions, including the VIE consolidation analysis, the lease classification, the PPA derivative question, and the capitalisation policy.
The project's current stage is research, not technical agenda addition. This is the earliest stage in FASB's standard-setting process. Research projects are monitored by FASB staff, who eventually present information to the board for the board to determine whether to add items to the technical agenda. Adding to the technical agenda is itself a pre-exposure draft stage. An exposure draft, comment period, and final standard would follow after technical agenda addition.
The practical timeline implication: no authoritative GAAP guidance on AI data center accounting is expected in 2026 or likely in 2027. The research phase alone may take 12 to 18 months. A technical agenda addition would follow, then an exposure draft, then a final standard with an effective date typically two or more years after issuance.
For Q3 2026 close, this means companies with AI data center investments are making accounting policy decisions without authoritative GAAP guidance and will be evaluated against whatever guidance FASB eventually issues. The documentation of the analysis and the accounting policy rationale is the only protection available.
The SAB 74 disclosure implication: the research project addition, while not yet a technical agenda item, may nonetheless trigger a SAB 74 disclosure obligation for companies where the data center accounting questions are material. SAB 74 requires disclosure when the impact of a pending standard or standard under development is expected to be material. A company making significant AI data center investments under accounting policies that may need to change when FASB issues guidance should assess whether SAB 74 disclosure is appropriate.
Standard #1: ASC 810, When Does an AI Data Center SPV Require Consolidation?
ASC 810, Consolidation, requires a reporting entity to consolidate a variable interest entity (VIE) if the reporting entity is the primary beneficiary of the VIE. A VIE is an entity that has insufficient equity to finance its activities without additional subordinated financial support, or an entity whose equity investors lack the power to direct the activities that most significantly affect the entity's economic performance.
For AI data center investments structured through SPVs or joint ventures, the consolidation analysis requires:
Step 1: determine whether the SPV is a VIE. If the SPV has been capitalised primarily with debt and a small equity investment that is insufficient to absorb expected losses without additional support, the SPV is likely a VIE. Many data center project finance structures use high leverage (the Hut 8 Beacon Point deal achieved 95% loan-to-cost), meaning the equity cushion is thin. Thin equity is the hallmark of a VIE.
Step 2: determine whether the company is the primary beneficiary. The primary beneficiary is the entity that (a) has the power to direct the activities that most significantly affect the VIE's economic performance, and (b) has the obligation to absorb losses or the right to receive benefits from the VIE that could potentially be significant to the VIE.
For a company that provides the computing workload that will run in the data center, and that has entered into a long-term capacity reservation agreement with the SPV, the power analysis requires assessing whether the workload commitment gives the company power over the data center's operations. A long-term capacity reservation that represents 100% of the data center's revenue and that contains the right to direct the timing and scale of the capacity deployment may give the company power over the SPV's most significant activities.
The consolidation consequence is material: if the SPV must be consolidated, its assets (the data center facility), liabilities (the project finance debt), and revenue (capacity fees) all appear in the company's financial statements. If the SPV is not consolidated, only the company's rights and obligations under the capacity arrangement appear, typically as an ASC 842 right-of-use asset and lease liability or as a service contract commitment.
The practical guidance gap: ASC 810 does not specifically address data center SPVs. The analysis applies the general VIE framework to facts and circumstances that FASB has not specifically addressed, meaning the primary beneficiary conclusion is highly judgment-dependent and will vary across companies with similar data center arrangements.
Standard #2: ASC 842, Is Your Capacity Arrangement a Lease or a Service Contract?
ASC 842-10-15-3 defines a lease as a contract that conveys the right to control the use of an identified asset for a period of time in exchange for consideration. The two key questions in the lease identification analysis are: is there an identified asset? Does the customer have the right to control the use of that asset?
For an AI data center capacity reservation arrangement, the identified asset question is the most consequential. If the data center operator has the substantive right to substitute different computing assets throughout the period of use (for example, substituting one rack of GPUs for another of equivalent capacity), no identified asset exists and the arrangement is a service contract, not a lease. If the customer has the right to use a specific, identifiable set of infrastructure assets (a specific set of servers in a specific data hall at a specific data center facility), an identified asset likely exists.
The substitution right has specific technical requirements under ASC 842-10-15-10: the supplier's substitution right is substantive only if the supplier has the practical ability to substitute alternative assets throughout the period of use and would benefit economically from exercising the substitution right. For AI data center arrangements, the substitution right analysis is complex because the computing assets (GPUs, networking equipment) are becoming obsolete rapidly and the operator would generally prefer to deploy newer, more efficient hardware where possible.
The control question: the customer has the right to control the use of an identified asset if the customer has both the right to obtain substantially all of the economic benefits from use of the asset and the right to direct the use of the asset throughout the period of use. For a capacity reservation arrangement where the customer dictates the workloads run in the facility but the operator manages the physical operations, the direction of use question depends on the specific contractual terms regarding workload scheduling, access rights, and operational control.
The lease vs service contract classification has a direct balance sheet impact. A lease requires recognition of a right-of-use asset and a corresponding lease liability at the present value of the future lease payments. A service contract is expensed as the service is received or recognised as a prepaid if consideration is paid in advance. For a 10-year capacity reservation arrangement with $100 million in annual payments, the lease treatment adds approximately $800 million in assets and liabilities (present value at current interest rates), while the service contract treatment adds nothing to the balance sheet at inception.
The practical determination: the AICPA's December 2025 conference specifically identified this determination as technically complex for AI data center arrangements. No Big 4 or standard-setter guidance has resolved how to apply the substitution right and control analysis to AI-specific capacity reservation structures as of Q3 2026.
Standard #3: ASC 815, Is Your Power Purchase Agreement a Derivative?
Power purchase agreements (PPAs) are long-term contracts between an electricity generator and a data center operator or AI company that lock in the price and volume of electricity to be supplied over the contract period. AI data centers require enormous and predictable electricity supply: a single large-scale GPU cluster may consume hundreds of megawatts continuously.
ASC 815-10-15-83 defines a derivative as a financial instrument or other contract with all of the following characteristics: it has an underlying variable and a notional amount; it requires no initial net investment or an initial net investment that is smaller than would be required for contracts with similar response to market changes; and it can be or requires net settlement.
Applied to a PPA for AI data center electricity supply:
Underlying and notional: a PPA has an underlying variable (the market price of electricity) and a notional amount (the contracted quantity of electricity). These conditions are typically met.
Initial net investment: a PPA typically requires no significant initial net investment (the contract price is paid as electricity is delivered). This condition is typically met.
Net settlement: this is the most judgment-intensive analysis. A PPA is not settled net in the ordinary course if the electricity is physically delivered and consumed. However, if the data center could be connected to the grid and the contracted electricity could be sold to third parties rather than consumed, a net settlement mechanism may exist.
The normal purchases and normal sales (NPNS) scope exception under ASC 815-10-15-22 allows companies to exclude from derivative accounting contracts for the delivery of goods that will be used or sold in the normal course of business. For a data center operator or AI company, electricity consumed in running AI workloads is part of normal operations. If the NPNS exception is elected, the PPA is recorded as a purchase commitment at the contracted price, not as a derivative measured at fair value.
The practical consequence of derivative vs non-derivative classification: a PPA classified as a derivative must be carried at fair value on the balance sheet with changes in fair value recorded through earnings each period. At current electricity prices and price volatility, a 10-year PPA with fixed pricing could have a fair value swing of tens to hundreds of millions of dollars in a single quarter. A PPA that qualifies for the NPNS exception is off-balance-sheet (subject to commitment disclosure) and its fair value is not recorded in the income statement.
The accounting policy election matters: the NPNS exception must be documented and designated at the inception of the PPA. A company that did not formally designate the NPNS exception at inception may not be able to apply it retroactively.
Standard #4: ASC 360, What AI Data Center Development Costs Are Capitalizable?
ASC 360-10-25-1 through 25-5 and ASC 350-40 (for internal-use software within the data center) govern the capitalisation of costs related to property, plant and equipment and internal-use software. For an AI data center, the capitalisation boundary is not obvious across all cost categories.
Clearly capitalisable: direct construction costs (materials, labour, equipment), external direct costs directly related to the asset, and interest costs during the construction period under ASC 835-3 (capitalisation of interest). These are standard PP&E capitalisation items that apply without AI-specific complexity.
Judgment-intensive categories specific to AI data centers:
GPU and accelerator costs: GPUs and other AI accelerators installed in a data center are equipment with a useful life that is typically two to five years (given rapid technology obsolescence), which is shorter than the facility's useful life of 20 to 40 years. The shorter useful life requires component accounting: the GPU cluster is depreciated separately from the building over its own shorter useful life.
Power infrastructure: the transformers, switchgear, and electrical distribution systems installed specifically for AI workloads may have different useful lives from the underlying data center building. These must be separately identified and depreciated.
Cooling systems: AI GPU clusters generate substantially more heat per square foot than traditional server racks. Custom liquid cooling infrastructure installed specifically for AI workloads is a capital cost, but the question is whether it is part of the building's structural cost (depreciated over the building's life) or a separate component with its own useful life (like the GPU cluster, it may become obsolete as GPU thermal profiles change).
Internal-use software and AI model development: costs of developing AI models run in the data center are governed by ASC 350-40 for internal-use software. Development-phase costs are capitalisable; preliminary project phase and post-implementation phase costs are expensed. The distinction between development and preliminary project phases in AI model training (which may span months and involve substantial computing costs) is a specific judgment that has not been authoritatively addressed.
Section 168(n) is addressed in the following section.
Standard #5: ASC 740 / Section 168(n), Can You Elect the QPP Deduction on a New AI Facility?
Section 168(n), the Qualified Production Property election under the OBBBA, allows a 100% deduction in the year placed in service for qualified production property, defined as tangible personal property used in qualified production activities including manufacturing, production, and certain construction activities.
The Section 168(n) question for an AI data center: is an AI data center a qualified production property facility? The OBBBA's definition of qualified production activities focuses on manufacturing and physical production. An AI data center, which runs GPU workloads to train AI models or to serve AI inference requests, is arguably a computing and data processing facility rather than a manufacturing or production facility in the traditional sense.
The IRS has not issued guidance specifically addressing whether AI data centers qualify for Section 168(n). The OBBBA's text uses the term "qualified production property" without specifically including or excluding data centers. The legislative history and the Treasury's regulatory authority will ultimately determine the answer, but as of Q3 2026, no definitive IRS guidance exists.
For tax provision purposes, a company that elects Section 168(n) treatment for a new AI data center and takes the 100% deduction in the year of placement in service is taking an uncertain tax position that may or may not be sustained on audit. Under ASC 740-10-25-6 through 25-7, an uncertain tax position must meet the more-likely-than-not recognition threshold (greater than 50% probability of being sustained on examination) before the tax benefit can be recognised. If the Section 168(n) position for an AI data center does not meet this threshold, the deduction cannot be recognised and a deferred tax asset must be established for the expected future deductibility of the asset under standard MACRS depreciation instead.
The tax provision documentation required: the technical analysis of whether the AI data center qualifies under Section 168(n), including the company's interpretation of the statute and the weight of the available evidence, must be documented in the tax provision workpapers before the Q3 estimated tax payment is made.
What Accounting Policy Decisions Must Your CFO Make Before Q3 Close and What SAB 74 Disclosures Does the FASB Project Require?
Five accounting policy decisions that must be documented before Q3 close for any company with a material AI data center investment.
Decision 1 (ASC 810): has the company performed a VIE analysis for each data center SPV in which it holds a variable interest? If so, what is the conclusion, and is the conclusion supported by an analysis of the power and economics tests under ASC 810-10-25-38A? The VIE analysis must be documented, reviewed by the external auditor, and reflected in the Q3 financial statements if consolidation is required.
Decision 2 (ASC 842): has the company assessed each capacity reservation arrangement for the identified asset and control of use tests? If the arrangement is classified as a lease, the right-of-use asset and lease liability must be reflected in the Q3 balance sheet. If classified as a service contract, the commitment must be disclosed in the commitments footnote.
Decision 3 (ASC 815): has the company assessed each long-term PPA for the derivative definition? If the NPNS exception applies, has it been formally documented and designated at the inception of each PPA? If derivative accounting is required, the fair value must be determined and recorded in Q3.
Decision 4 (ASC 360): has the company established a capitalisation policy that separately identifies the components of the AI data center (building, power infrastructure, GPU clusters, cooling systems, internal-use software) with appropriate useful lives for each? The component accounting policy must be documented and consistently applied.
Decision 5 (ASC 740): has the company assessed the Section 168(n) QPP position for any AI data center placed in service in 2026? If the position is taken, the more-likely-than-not threshold analysis must be documented in the tax provision.
SAB 74 disclosure: the FASB's April 6 research project, while in a pre-technical-agenda stage, may require SAB 74 disclosure for companies where the data center accounting questions are material. The disclosure should describe the project's scope, the current pre-guidance status, and the company's preliminary assessment of the potential impact on its accounting policies for AI data center investments.
What Does the SEC Comment Letter Record Show About Data Center Accounting Gaps?
The SEC's comment letter record on data center and cloud infrastructure accounting provides the best available signal of where the SEC staff currently focuses when reviewing financial statements with significant data center investments. Three recurring themes from the comment letter record:
Lease vs service contract classification: the SEC has asked companies to more specifically describe the basis for their classification of data center capacity arrangements, including the identified asset and control of use analysis. Generic statements that an arrangement "does not meet the definition of a lease" without describing the specific substitution rights analysis and the control of use assessment have generated follow-up comments.
VIE disclosure adequacy: the SEC has asked companies with data center investments through SPVs or joint ventures to describe the nature of their variable interests, the basis for their consolidation conclusions, and the financial statement impact of consolidation or non-consolidation. In several cases, the SEC has asked whether changes in the data center's economic circumstances during the period would affect the VIE analysis.
PPA commitment disclosures: the SEC has asked companies with long-term PPAs to disclose the aggregate amount of purchase commitments under those agreements by year for the next five years and thereafter, even when the PPA is classified as a service contract or when the NPNS exception is applied. The disclosure of PPA commitments in the commitments footnote is required regardless of the derivative accounting conclusion.
The SEC's AI-powered filing review capability (confirmed from Law.com, covered in the companion August 16 blog) will apply these same focus areas at scale across all public companies with material data center investments. A company whose disclosures do not address the lease classification basis, the VIE analysis, or the PPA commitment amounts is a statistical outlier relative to peers who have been asked to include this information and will generate a comment letter.
Frequently Asked Questions
What GAAP standards apply to AI data center accounting?
A typical AI data center arrangement simultaneously implicates five GAAP standards: ASC 810 (consolidation of data center SPVs as VIEs), ASC 842 (classification of capacity arrangements as leases or service contracts), ASC 815 (classification of power purchase agreements as derivatives), ASC 360 (capitalisation of data center development and equipment costs), and ASC 740 (tax accounting for the Section 168(n) qualified production property election). No single GAAP standard or authoritative guidance document addresses all five simultaneously.
Is an AI data center capacity arrangement a lease under ASC 842?
It depends on whether the arrangement involves an identified asset and gives the customer the right to control the use of that asset. If the data center operator has the substantive right to substitute different computing assets throughout the period of use (different GPUs, different server racks), no identified asset exists and the arrangement is a service contract, not a lease. If the customer has the right to use a specific, identified set of infrastructure assets and can direct how those assets are used, a lease likely exists. The analysis requires examination of the specific contractual terms and the economic substance of the substitution right.
Does a data center SPV require consolidation under ASC 810?
The consolidation analysis under ASC 810 requires a two-step assessment: (1) whether the SPV is a VIE (typically yes for project finance structures with thin equity relative to total capitalisation), and (2) whether the company is the primary beneficiary (has power over the most significant activities and has the obligation to absorb losses or the right to receive benefits). AI data center capacity reservation agreements that represent substantially all of the SPV's revenue may give the customer power over the SPV's most significant activities, which could make the customer the primary beneficiary required to consolidate.
Is a power purchase agreement for an AI data center a derivative under ASC 815?
A PPA meets the derivative definition if it has an underlying variable (electricity price), a notional amount (contracted volume), no significant initial net investment, and can be or requires net settlement. For PPAs where electricity is physically delivered and consumed in normal operations, the normal purchases and normal sales exception (ASC 815-10-15-22) may exclude the PPA from derivative accounting. The NPNS exception must be formally designated at the inception of each PPA. If the exception is not designated, the PPA must be measured at fair value through earnings each period.
What AI data center costs are capitalizable under ASC 360?
Clearly capitalisable: direct construction costs, external direct costs, and interest during construction. Judgment-intensive: GPU and accelerator costs (capitalisable, with shorter useful lives than the building requiring component accounting), custom cooling systems (capitalisable, with separate useful life assessment), and AI model development costs (governed by ASC 350-40, with development-phase costs capitalisable and preliminary project phase costs expensed).
Does Section 168(n) apply to an AI data center facility?
The IRS has not issued guidance specifically addressing whether AI data centers qualify as qualified production property under Section 168(n) of the OBBBA. The question turns on whether AI model training and inference constitute qualified production activities under the statute. If the Section 168(n) position is taken, it must meet the more-likely-than-not recognition threshold under ASC 740-10-25-6 for the tax benefit to be recognised in the financial statements.
Key Takeaways
- FASB Chair Rich Jones added a research project on data infrastructure investments on April 6, 2026. It is in the research phase, two stages before an exposure draft. No authoritative GAAP guidance on AI data center accounting is expected in 2026 or likely in 2027. Companies are making policy decisions now without guidance.
- Five GAAP standards apply simultaneously: ASC 810 (VIE consolidation), ASC 842 (lease vs service contract), ASC 815 (PPA derivative), ASC 360 (capitalisation), and ASC 740 (Section 168(n) QPP election). No authoritative standard or guidance document addresses the interaction among all five.
- ASC 810: data center SPVs with thin equity relative to project finance debt are typically VIEs. The primary beneficiary analysis focuses on whether the capacity customer's long-term reservation agreement gives it power over the SPV's most significant activities.
- ASC 842: the substitution right and control of use analyses are the critical tests. A capacity arrangement with a genuine substitution right (operator can and would substitute alternative assets) is a service contract; without a genuine substitution right, it is likely a lease requiring right-of-use asset and lease liability recognition.
- ASC 815: the PPA derivative analysis turns on the net settlement criterion. The NPNS exception, if formally designated at inception, prevents derivative accounting for PPAs where electricity is physically delivered in normal operations. Failure to designate NPNS at inception requires fair value accounting with income statement volatility.
- ASC 360: component accounting is required for AI data center equipment with useful lives shorter than the building (GPU clusters, cooling systems). The Section 168(n) QPP election for AI data centers requires an uncertain tax position analysis under ASC 740 because no IRS guidance confirms AI data centers qualify.
- Five Q3 policy decisions to document: VIE consolidation conclusion, lease vs service contract classification, NPNS designation for each PPA, component accounting policy with useful lives, and Section 168(n) uncertain tax position assessment.
- SAB 74 disclosure: companies with material AI data center investments should assess whether the FASB research project requires disclosure, even in its pre-technical-agenda stage, where the potential impact on accounting policies is material.
