## Guidance Note on Cloud Computing (IMF)

## Source details

**Canonical URL:** [Guidance Note on Cloud Computing (IMF)](https://www.imf.org/-/media/files/data/statistics/bmp7/implementation-support/cloud-computing-compilation-cg.pdf)

## Other formats

- [Markdown version](/-/media/files/data/statistics/bmp7/implementation-support/cloud-computing-compilation-cg.pdf.md)
- [Structured JSON version](/-/media/files/data/statistics/bmp7/implementation-support/cloud-computing-compilation-cg.pdf.json)

---

### Background and definition
- Cloud computing shifts computing activities from local devices to remote data centers accessed via networks, commonly referred to as “the cloud.”
- Cloud computing has replaced owned computing and communication equipment and software with purchased ICT services, and is used in the delivery and production of many digital services provided over the internet, including AI computing services.
- Cloud computing services are defined in GN DZ.8 Measurement of Cloud Computing in National Accounts as “computing, data storage, and related IT services accessed remotely over a network, supplied on demand and with measured resource usage.”
- The White Cover (Pre-Edited) Versions of the Integrated Balance of Payments and International Investment Position Manual, Seventh Edition (BPM7) and the 2025 SNA—both released in March 2025—introduced a digitalization chapter featuring cloud computing concepts discussed in the GN, and a globalization chapter covering MNE-related topics.
- Note on survey applicability: Applies to business with 10 employees or more (except Japan for which applies to business with 100 employees or more). Source: OECD

### Industry structure and economic implications
- Cloud computing is a global industry concentrated in a few large providers that are part of multinational enterprises (MNEs).
- Growth of intangible assets (data and intellectual property) complicates measurement and attribution because such assets can be moved across economies and held in different organizational units by MNEs.
- Cloud computing concentrates ICT investment patterns, shifting from broad diversification across many industries toward more concentrated business models.

### Principal categories of cloud computing services
- Infrastructure as a Service (IaaS): Virtualized computing resources such as servers, storage, and networking. Examples: Amazon Web Services (AWS) EC2, Microsoft Azure Virtual Machines.
- Platform as a Service (PaaS): Frameworks for developers to build, test, and deploy applications. Examples: Google App Engine, Microsoft Azure App Service.
- Function as a Service (FaaS) (subset of PaaS): Event-driven execution; examples: AWS Lambda, Google Cloud Functions, Azure Functions.
- Software as a Service (SaaS): Fully functional applications delivered over the web; examples: Gmail, Microsoft 365, Salesforce.
- Business Process as a Service (BPaaS): Automation and management of business processes using cloud platforms; examples: Automatic Data Processing (ADP) payroll, Workday, Salesforce Service Cloud.
- Additional use case: cloud mining — users rent computing power for crypto asset mining rather than operating hardware themselves.

### U.S. BEA satellite-account experience (summary)
- BEA definition: cloud services are computing services customers access from a shared pool of configurable computing resources in a flexible and on-demand way, without active management by the customer; primary NAICS industry: NAICS 518210 “data processing, hosting, and related services.”
- Data and methodology changes:
  - Initial estimates used Economic Census product revenue within NAICS 518210 but overestimated cloud services because product categories included unrelated internet and data products.
  - From 2013–forward, BEA used International Data Corporation (IDC) cloud services revenue data to estimate cloud services output.
  - For 2005–2012, values were back-casted from the 2013 value using growth rates from public financial filings for major cloud companies (AWS, Salesforce, Google, Microsoft, IBM).
- Key statistics:
  - cloud services represented 7.5 percent of digital economy GDP in 2022;
  - cloud services had the fastest growth in current-dollar value of any digital economy component between 2017 and 2022, with an annual average growth rate of 27.2 percent.
- BEA publication timing: initial estimates published August 2020; digital economy satellite account release noted December 2023.

### Balance of payments (BOP) recording and compilation guidance
- Recording principle: Cloud computing services are recorded as computer and information services in the BOP and are attributed to the location where they are used; they are typically consumed at the location where the related production process occurs.
- Distinguish billing entity vs. actual location of service delivery: resource pooling and dynamic workload allocation can obscure the production site, complicating determination of whether a transaction is domestic or cross-border and correct partner attribution.
- Compiler actions:
  - Attribute transactions to the location where the service is provided, using supplementary data sources (enterprise surveys, user console interfaces).
  - Consider regulatory requirements affecting geographic attribution (example: Lei Geral de Proteção de Dados (LGPD) in Brazil may require data centers within Brazil or contractual/technical safeguards for processing abroad).
- BPM7 and 2025 SNA guidance highlights:
  - Users with a long-term contract for dedicated access to a server in a cloud data center are economic owners of the server if operating risk is borne by the user, making the contract a financial lease (2025SNA 11.99-11.101; BPM7 16.16).
  - BPM7 does not distinguish treatment of licenses to use based on intended use duration or whether the licensee assumes risks and rewards of ownership (BPM7 11.100).
  - EBOPS 2026 framework improves alignment with national accounts by facilitating classification of transactions related to intellectual property products according to intended use.

### Cross-border transaction scenarios and recording approaches
- Case 1 — simple cross-border purchase:
  - Users in Country A purchase cloud services from a company based in Country B and pay service charges of 100.
  - In Country A, the payment is recorded as computer services in BOP.
- Case 2 — resident-subsidiary intermediary:
  - Users access services through a local subsidiary in Country A; payments made to the local subsidiary are resident-resident transactions and thus not recorded in BOP as imports; dividends from the subsidiary to the parent are recorded under investment income.
- Case 3 — complex MNE billing structure (numerical example):
  - Users in Country A pay 100 to the headquarters in Country C for computing services.
  - Headquarters transfers 60 to the subsidiary in Country B to cover data center operating expenses (implied remainder 40 as earned income/dividend flow).
  - Observed transactions: payment of 100 from Country A to Country C; payment of 60 from Country C to Country B.
  - Ideal (first-best) approach:
    - Record imputed payment of computing service charges from Country A to Country B (100) and imputed payment of dividends from Country B to Country C (40).
    - Ensure imputations are consistent with treatments in counterpart economies.
  - Practical (second-best) approach:
    - Record transactions based on actual payments when reliable information is insufficient to attribute service production location (i.e., record 100 from Country A to Country C and 60 from Country C to Country B).
  - Guidance:
    - Use imputation only when supported by reliable information; otherwise record observed payments.
    - Compilers are encouraged to collaborate with partner economies to ensure consistency.

### Compilation challenges and recommended tools
- Major challenges:
  - Identifying production location when billing entity differs from service provider location.
  - Attributing partner economy in cross-border transactions given MNE structures and intra-group arrangements.
  - Distinguishing reimbursements or cost-sharing transfers from service provision, and recognizing hidden dividends and investments embedded in intra-group arrangements.
- Recommended tools and data sources:
  - Enterprise surveys, user console interfaces, regulatory compliance information (e.g., data localization requirements), alternative commercial data sources (e.g., IDC), analysis of financial filings for major cloud firms.
  - Collaboration with counterpart compilers in partner economies to achieve symmetric treatments and improve attribution accuracy.

### Partner-economy attribution: first-best vs. second-best
- First-best approach: imputation to allocate provision of services to the country ultimately providing the service; faces “substantial operational challenges” and “could introduce asymmetries across countries.”
- Second-best approach: “considered the most practical method for partner economy attribution.”
- Considerations:
  - Attributing the partner economy to the headquarters is “conceptually appealing” because headquarters “typically bears the risks and rewards associated with the contract” but “diverges from the residency-based concept.”
  - Choice should consider conceptual rationale, data availability, and compilation constraints.
  - Compilers should "devote adequate resources to understanding the operational structures of cloud service providers" through direct engagement where feasible.
  - In practice, when unsure about service counterparts or symmetric treatment, “recording transactions according to actual payments is a practical solution.”

### SaaS, PaaS, IaaS, BPaaS and transaction attribution
- SaaS/PaaS providers often rely on supplementary cloud resources including IaaS; end users may lack details about underlying contracts and data processing locations.
- Guidance:
  - “For SaaS and PaaS transactions, the partner economy should be attributed to the direct party of the contract.” This guidance “may be revisited in the future” as business models evolve.
- BPaaS classification:
  - If BPaaS is “primarily focused on providing business advice, strategic guidance, or management support,” classify under “professional and management consulting services.”
  - If BPaaS “is centered on delivering automated computing solutions, leveraging cloud infrastructure to execute standardized tasks,” classify as “computing services.”

### European Union context and survey findings
- EU Single Market effects:
  - Companies may “establish their legal headquarters in any EU Member State,” often separate from physical data centers.
  - The invoicing country may not coincide with the country hosting the data centres, forcing compilers to “rely solely on financial flows to determine the counterpart country (second-best approach),” which can lead to “significant misallocation of current transactions.”
  - Even the first-best approach has “severe practical limitations” because enterprises “often” cannot determine the specific country where services are provided, despite GDPR provisions.
- Eurostat survey (September 2025) findings:
  - Eurostat “conducted a survey among European Union Member States” and “Approximately half of the reporting countries responded.”
  - Respondents highlighted “significant challenges in gathering data for most cross-border cloud computing activities.”
  - Countries able to collect some information “generally rely on direct surveys” with “limited use of administrative data sources.”
  - Limitations include differentiating cloud service types and contract maturities.

### FDI, data centers, and allocation of multi-territory enterprise operations
- MNE investments in cloud infrastructure are significant for BOP; major cloud providers are MNEs with operations in multiple countries connected through international cable networks.
- Example statistic: “Amazon Web Services (AWS) has 117 Availability Zones worldwide.”
- Investments abroad by cloud providers are recorded as FDI in BOP; treatment of MNEs should follow BPM7.
- For multi-territory enterprises where parent entities or distinct branches cannot be clearly identified:
  - “It is necessary to allocate the enterprise's total operations among the respective economic territories.”
  - Allocation should use available data best representing contributions to actual operations (examples: equity shares, equal allocations, or distributions based on operational factors such as tonnage or wages).
  - Where tax authorities have recognized a multi-territory structure, any established prorating formula "should serve as the basis for statistical reporting." (BPM7 4.13, 4.66, 4.68)

### Establishing, operating, hosting, and leasing of data centers
- Construction, equipping, and operation of data centers involve resident-nonresident transactions and must be measured accurately for BOP.
- Hosting and co-location services:
  - Included in computer services in BOP.
  - Exported when “foreign-owned IT assets” are hosted in a resident data center.
  - Imported when “locally owned ICT assets are hosted in a nonresident data center.” (BPM7 16.19)
- Leasing of computers without an operator is typically included in operating leasing.
- Leasing of data center buildings to cloud computing enterprises is typically included under financial lease “provided the lessee bears the operating risks.” (BPM7 11.105, 16.17)

### Intellectual property products (IPPs), software, and classification of investment
- Cloud computing has produced asset categories including “data assets, software, nonfungible tokens, and artificial intelligence (AI) systems,” involving significant IPPs and measurement complexities.
- IPP transactions within MNEs raise identification challenges, particularly “economic ownership.”
- IPP transactions outside MNE groups are generally recorded under “charges for the use of intellectual property n.i.e.” (BPM7 11.96).
- Hardware investment recording depends on asset ownership and risk:
  - If the remote provider “bears the operating risk of the IT equipment,” classify as cloud computing and the provider is the owner.
  - If the client “bears the operating risk,” classify as co-location rather than cloud services.
- Compilers should identify own-account investment in IT equipment; "dedicated survey questions could be used" and valuation should be "at sum of costs, including the costs of materials and the workers who design and build the equipment."
- Software classification and ISIC mapping:
  - ISIC class 582, Software publishing: companies primarily providing access to software they hold copyright for (may include some large SaaS providers).
  - ISIC class 631, Computing infrastructure, data processing, hosting and related activities: companies primarily providing other cloud computing services, including IaaS.
  - If a company primarily develops software, it is classed as a software publisher even if offered as SaaS; providing access to software developed by others is classified as cloud services.
- Software license capitalization rule:
  - "Software licenses of one year or more of use are classified as investment and capitalized."
  - "Licenses for use of less than one year are considered as intermediate consumption of software services." (2025 SNA 11.99-11.101)

### Data sources, surveys, and ITRS adjustments
- Valuable data items from global cloud providers include:
  - Payments within an MNE to each local establishment, including payments on behalf of the local establishment rerouted through that establishment.
  - Country-level breakdowns of total provision of cloud computing services.
  - Gross flows of cloud computing services to generate more precise estimates of international transactions.
- Surveys of MNEs should capture:
  - gross flows of cloud computing services by country,
  - intra-group transactions,
  - licensing arrangements,
  - geographic distribution of cloud service production.
- Preliminary user surveys are advisable before integrating cloud questions into existing surveys; recommended items include types of cloud services used, purposes, provider locations, value of cross-border payments, and instances where billing address differs from data center location.
- The ITRS often reflects the location of the billing entity rather than the actual service provider, which can misclassify partner economies in BOP statistics; compilers are encouraged to refine ITRS data using supplementary sources.

### Price and volume measurement guidance
- Pricing features:
  - IaaS and PaaS data storage/transfer usually priced per gigabyte with tiered discounts for high-volume users.
  - SaaS typically charged as subscription per user per month with volume/term discounts; usage-based charges per user session are less frequent.
  - IaaS and PaaS computing capacity is typically charged per hour, sometimes per second.
- Hourly fees vary based on time of use, server location, and amount of memory used per computing hour.
- Practical pricing approach:
  - Use the bill of a non-specified high-volume user in the reference period to set the services to be priced; in future periods, obtain current provider rates for the same specified services.

### Conclusions, recommendations, and institutional practices
- Cloud computing complicates international transaction recording due to dynamic resource allocation, intra-group transfers, absence of explicit cross-border payments, and tax-related obfuscation (IPPs relocation, transfer pricing).
- Ongoing challenges include identifying the location of actual computing or data storage; resource pooling and dynamic allocation make pinpointing service provision difficult.
- Recommended methodologies and practices:
  - Consider criteria or methodologies for allocating the economy where the service was provided using supplementary data sources, enterprise surveys, or technical metadata.
  - Document and disclose practices in metadata as part of data quality assessment frameworks; specify the approach employed (such as the first-best or second-best).
  - Improve data collection using a multi-faceted approach: advanced analytics, harmonized reporting standards, data from surveys, ITRS, contracts, network activity, and user interfaces.
  - Collaborate across jurisdictions, establish data sharing and data exchange arrangements with economies that have access to relevant information, and use Compilers Hub.
- Joint efforts aim to enhance the reliability and accuracy of measuring economic impact within national and international accounts.

### Questions for global consultation
- Do you agree with the first-best approach and the second-best approach illustrated in Figure 3?
- Do you agree with the treatment of transactions involving SaaS and PaaS?
- Do you agree with the guidance on price and volume measures of cloud computing as outlined in Section F?
- Do you have any further comments, or country experiences that could be shared to enrich the note?

*Prepared by Haruko Sakai, Venkat Josyula, and Andrew Baer (Statistics Department, IMF), with contributions from Matthias Ludwig, Robert Leisch, Felipe Correa, and Marco Tamburro (Eurostat); Jennifer Bruner (U.S. Bureau of Economic Analysis); and Tomoko Mori, and Akiko Kobashi (Bank of Japan).*

### 1.      Digitalization and the integration of digital technologies are changing work processes and

### Digitalization and the integration of digital technologies are changing work processes and 

### Background and definition
- Cloud computing shifts computing activities from local devices to remote data centers accessed via networks, commonly referred to as “the cloud.”
- Cloud computing has replaced owned computing and communication equipment and software with purchased ICT services, and is used in the delivery and production of many digital services provided over the internet, including AI computing services.
- Cloud computing services are defined in GN DZ.8 Measurement of Cloud Computing in National Accounts as “computing, data storage, and related IT services accessed remotely over a network, supplied on demand and with measured resource usage.”
- The White Cover (Pre-Edited) Versions of the Integrated Balance of Payments and International Investment Position Manual, Seventh Edition (BPM7) and the 2025 SNA—both released in March 2025—introduced a digitalization chapter featuring cloud computing concepts discussed in the GN, and a globalization chapter covering MNE-related topics.
- Note on survey applicability: Applies to business with 10 employees or more (except Japan for which applies to business with 100 employees or more). Source: OECD

### Industry structure and economic implications
- Cloud computing has become a global industry with widespread use across businesses, governments, and households; various entities acquire hardware and software services both internationally and domestically, relying on a few large cloud computing providers that are part of multinational enterprises (MNEs).
- Growth of intangible assets (data and intellectual property) as sources of value complicates measurement and attribution because such assets can be moved across economies and held in different organizational units by MNEs.
- Cloud computing concentrates ICT investment patterns, shifting from broad diversification across many industries toward more concentrated business models.

### Types of cloud computing services (principal categories)
- Infrastructure as a Service (IaaS): Virtualized computing resources such as servers, storage, and networking. Examples: Amazon Web Services (AWS) EC2, Microsoft Azure Virtual Machines.
- Platform as a Service (PaaS): Frameworks for developers to build, test, and deploy applications. Examples: Google App Engine, Microsoft Azure App Service.
- Function as a Service (FaaS) (subset of PaaS): Event-driven execution; examples: AWS Lambda, Google Cloud Functions, Azure Functions.
- Software as a Service (SaaS): Fully functional applications delivered over the web; examples: Gmail, Microsoft 365, Salesforce.
- Business Process as a Service (BPaaS): Automation and management of business processes using cloud platforms; examples: Automatic Data Processing (ADP) payroll, Workday, Salesforce Service Cloud.
- Additional use case: cloud mining — users rent computing power for crypto asset mining rather than operating hardware themselves.

### U.S. BEA satellite-account experience (box summary)
- BEA definition: cloud services are computing services customers access from a shared pool of configurable computing resources in a flexible and on-demand way, without active management by the customer; primary NAICS industry: NAICS 518210 “data processing, hosting, and related services.”
- Data sources and methodology changes:
  - Initial estimates used Economic Census product revenue within NAICS 518210 but overestimated cloud services because product categories included unrelated internet and data products.
  - From 2013–forward, BEA used International Data Corporation (IDC) cloud services revenue data to estimate cloud services output.
  - For 2005–2012, values were back-casted from the 2013 value using growth rates from public financial filings for major cloud companies (AWS, Salesforce, Google, Microsoft, IBM).
- Key statistics: cloud services represented 7.5 percent of digital economy GDP in 2022; cloud services had the fastest growth in current-dollar value of any digital economy component between 2017 and 2022, with an annual average growth rate of 27.2 percent.
- BEA publication timing: initial estimates published August 2020; digital economy satellite account release noted December 2023.

### Compilation guidance and balance of payments (BOP) recording
- Recording principle: Cloud computing services are recorded as computer and information services in the BOP and are attributed to the location where they are used; they are typically consumed at the location where the related production process occurs.
- Distinguish billing entity vs. actual location of service delivery: resource pooling and dynamic workload allocation can obscure the production site, complicating determination of whether a transaction is domestic or cross-border and correct partner attribution.
- Compilers should:
  - Attribute transactions to the location where the service is provided, using supplementary data sources (enterprise surveys, user console interfaces) to improve attribution accuracy.
  - Consider regulatory requirements that affect geographic attribution (example: Lei Geral de Proteção de Dados (LGPD) in Brazil imposes requirements on location and handling of personal data that may require data centers within Brazil or contractual/technical safeguards for processing abroad).
- BPM7 and 2025 SNA guidance:
  - Users with a long-term contract for dedicated access to a server in a cloud data center are considered economic owners of the server if operating risk is borne by the user, making the contract a financial lease (2025SNA 11.99-11.101; BPM7 16.16).
  - BPM7 does not distinguish treatment of licenses to use based on intended use duration or whether the licensee assumes risks and rewards of ownership (BPM7 11.100).
  - EBOPS 2026 framework improves alignment with national accounts by facilitating classification of transactions related to intellectual property products according to intended use.

### Cross-border transaction scenarios and recording approaches
- Illustrative simple case (Case 1):
  - Users in Country A purchase cloud services from a company based in Country B and pay service charges of 100.
  - In Country A, the payment is recorded as computer services in BOP.
- Resident-subsidiary case (Case 2):
  - Users access services through a local subsidiary in Country A; payments made to the local subsidiary are resident-resident transactions and thus not recorded in BOP as imports; dividends from the subsidiary to the parent are recorded under investment income.
- Complex MNE billing structure (Case 3) — numerical example and recording options:
  - Users in Country A pay 100 to the headquarters in Country C for computing services.
  - Headquarters transfers 60 to the subsidiary in Country B to cover data center operating expenses (implied remainder 40 as earned income/dividend flow).
  - Observed transactions: payment of 100 from Country A to Country C; payment of 60 from Country C to Country B.
  - Ideal (first-best) approach:
    - Record imputed payment of computing service charges from Country A to Country B (100) and imputed payment of dividends from Country B to Country C (40).
    - Ensure imputations are consistent with treatments in counterpart economies.
  - Practical (second-best) approach:
    - Record transactions based on actual payments when reliable information is insufficient to attribute service production location (i.e., record 100 from Country A to Country C and 60 from Country C to Country B).
  - Guidance:
    - Imputation for the ideal approach should only be adopted when supported by reliable information; where evidence is insufficient, record observed payments as a practical basis.
    - Compilers are encouraged to collaborate with partner economies to ensure consistency and accuracy.

### Compilation challenges and recommended tools
- Major challenges:
  - Identifying production location when billing entity differs from service provider location.
  - Attributing partner economy in cross-border transactions given MNE structures and intra-group arrangements.
  - Distinguishing reimbursements or cost-sharing transfers from service provision, and recognizing hidden dividends and investments embedded in intra-group arrangements.
- Recommended tools and data sources:
  - Enterprise surveys, user console interfaces, regulatory compliance information (e.g., data localization requirements), alternative commercial data sources (e.g., IDC), and analysis of financial filings for major cloud firms.
  - Collaboration with counterpart compilers in partner economies to achieve symmetric treatments and improve attribution accuracy.

*Prepared by Haruko Sakai, Venkat Josyula, and Andrew Baer (Statistics Department, IMF), with contributions from Matthias Ludwig, Robert Leisch, Felipe Correa, and Marco Tamburro (Eurostat); Jennifer Bruner (U.S. Bureau of Economic Analysis); and Tomoko Mori, and Akiko Kobashi (Bank of Japan).*

### 17.      It is agreed that the second-best approach is the most practical method for partner

### 17.      It is agreed that the second-best approach is the most practical method for partner economy attribution.

### Partner-economy attribution: approaches and guidance
- The "first-best" approach involves imputation and allocating provision of services to the country ultimately providing the service, but:
  - It faces "substantial operational challenges" and "could introduce asymmetries across countries."
  - The physical location of data processing "often changes and cannot be easily determined, even by the service providers themselves."
- The "second-best" approach is "considered the most practical method for partner economy attribution."
- Attributing the partner economy to the headquarters:
  - Is "conceptually appealing" because headquarters "typically bears the risks and rewards associated with the contract."
  - "Diverges from the residency-based concept."
- Choice between approaches should consider conceptual rationale, data availability, and compilation constraints.
- Compilers are encouraged to "devote adequate resources to understanding the operational structures of cloud service providers, including through direct engagement where feasible, to support informed and consistent application of the guidance."
- In practice, when unsure about service counterparts or symmetric treatment between economies, "recording transactions according to actual payments is a practical solution."

### Billing entity, headquarters, and illustrative cases
- The billing entity may differ from the headquarters; e.g., "several billing centers based in Luxembourg handle invoicing for cloud services, even though they do not have cloud-related IPPs or company headquarters there."
- Appendix II provides further elaboration and illustrative cases on how different business structures affect partner-economy attribution.
- In certain contexts—such as within the European Union—"enhanced data availability and coordination may facilitate the application of the first-best approach, although practical challenges remain."

### SaaS, PaaS, IaaS and transaction attribution
- Transactions involving SaaS and PaaS present additional complexities because:
  - SaaS/PaaS providers frequently rely on supplementary cloud computing resources, including IaaS.
  - Final users "may not have access to details regarding contracts between the SaaS/PaaS providers and the IaaS providers," especially about data processing and storage locations.
- Example: ERP solutions delivered as SaaS may rely on underlying infrastructure provided by third-party IaaS providers.
- When end users contract with a direct party (e.g., a foreign SaaS/PaaS provider), "attributing the counterpart economy of the transaction to the direct party of the contract is a standard and effective practice."
- Guidance: "For SaaS and PaaS transactions, the partner economy should be attributed to the direct party of the contract." This guidance "may be revisited in the future" given evolving business models.

### Complex transaction structures (BOX2) — illustrative diagram figures and amounts
- Practical complexities:
  - Resident users may use data centers both within and outside the country; both residents and nonresidents may utilize data centers within the country.
  - Transactions of cloud computing service charges can occur within a cloud computing enterprise group.
  - When unsure, "recording transactions according to actual payments is a practical solution."
- Diagram annotations (preserve exact figures as presented):
  - Computing Service Charges 100 + α
  - Operating expenses of data centers 60
  - Computing Service Charges 100

### BPaaS classification
- BPaaS can range from "purely digital automation" to "business process outsourcing" involving significant expert input.
- Classification guidance:
  - If BPaaS is "primarily focused on providing business advice, strategic guidance, or management support," classify under "professional and management consulting services."
  - If BPaaS "is centered on delivering automated computing solutions, leveraging cloud infrastructure to execute standardized tasks," classify as "computing services."
- Compilers must remain vigilant distinguishing computer services from other service categories as cloud offerings diversify and boundaries shift.

### European Union context (BOX3): data, single market, and survey findings
- The EU Single Market allows companies to provide services across Member States and "establish their legal headquarters in any EU Member State," often in locations different from physical data centers.
- Consequence: "The invoicing country where the cloud computing enterprise headquarter is located may not coincide with the country hosting the data centres," forcing BOP compilers to "rely solely on financial flows to determine the counterpart country (second-best approach)," which can lead to "significant misallocation of current transactions."
- Even the first-best approach has "severe practical limitations" because enterprises "often" cannot determine the specific country where services are provided, despite GDPR provisions.
- Eurostat survey (September 2025):
  - Eurostat "conducted a survey among European Union Member States to assess the current capabilities of reporting countries in compiling data on cloud computing activities."
  - "Approximately half of the reporting countries responded."
  - Respondents highlighted "significant challenges in gathering data for most cross-border cloud computing activities."
  - Countries able to collect some information "generally rely on direct surveys" (e.g., International Trade in Services Survey) with "limited use of administrative data sources such as business registers, tax records, and payment data."
  - Respondents noted limitations in current surveys that "may potentially be addressed with the introduction of the cloud computing topic in the new Digitalization chapter in the BPM7 manual."
  - Difficulties include differentiating between cloud computing service types and contract maturities (short-term vs. long-term).

### Foreign Direct Investment (FDI), data centers, and recording in BOP
- MNEs' foreign direct investment in cloud infrastructure is "significant for BOP."
- Major cloud providers are MNEs with clients and operations in multiple countries, enabling cross-border data flows and establishing local data centers connected through international cable networks, including undersea cables.
- Example statistic preserved: "Amazon Web Services (AWS) has 117 Availability Zones worldwide."
  - Footnote definition: "Availability Zone in AWS is one or more discrete data centers with redundant power, networking, and connectivity in an AWS Region."
- Investments abroad by cloud providers are recorded as FDI in BOP; treatment of MNEs should follow BPM7.
- For multi-territory enterprises with integrated operations across multiple economic territories, if parent entities or distinct branches cannot be clearly identified:
  - "It is necessary to allocate the enterprise's total operations among the respective economic territories."
  - Allocation should use available data best representing contributions to actual operations (examples: equity shares, equal allocations, or distributions based on operational factors such as tonnage or wages).
  - Where tax authorities have recognized a multi-territory structure, any established prorating formula "should serve as the basis for statistical reporting." (BPM7 4.13, 4.66, 4.68)

### Establishing, operating, hosting, and leasing of data centers
- Establishment and operation of data centers are recorded in the BOP "if it involves resident-nonresident transactions."
- Ongoing operation (capital expenditures, maintenance, technological upgrades) is relevant for both initial investment and subsequent production of services.
- Data center legal/economic/organizational forms include wholly owned local subsidiaries, joint ventures, or contracts through specialized local partners.
- Cross-border transactions related to data center construction, leasing, and equipping must be measured accurately.
- Hosting and co-location services:
  - Are included in computer services in BOP.
  - Exported when "foreign-owned IT assets" are hosted in a resident data center.
  - Imported when "locally owned ICT assets are hosted in a nonresident data center." (BPM7 16.19)
- Leasing of computers without an operator is typically included in operating leasing.
- Leasing of data center buildings to cloud computing enterprises is typically included under financial lease "provided the lessee bears the operating risks." (BPM7 11.105, 16.17)

### Intellectual property products (IPPs), software, and classification of investment
- Cloud computing has produced new asset categories: "data assets, software, nonfungible tokens, and artificial intelligence (AI) systems," involving significant IPPs and measurement complexities.
- Transactions within MNEs involving IPPs raise identification challenges, particularly "economic ownership."
- BPM7 and the 2025 SNA include a dedicated chapter on globalization addressing MNE structures and IIP-related statistical challenges (BPM7 15.13–27, 15.33–37).
- IPP transactions outside MNE groups are "generally identifiable through market transactions" and are often recorded under "charges for the use of intellectual property n.i.e." (BPM7 11.96).
- Recording hardware investment depends on asset ownership and risk:
  - If the remote provider "bears the operating risk of the IT equipment," classify as cloud computing and the provider is the owner.
  - If the client "bears the operating risk," classify as co-location rather than cloud services.
- Compilers should identify own-account investment in IT equipment by cloud providers; "dedicated survey questions could be used" and valuation should be "at sum of costs, including the costs of materials and the workers who design and build the equipment."
- Software classification and ISIC mapping:
  - Companies that primarily provide access to software they hold copyright for are classed in ISIC class 582, Software publishing (may include some large SaaS providers).
  - Companies that primarily provide other cloud computing services, including IaaS and related services, are classed in ISIC class 631, Computing infrastructure, data processing, hosting and related activities.
  - If a company primarily develops software, it is classified as a software publisher even if offered as SaaS; providing access to software developed by others is classified as cloud services.
- Software license capitalization rule:
  - "Software licenses of one year or more of use are classified as investment and capitalized."
  - "Licenses for use of less than one year are considered as intermediate consumption of software services." (2025 SNA 11.99-11.101)

### Data sources and surveys
- Information from global cloud computing service providers can significantly improve understanding of exports and imports of these services.
- Valuable data items include:
  - Payments within an MNE to each local establishment, including payments on behalf of the local establishment rerouted through that establishment.
  - Country-level breakdowns of total provision of cloud computing services.
  - If gross flows of cloud computing services are available, it is possible to generate more precise estimates of international transactions in these services.

*Source: BPM7 and related cloud computing compilation guidance and the 2025 SNA.*

### 40.      Surveys of MNEs can improve measurement of cloud service exports and imports. Surveys

### Surveys of MNEs can improve measurement of cloud service exports and imports. Surveys

### Surveys of MNEs: purpose and design
- Targeted surveys of multinational enterprises (MNEs) should capture detailed information on:
  - gross flows of cloud computing services by country,
  - intra-group transactions,
  - licensing arrangements,
  - geographic distribution of cloud service production.
- International cooperation can assist national statistics offices in developing consistent estimates. The UNECE Guide to Sharing Economic Data in Official Statistics discusses international data sharing for enhancing measurement of MNEs and international transactions.
- Ongoing cooperation between national accounts specialists and balance of payments statisticians is necessary to develop further guidelines for data sharing.

### Surveys on cloud computing services users and complementary data sources
- Preliminary user surveys are advisable to obtain a general overview of cloud-related transactions before integrating questions into existing surveys (trade in services surveys, enterprise surveys).
- Recommended preliminary survey items include:
  - types of cloud services utilized (IaaS, PaaS, SaaS, BPaaS),
  - purposes of usage (production, administration, or both),
  - geographical location of cloud service providers,
  - value of cross-border payments related to cloud services,
  - instances where the billing address differs from the data center location.
- Sample preliminary survey form is provided in Appendix I (sections and questions preserved as in the Guidance Note).
- Alternative and supplementary data sources:
  - user-level analytics platforms (web traffic, application usage, digital transaction data),
  - these can capture cross-border usage patterns by households and smaller users not covered in enterprise surveys,
  - while partner-economy attribution may be incomplete, such sources enhance coverage for many small transactions.

### International Transactions Reporting System (ITRS) and billing vs. service location
- The ITRS often reflects the location of the billing entity rather than the actual service provider, which can misclassify partner economies in BOP statistics.
- Example scenario: Company in Country A pays an annual fee to a cloud service provider headquartered in Country C, but the data center delivering the service is in Country B; ITRS would record Country C unless adjusted.
- The ideal recording is the imputed payment from Country A to Country B (refer to Case 3 in the Guidance Note).
- Compilers are encouraged to refine ITRS data using supplementary sources, enterprise surveys, and collaboration with partner economies.

### Other and emerging data sources
- Other useful sources include administrative records, financial disclosures, industry reports, technical metadata, regulatory filings, tax documents, commercial datasets, company financial statements, and licensing contracts.
- Metadata from user interfaces and network logs can support attribution of service delivery when billing entities differ from actual providers.
- Research on quantifying investments in servers and software hosted in foreign locations by users of hosting services is recommended.
- Emerging sources: blockchain-based ledgers, IoT-generated data (fitness trackers, smart home appliances, industrial sensors), third-party data providers and industry intelligence firms—useful as complementary inputs for estimation and high-frequency transactions.

### Box4 — Measuring cloud computing services in U.S. International Economic Accounts (summary)
- BEA collected cloud computing information via:
  - Activities of Multinational Enterprises (AMNE) surveys (2019 and 2024 Benchmark Surveys of United States Direct Investment Abroad; 2022 Benchmark Survey of Foreign Direct Investment in the United States),
  - questions requested values for sales/gross operating revenues, costs and expenses (excluding employee compensation), and number of employees engaged in providing cloud computing services.
- Defined cloud computing services as: "Computing services that customers can access from a shared pool of configurable computing resources in a flexible and on-demand way, without active management by the customer..."
- Observed challenges in special data collection on direct investment surveys:
  - low response rate on cloud questions,
  - difficulty communicating the concept of cloud computing to respondents; some firms misreported internal IT cloud migration as cloud service sales.
- BEA collects cloud computing together with data storage under 'computer services' in quarterly International Transactions Accounts; detailed annual statistics for cloud computing and data storage begin with data for 2006.
- Market research data (e.g., IDC, Gartner) have been used to refine estimates of cloud services output.

### Compilation challenges — identification of counterpart economy and MNE structures
- Key difficulties in identifying the location of service provision due to:
  - resource pooling and global distribution of data centers,
  - dynamic shifting of workloads across jurisdictions,
  - redundant data storage for reliability and continuity.
- Internal MNE operations complicate measurement:
  - software updates developed at headquarters may be deployed globally as part of SaaS offerings,
  - unpriced cross-border data flows from workload/data movement between affiliates create gaps in recorded economic value of intra-group transactions.
- Tax-related factors:
  - MNEs in sectors with substantial IPPs may allocate production to low-tax jurisdictions by relocating IPPs and manipulating transfer pricing; cloud computing is notably impacted by these practices.
  - Baer et al. (2020) indicate that Ireland and the Netherlands—representing only 2 percent of the OECD countries’ GDP—were responsible for 53 percent of OECD countries’ exports of computer services in 2016.
- Multi-layered structure (IaaS, PaaS, SaaS, BPaaS) complicates tracing origin/destination, as different layers may involve different providers, locations, and contracts.
- Classification needs:
  - Clarification required on management and support services (e.g., BPaaS) and whether BPaaS should be categorized as professional and management consulting services enabled by IT resources.
  - Segmenting cloud computing services to identify BPaaS would enhance understanding.

### Price and volume measurement challenges and practical guidance
- Pricing features vary by cloud service:
  - IaaS and PaaS data storage/transfer usually priced per gigabyte with tiered discounts for high-volume users.
  - SaaS typically charged as subscription per user per month with volume/term discounts; usage-based charges per user session are less frequent.
  - IaaS and PaaS computing capacity is typically charged per hour, sometimes per second.
- Hourly fees vary based on:
  - the time when computing is used (higher fees for higher demand; discounts for pre-purchasing time),
  - the location of the server (busier servers charge higher rates; constraints from privacy laws),
  - the amount of memory used per computing hour.
- Practical pricing approach:
  - Use the bill of a non-specified high-volume user in the reference period to set the services to be priced; in future periods, obtain current provider rates for the same specified services.

### Conclusions and way forward — recommendations and institutional practices
- Cloud computing complicates international transaction recording due to dynamic resource allocation, intra-group transfers, absence of explicit cross-border payments, and tax-related obfuscation (IPPs relocation, transfer pricing).
- Ongoing challenges include identifying the location of actual computing or data storage; resource pooling and dynamic allocation make pinpointing service provision difficult.
- Suggested methodologies and practices:
  - Consider criteria or methodologies for allocating the economy where the service was provided using supplementary data sources, enterprise surveys, or technical metadata.
  - Countries should document and disclose practices in metadata as part of data quality assessment frameworks; specify the approach employed (such as the first-best or second-best).
  - Improved data collection and a multi-faceted approach are necessary: advanced analytics, harmonized reporting standards, data from surveys, ITRS, contracts, network activity, and user interfaces to better reveal resource use and allocation.
  - Collaboration across jurisdictions, data sharing and data exchange arrangements with economies that have access to relevant information, and use of Compilers Hub are essential.
- The joint efforts enhance the reliability and accuracy of measuring economic impact within national and international accounts.

### Questions posed for global consultation
- Do you agree with the first-best approach and the second-best approach illustrated in Figure 3?
- Do you agree with the treatment of transactions involving SaaS and PaaS?
- Do you agree with the guidance on price and volume measures of cloud computing as outlined in Section F?
- Do you have any further comments, or country experiences that could be shared to enrich the note?

---


_Source: https://www.imf.org/-/media/files/data/statistics/bmp7/implementation-support/cloud-computing-compilation-cg.pdf_