On Dec. 5, 2023, S&P Global Ratings published a request for comment (RFC) on its proposed global framework for rating data center securitizations. Following feedback from the market, we finalized and published our criteria, titled "Data Center Securitizations: Global Methodology And Assumptions," on June 13, 2024.

We'd like to thank investors, issuers, and other intermediaries who provided feedback. This RFC process summary provides an overview of the changes between the RFC and the final criteria, and the rationale behind those changes.

External Written Comments Received From Market Participants That Led To Significant Analytical Changes To The Final Criteria

Significant-tenant lag

Comment:  A market participant asked about our rationale for excluding optional contractual renewals from the single-tenant lag assumption. In addition, multiple market participants inquired about how the single-tenant lag period is defined given cloud providers are unlikely to move due to their large-scale investments, moving large deployments is challenging, and there are currently limitations on supply and power. Moreover, multiple market participants commented that when determining the appropriate lag, the approach should consider lease termination notification periods as well as the location, quality, and homogeneity of the data center. Another market participant commented that the single-tenant lag stress should be applied on the data-center level and not on the building.

Response:  Our methodology accounts for the contractual nature of the in-place lease contracts. Tenants are not required to renew. Instead, such renewal clauses provide tenants future optionality and price protection against significant lease rate increases. The methodology looks to account for the risk of vacancy, which would be more acute for data centers occupied by a significant tenant.

We developed the timeframes in table 4 of the criteria using steady-state performance indicators to determine stress cases for different economic environments. However, we acknowledged the potential differing impact based on the size and the location of the facility, and we created bands around the lag time at each rating category. At the same time, when determining whether to apply the lag stress, we've updated the approach to account for contractual notification periods and staggering lease maturities. In addition, we've expanded its application beyond single tenants to capture significant tenant exposure, whereby a single tenant may not occupy the entirety of a property but represents the overwhelming majority of the exposure. Regarding asset quality and homogeneity, the asset quality or similarity of a property type doesn't imply that a new tenant could seamlessly replace the in-place tenant without downtime.

The framework applies the significant tenant stress by specific building to represent the idiosyncratic risk associated with re-letting a given property. Also, buildings are typically legally ring-fenced from one another and, consequently, can be disposed of separately.

Renewals on property leases

Comment:  A couple of market participants commented that when customer contracts extend beyond the life of the property lease, our analysis should anticipate future contractual extensions. One of the same market participants believed we should include recovery credit for the operations' mechanical and electrical equipment at the end of the property lease term.

Response:  We've updated the methodology to typically provide credit to property leases through the time horizon associated with the utility of the property, subject to there being lease extension options within the property lease contract. This anticipates the renewal of a property lease when a contract matures within such a timeframe. With respect to mechanical and electrical equipment, there isn't a longstanding and robust secondary market for such assets. Furthermore, depreciation of such assets may be significant over the course of a transaction's life.

Ground leases

Comment:  A market participant questioned the limitation on exposure to ground leases given that a SASB CMBS transaction backed by a single property (100% of the portfolio) with exposure to a ground lease can be rated, assuming the ground lease extends beyond the legal final maturity of the financing.

Response:  We've updated the scope to remove the limitation on exposures to ground leases where the corresponding data centers are owned by the transaction. We've also clarified that the scope includes transactions with minimal exposure to land ownership where the corresponding data center is not owned.

Redundancy sub-score

Comment:  Multiple market participants commented that the calibration of the redundancy sub-score doesn't represent the current data center landscape in the U.S. and Canada. This is because facilities have increasingly moved toward handling network redundancy by way of non-physical solutions, including availability zones and software. As a result, many tenants believe it's not cost effective to pay for 2N because N+1 redundancy has become the typical standard for new construction. Another market participant stated that facilities that exhibit redundancy above the market standard are at risk of lower valuations and increased liquidation timing given the additional maintenance required. Moreover, multiple market participants asserted that downtime is a combination of equipment and operational excellence and that performance against contractual commitments is a better indicator of future performance.

Response:  We've updated the implementation of the redundancy sub-score such that a score of '5' no longer represents 2N. Rather, it addresses a facility with concurrently maintainable data center infrastructure, dual power feed to the racks, and N+1 or above redundancy on critical data center infrastructure components, including power generation, UPS, and cooling. In addition, we have removed references to uptime institute scores.

With respect to property valuations, our property underwriting reflects estimates for lease revenues and expenses associated with various redundancy levels, including maintenance capital expenditures. Given that the property valuation would account for the higher operating expense ratio associated with increased redundancy, we don't adjust our liquidation timing assumptions.

Regarding downtime performance, our operational risk assessment framework considers the manager's experience and historical performance in determining the maximum rating on the transaction.

Weighting of sub-scores in U.S. and Canada

Comment:  A market participant commented that the age sub-score should carry less weight than the power efficiency sub-score in the determination of the property-level utility score. Similarly, a market participant commented that these attributes are overweighted because a newer property will most likely have a higher power efficiency score. Another market participant didn't believe the location score carried sufficient weight for the top locations. Someone else commented that the cost of power sub-score appears overweighted given that there are top-tier markets with high power costs.

Response:  The methodology focuses on the competitive profile of each property given the shorter-term nature of the lease contracts compared with the maturity profile of the liabilities. In general, we believe a data center's competitive position is reflected in the location, age, cost of power, redundancy, and power efficiency. The weighting of these attributes helps rank-order the potential impact from each component.

We consider the age of the data center to be from the latest major renovation date. The age of a building is correlated with the age of the power and cooling equipment, uptime performance, designed and operating power efficiency, and other elements that help to inform the data center's competitive position. Over time, core mechanical and electrical components become less reliable, more expensive to maintain, and significantly riskier. Assuming similar pricing and location, new properties are expected to be more attractive and competitive than older ones. Overall, these elements inform our weighting for the age of the build. In addition, the power efficiency score allows us to differentiate facilities by a factor beyond just the age of the building. While newer facilities will likely have improved power usage effectiveness (PUE) ratios, different facilities may have different efficiency levels. At the same time, upgrades might not target power efficiency.

With respect to the location score, we've increased the percentage weighting for the U.S. and Canada to recognize that latency and network effects are paramount in determining the long-term competitive nature of a property. In addition, the location is reflected in the S&P Global Ratings value through assumptions such as vacancy and cap rates.

Regarding the cost-of-power sub-score, there are top-tier markets with both high and low power costs. Data centers with lower power costs are expected to have a competitive advantage over the longer term and would likely be less susceptible to price competition from facilities that can offer similar levels of latency. Nevertheless, we've lowered the weighting of the cost-of-power score to reflect the significant role of location in determining a property's long-term competitive profile.

External Written Comments Received From Market Participants That Did Not Lead To Significant Analytical Changes To The Final Criteria

Additional U.S. and Canadian markets and jurisdictions outside of North America

Comment:  Multiple market participants asked how the criteria would be applied to additional markets within the U.S. and Canada and to jurisdictions outside the U.S. and Canada given that Appendix 3 only includes our benchmark assumptions and weightings for the five standard utility attributes within the U.S. and Canada.

Response:  The criteria scope is global and can be applied to additional markets in the U.S. and Canada and new jurisdictions without updating Appendix 3 to include such benchmark assumptions and associated weightings. Nevertheless, we've clarified that as asset pools expand, Appendix 3 will be updated for new markets and jurisdictions.

Self-insurance

Comment:  A market participant inquired about how the framework would address self-insurance in relation to property insurance.

Response:  We have clarified in the insurance section that in limited cases, self-insurance will be considered for highly rated tenants when there is evidence of structural mitigants to negative ratings migration as well as consequences for failure to replace.

Diversity and concentration

Comment:  Multiple market participants asked about the definition of "significant tenant concentration," while another market participant noted that many references to diversity and concentration are not defined.

Response:  We've included references to property count when describing property diversity and concentration. With respect to tenant diversity, we've clarified that this relates to the diversity of the in-place tenants. We've also clarified "significant tenant concentration," which looks to capture event risk related to pools that are significantly exposed to a low number of tenants.

Disposition periods

Comment:  Multiple market participants commented that the disposition period of 36 months prior to the earliest legal final maturity of a series of notes within the trust for an 'A' liability appears to be punitive when considering recent market trends. In addition, multiple market participants commented that the disposition periods are unnecessary given that the cumulative stress applied to the property value when compared to current appraisals would classify as a distressed sale. Another market participant asserted that having the disposition periods prior to the maturity of the shortest-dated liability doesn't account for the utility of the property.

Response:  The current data center landscape is an expansionary environment with limited supply and high demand, which supports market dispositions of under a year. Nevertheless, we believe that such an environment is unlikely to persist throughout the long maturity profiles associated with ABS data center transactions. In addition, we haven't yet seen a data center downturn with waning demand and excess supply that would be representative of an 'A' economic environment. The disposition periods reflect stressed timeframes that we believe would be representative of liquidation timeframes during stressed economic periods, which would likely be accompanied by severe property-value declines. In respect to the legal final maturity, the approach accounts for liquidation proceeds being needed to repay the debt on these transactions. As a result, the disposition period is imposed prior to the earliest legal final maturity. At the same time, we've clarified that we consider the maturity profile of outstanding liabilities when determining additional stresses to account for transaction disposition periods that are shorter than our estimated disposition periods.

Forward starting leases

Comment:  A market participant commented that credit for forward-starting leases should not be capped given they represent executed agreements. Multiple market participants commented that when a tenant has reached its ready-for-service date for an existing phase, it's very unlikely that the tenant won't occupy additional phases. This is because it has already commenced operations at the site, and the additional capex required for fit-out is accounted for in the appraised value.

Response:  The proposed thresholds align with the typical credit provided to forward starting leases within our liquidation values for our CMBS approach while considering the nature of the exposure and relevant mitigants. The longer the exposure period, and the lower the proportion of the facility that is fully complete, the less credit given to future leases to the extent that there are no structural mitigants. We've clarified in the framework that we also consider the credit strength of the tenants when assessing the nature of the exposure. In addition, such forward-starting leases are not risk free, as they often include tenant optionality based upon completion timelines, which are often subject to delay and cost overruns. Moreover, there is the potential for lease modifications based on completion delays and possible performance obligations on the part of the utility provider. At the same time, the capex is generally not included as part of the collateral of the transaction. For the appraised value, the as-is value doesn't typically include credit for future phases.

Base utilization stress

Comment:  Multiple market participants commented that the base utilization stress for hyperscale facilities is high given the contractual nature of the leases, the mission-critical nature of the data centers to the in-place tenants, and the large levels of invested capital on the part of tenants. Another market participant commented that the base utilization should account for historical churn, which would indicate that data center performance is partly insulated from recessionary risks, as evidenced by performance during the COVID-19 shutdown. In addition, the same market participant commented that the approach doesn't account for improvements in utilization over time. Another market participant asked how we would differentiate the utilization stress when a lease contract is predicated on leased capacity versus utilized capacity. Separately, a market participant asked how the utilization approach captures tenant diversification, as single-tenant facilities can have strong credit but substantial churn risk related to the risk associated with business applications reaching end of life, business consolidation through cost-cutting measures or M&A, service-level agreement violations, etc.

Response:  The base utilization stress accounts for fluctuations in delinquency and occupancy over time and is tiered by rating category to address the impact of economic conditions. We would set a base utilization stress that reflects historical performance and the outlook for the relevant sectors while considering tenant diversity and lease tenor. We would then apply rating-specific multipliers to represent the increase in cash flow volatility that we expect higher-rated liabilities to be able to withstand. We determine base utilization at the property level and account for potential migration in tenant quality and concentration over time. For example, the framework recognizes that single-tenant properties with long-term leases to investment-grade tenants are expected to exhibit lower cash flow volatility based on contractual performance. A 3%-9% stressed rate is representative of an 'A' category stress for a wholesale facility and not the current expansionary market conditions of limited supply and excess demand, which we would characterize at a stress level below 'B'. In addition, the COVID-19 period would not be considered a stressed environment for the data center industry and wouldn't be consistent with a stress level above 'B'.

Regarding evolving utilization rates, the framework provides limited credit to forward-starting leases under our revenue assumptions but doesn't account for improvements in utilization over time. When contracts are usage based, we've clarified under Step 3 that we'll consider historical utilization while accounting for contractual minimums and the transaction's exposure to such leases to determine our revenue assumptions. In addition, we've clarified in the base utilization section that we would likely assume a higher utilization stress to recognize the potential for increased volatility in lease payments over time.

In terms of concentration risk, the methodology acknowledges that single-tenant properties will likely have a lower base utilization stress but are subjected to a renewal lag to capture such churn risk. More diversified properties would have a higher base utilization stress but wouldn't experience lag.

Location sub-score

Comment:  A market participant suggested that location shouldn't be used as a sub-score in the determination of utility of a hyperscale data center, given the level of investment on the part of the lessee when establishing an availability zone. In addition, they commented that a single-tenanted facility with a large and diversified end-user base should benefit from a positive location score in the same way that a facility in a tertiary-market may benefit under the framework from strong demand and diversified tenants. A similar comment was raised by multiple market participants, who asserted that the importance of location is increasingly driven by proximity to individual end users. Another participant commented that facilities in moderate to weak markets with long-term leases or large campuses, strong operating performance, and network superiority should carry higher location sub-scores.

In addition, a market participant commented on the correlation between the cost-of-power and location sub-scores. Separately, market participants emphasized that the location sub-score should account for latency, total cost of operations, future development, the presence of cloud providers, and utilization and absorption rates. Moreover, multiple market participants asked how these sub-scores will evolve over time while asserting that the proposed calibration of location scores didn't account for the growth in the sector and the expansion outside of key markets.

Response:  The framework assigns a competitive risk ranking to each property within a given portfolio, which indicates the property's expected liquidation profile based upon potential exposure to sector evolution and obsolescence. This rank ordering doesn't account for the role of the in-place tenant given that the contractual lease terms tend to be significantly shorter than the stated liability maturity profile.

We don't account for contractual renewals in our analysis. Instead, we consider the attractiveness of a given property to the market over time based on characteristics such as location. Proximity to the end-user base--as measured through latency--is a key attribute of a location sub-score. Facilities that offer both low latency to end users and form part of a large data center market are expected to remain competitive over the long term and receive higher location scores.

To recognize that latency and network effects are paramount in the determination of the long-term competitive nature of a property, we've increased the weighting for the location sub-score in the U.S. and Canada. At the same time, our property-level analysis could lead us to assign an attribute score that differs from the benchmark assumption. For example, our view of the location score could be bolstered if a facility in a tertiary market benefits from strong demand and diversified tenants. However, a single-tenanted property with a large and diversified end-user base doesn't offset the risk of non-renewal from the in-place lessee in the same way that a diversified pool of tenants would be perceived to provide increased cash-flow stability.

Regarding the correlation between the location and cost-of-power sub-scores, the data center landscape includes primary markets with both high and low power costs. The location sub-score captures both latency and network effects as reflected in the size of clusters. On the other hand, the cost-of-power sub-score recognizes that facilities in markets with higher power costs could be challenged by facilities in lower-cost markets that can offer similar levels of latency.

As for latency and total cost of operations, the location sub-score accounts for latency and elements that affect cost, such as tax incentives, while the cost-of-power sub-score addresses a key element of total cost. These sub-scores collectively account for 50% of the utility score.

With respect to utilization rates and absorption rates, we expect them to be captured in the location score through the consideration of the size of the market. We would generally expect these rates to be favorable in top markets over the longer term given higher demand. We expect secondary and tertiary markets to be more exposed to volatility in vacancy and absorption rates. While strong markets will likely exhibit robust cloud deployments, the entrance of cloud providers into a market doesn't mean that it's a top market.

Over time, we will adjust the location scores as the data center industry evolves to account for developments in the market landscape, including changes in the size of data center clusters and shifting data center demand. The location sub-scores in Appendix 3 reflect the current landscape in the U.S. and Canada.

Cost-of-power sub-score

Comment:  Multiple market participants suggested that the cost of power should not be used as a sub-score in the determination of the utility of a hyperscale data center because it isn't among the most critical determinations of availability zone deployments. Similarly, another market participant noted that the total cost (power costs, lease rates, and equipment costs) is relevant for the establishment of data center facilities. In addition, multiple market participants stated that the relevance of the cost of power is minimal within the broader cost/benefit analysis of remaining at a data center. Another market participant asserted that unless properly weighted, the sub-score will be overly punitive to locations that already have well-established and impossible-to-relocate data ecosystems and availability zones, such as Silicon Valley, but may have higher power costs due to a variety of factors. Moreover, multiple market participants commented that higher-cost markets are being negatively affected, even though they might be high-value markets with less obsolescence risk.

Another participant stated that because cloud providers can't consider cheaper power in distant lower-cost markets when establishing their infrastructure, the location score provides sufficient differentiation. In addition, multiple market participants commented that the cost of power is more likely a competitive consideration within a market but not across markets. Similarly, another market participant commented that there is a narrow range of power costs within markets. The same participant asked about the frequency of calibration of the benchmark assumptions in Appendix 3 given that the cost of power can evolve quickly.

Response:  The cost of power factors into the overall competitive dynamics of a given facility, with higher costs reducing competitiveness. Nevertheless, we have decreased the weighting of the cost-of-power sub-score to recognize the impact that the cost of power has in the cost/benefit analysis of a facility. We also increased the weighting of the location sub-scores to reflect the importance of location, which measures the strength of a given market based upon latency, network effects, and elements that affect total cost, such as tax incentives. The methodology focuses on the competitive profile of each established property given the shorter-term nature of the lease contracts compared with the maturity profile of the liabilities. Therefore, the approach accounts for the contractual nature of the in-place lease contracts but doesn't assume tenant renewal of existing contracts. Tenants aren't required to renew, but such renewal clauses provide future optionality.

Data centers with lower power costs are expected to have a competitive advantage over the longer term and would likely be less susceptible to price competition. In addition, higher power costs incentivize the creation of competing markets that can service existing locations at a discount. With respect to the benchmark assumptions related to the cost of power, they will be updated periodically to reflect the evolving power landscape, though we would expect that high- and low-cost regions would remain in their relative positions.

Age-of-build subscore

Comment:  A market participant commented that the age-of-build score should account for phased development, as data center facilities are often built in phases. The same participant also noted that continuous upgrades will mitigate the impact of a building's age while stating that supply and demand are more relevant than a facility's age. Separately, the same participant commented that tenant improvements increase the longevity of the utility of a property, while the potential for upgrading older facilities should be considered. Another market participant noted that in certain high-demand areas, the age of build may be affected by limitations on supply.

Response:  The framework accounts for major improvements over time in determining the age of build sub-score, including retrofitting. However, we've clarified the methodology by specifying that we will take the average age of the building when a facility is built in phases. In terms of supply and demand, the framework rank-orders facilities based on their competitive nature, with older properties expected to have weaker demand and greater market-value volatility over time. The current landscape--characterized by under-supply and significant demand--might not be representative of future market dynamics for these long-dated transactions. The approach doesn't consider the role of tenant improvements, which are typically not contractual obligations.

Regarding property upgrades, the framework considers upgrades upon completion but does not anticipate such upgrades at the expense of the sponsor or tenants in our analysis. In terms of high-demand areas, an older building would receive a lower utility score--all else being equal--than a newer building to account for the older equipment and weaker competitiveness against newer properties that may feature more efficient designs. At the same time, the location sub-score would capture the strength of the given market.

PUE sub-score

Comment:  Multiple market participants highlighted that power efficiency should only be compared within a market, as regional climate must be considered in addition to the design of a facility. A similar comment was raised by another market participant, who highlighted that a property with a stronger PUE might not benefit from the same demand as a property with a weaker PUE. The same participant asserted that PUE may be unfairly discounted at a site based on tenant usage and that off-setting mitigants--such as regional or state sales tax--should be used to adjust the PUE subscore. Separately, a market participant suggested that we use total power usage effectiveness (TUE) instead because PUE lacks specificity and is not standardized across operators.

Response:  The utility function looks to rank-order facilities across markets by their competitive position based on key characteristics, including PUE. Higher PUEs leave facilities vulnerable to evolution in the industry and therefore challenge their longer-term competitive nature. While regional climate could affect PUE, we don't expect a large discrepancy in the average PUE across regions in the U.S. and Canada.

With respect to tenant usage, the framework anticipates weaker PUE performance during ramp-up. In addition, we've added language to the methodology to consider actual performance alongside the efficiency level indicated by infrastructure design. However, the approach doesn't consider off-setting mitigants in evaluating the PUE sub-score. Nevertheless, tax incentives are captured under the location sub-score, which looks to address total cost outside of power costs. In respect to TUE, this metric isn't readily available across markets and jurisdictions.

Lease renewal terms

Comment:  Multiple market participants commented that the assumption of setting the lease renewal term at the lesser of the portfolio weighted average original lease term and five years is overly punitive for issuers that have a weighted average original lease term of more than five years. Instead, they believed the renewal lease term should be set at the weighted average original lease term.

Response:  The framework accounts for typical optional renewal terms of five years under data center lease contracts. To the extent that an in-place tenant is not renewing, which is our base-case assumption, then we assume that the new tenant would renew at a similar term to the lease option from the in-place tenant. In addition, in a market environment characterized by less demand and more supply, we would expect that lease terms could shorten. Nevertheless, the length of the renewal term doesn't significantly affect the aggregate lease collections under the framework. However, we've updated the framework to consider adjustments when we see significantly longer original leases terms or to address higher lease renewal terms.

Credit quality of the tenant

Comment:  A market participant commented that low leverage and strong assets don't compensate for a weak pool of tenants, especially if there are no structural protections for cash flow coverage. Another market participant had a contrasting view regarding the ability of a strong tenant to offset a weaker utility score for a property in a tertiary market. Similarly, a market participant noted that the framework provides limited differentiation for the ratings on in-place tenants.

Response:  The framework captures the relative strength of the asset pool through the utility score. The strength of the tenants is reflected in the recessionary period utilization stress, while transaction leverage is captured in our cash flow analysis. We factor performance tests into our cash flow runs, and under our analysis, a transaction with weaker triggers would be expected to have greater cash-flow leakage. Transactions with greater debt service coverage typically perform better under our analysis, as they pay down faster when structural protections are in place.

To the extent a property portfolio is viewed as having high utility, the methodology assumes that such properties would be re-leased irrespective of the performance of the in-place tenants over a longer time horizon. At the same time, the analysis would capture the performance of the in-place tenant pool through the utilization stress. Given the shorter-term nature of lease contracts compared to the liability maturity profiles, the methodology deemphasizes the impact of the in-place tenant ratings while focusing on the utility of the properties. As a result, a strong tenant would not directly offset a utility score that was affected by a weak location sub-score. Nevertheless, the strong tenant could reduce the level of utilization stress applied.

Operational risk

Comment:  A market participant noted that the proposed framework didn't explicitly incorporate an assessment of the operator of the data center.

Response:  The framework states that we apply our global framework for assessing operational risk in structured finance transactions to assess whether the structure explicitly describes and assigns responsibilities that could weigh on the rating if they're not performed as agreed.

Triple-net leases

Comment:  A market participant commented on the approach to grossing-up revenue and expense items in relation to triple-net leases to reflect that future leases might not be triple-net. They suggested that there should only be grossing-up for presumed renewals and not for the in-place lease. In addition, multiple market participants commented that this is a different approach to that taken in the triple-net lease criteria, in which a triple-net lease tenant is not assumed to migrate to a gross lease.

Response:  This approach applies to turnkey facilities, whereby the special-purpose vehicle (SPV) may become responsible for the maintenance costs of the facility over time. Under the framework, we don't assume renewal to the same tenant upon termination of the initial in-place lease. As a result, from the onset of the analysis, we model the potential for such lease migration by estimating expense margins while grossing-up the lease contracts. This is not intended to affect credit given to cash flows during the initial lease term. With respect to the triple-net lease criteria, the lease contracts are typically restricted to double-net or triple-net leases, thereby reducing such transition risk. Moreover, the triple-net lease criteria anticipate the potential for additional costs that the SPV may bear over time.

Project financing

Comment:  A couple of market participants questioned how the proposed criteria would address project finance transactions after the construction phase. In addition, a market participant asked how the framework would be applied to shorter maturity profiles given the lack of depth for long-dated liabilities in newer markets.

Response:  The criteria scope outlines that projects with unmitigated construction risk will be analyzed under our project finance methodology, including during the operations phase. For transactions without construction risk or with mitigated construction risk, we expect to analyze such transactions using the data center criteria assuming other conditions are met, such as legal risk consistent with a structured finance rating.

The methodology is agnostic to financing structure and focuses on the transaction-specific legal final maturity. It doesn't assume refinancing of a shorter-dated maturity profile. Instead, the framework would model the legal final maturity associated with any such transaction.

Retail versus wholesale facilities

Comment:  Multiple market participants commented that the differences between retail and wholesale facilities warrant different approaches for analyzing both asset types, including for tenant churn and stickiness. Similarly, a market participant asserted that a wholesale facility is intrinsically more valuable than a retail facility given the continued trend of migration to the cloud, the credit quality of the customers, and the longer lease terms. In addition, another market participant commented that because wholesale properties (including built-to-suit facilities) have a much higher propensity to renew their leases due to the importance of data centers to their operations, the large costs associated with moving to a new location, and limited options to move, such facilities may be unfairly scored under the utility framework.

Response:  The approach accommodates all property types and business models with significant room for differentiation, including through the utility function, property valuations, and the utilization function. Given the current landscape of data centers in the U.S. and Canada, we would expect purpose-built wholesale facilities to carry more favorable underwriting assumptions, including cap rates and vacancy assumptions, lower utilization stresses, and higher utility scores. Nevertheless, given that lease terms are typically significantly shorter than the legal final maturity of data center transactions, the analysis focuses on the utility of the property. At the same time, the line between retail and wholesale has started to blur as more hybrid business models emerge.

Legal final maturity

Comment:  A market participant commented that ratings with a couple of years remaining before the anticipated repayment date should be less affected under the proposed methodology given that the change in the approach to market value declines looks to capture the long-term risk of volatility in property valuations.

Response:  Our data center ABS ratings address timely interest and ultimate principal repayment by the legal final date. As per our ratings definitions: "If the terms of an instrument specify a final maturity date (or legal maturity date) as well as an earlier, projected principal repayment schedule--sometimes referred to as the expected maturity date or the target amortization schedule--including instruments with a temporary principal write-down feature, the imputed promise is payment of principal due at final maturity in cash. As a result, missing a projected payment does not constitute a default, because it does not breach the imputed promise."

Market value declines

Comment:  A market participant commented that data centers should be treated as a separate property sector from industrial properties in terms of property value haircuts due to different supply/demand dynamics and differentiation in geographical concentrations. Another market participant commented that the calibration of the market value decline haircuts to those used for industrial properties doesn't address their expectations for milder performance volatility in the data center sector. In addition, a market participant commented that the haircuts don't consider the pace of inflation over the duration of a data center securitization, the capex an operator will invest into their assets, and the demand from the tenants to purchase the properties.

Separately, a market participant asked whether the rating on a tenant is considered when adjusting market value decline assumptions for property concentration. Another market participant disagreed with the proposed alignment of property value haircuts with our CMBS criteria given that data center ABS transactions have fee title to the properties as well as structural features that should reduce the risk of the property manager selling assets at highly distressed prices.

Response:  The framework aligns the LTV thresholds with those used for CMBS transactions backed by data centers, thereby acknowledging that the best alternative use of such properties would be an industrial facility. The current data center landscape has limited supply and high demand. We've not yet seen a data center downturn with waning demand and excess supply that would be representative of an 'A' economic environment. Nevertheless, we do have separate cap rates to address the specific dynamics of this sub-sector.

With respect to inflation, the proposed methodology accounts for rent escalations in our cash flow analysis, including during recessionary periods. However, the S&P Global Ratings value assumes the operating expense ratio to remain relatively stable through the cycle because inflation will have a similar impact on both lease revenue and expenses. Typically, the level of capex is commensurate with the maintenance needs of a given facility to sustain the in-place lease rate and occupancy level. The demand from tenants to purchase these properties reflects the current supply/demand imbalance. Regarding adjustments for property concentration, given the longer time horizon in data center securitizations compared with a typical CMBS transaction, we may provide minimal credit to investment-grade tenants in our property valuations, including when there's a property concentration.

The framework maintains the existing alignment with our CMBS property evaluation methodology in deriving the S&P Global Ratings property value through property-level analysis. This approach uses property type-specific revenue and expense line items--which are predominantly paid by the landlord except electricity--to calculate a net operating income before the application of a market stable cap rate. Rents are a core driver of our real estate analysis and are a significant driver of the S&P Global Ratings value. Our cap rate assumptions vary by property characteristics and across markets (primary, secondary, and tertiary). In addition, the criteria provide flexibility to address our cap rate assumptions based on market conditions.

The methodology recognizes the property fee interest by providing credit to lease cash flows in the analysis, which aren't directly considered in our CMBS approach. At the same time, the framework extends the CMBS approach to market-value declines, thereby aligning our approach with data center CMBS transactions. Nevertheless, the longer liquidation time afforded to ownership when compared with a mortgage refinancing doesn't mitigate the risk for a cyclical downturn at the point that assets would need to be liquidated. Moreover, CMBS transactions often exhibit sufficiently long-tail periods of five to eight years, which provide flexibility in liquidation timing. Lastly, the disposition periods in data center ABS transactions are often short term, thereby increasing the potential for market-value risk.

Lease rate declines

Comment:  Multiple market participants commented that the lease rate declines shouldn't be anchored to property valuations. Instead, they felt that rates should be driven by empirical data linked to supply/demand dynamics while considering the complexity and operation risk of moving locations. In addition, they cited the current strong pricing dynamics for data center operators, underpinned by robust demand in the data center industry in the U.S. and Canada, including during the COVID-19 shutdown. One of these market participants commented that because churn metrics typically include any price reductions upon renewal, the base utilization would likely result in double-counting rate reductions.

Response:  The framework includes cumulative lease rate declines that are applied over the liquidation time horizons, as indicated by the utility scores. The methodology generally looks to align the market-value decline assumptions assigned to the properties with the all-in lease rate declines. This approach reflects that in the analysis, changes to S&P Global Ratings property values are mainly driven by lease rate declines given that the analysis holds expense ratios and cap rates constant. Regarding empirical data, recent experience through the COVID-19 shutdown wouldn't be considered a stressed environment for the data center industry and wouldn't be consistent with stress levels above 'B'. Indeed, this period was quite beneficial for the data center industry.

In respect to the double counting of rate reductions, the proposed utilization function considers tenant performance and doesn't adjust for lease rate reductions from churn.

Property valuations and rating cap

Comment:  A market participant commented that valuation assumptions between CMBS- and ABS-backed data center transactions shouldn't be aligned given the difference in liability maturity profiles, with CMBS transactions traditionally exhibiting shorter time horizons. In addition, multiple market participants asked why the methodology typically caps ratings at 'A+' when other asset classes--including CMBS, SASB, and triple-net leases--can be rated as high as 'AAA'. These market participants asserted that higher ratings would be supported by:

• High tenant credit quality;
• The long contractual tenor of cash flows;
• Large-scale investments by tenants;
• The mission-critical nature of the facilities;
• Projections for long-term property valuations given future demand for computational capacity;
• More limited--yet sufficient--historical valuation and performance data for large-scale data centers;
• The stability of the cash flows given the ownership structure of the assets; and
• Soft-maturity anticipated repayment dates (ARDs) that reduce refinancing risk when compared to CMBS.

Response:  The methodology generally aligns the valuation assumptions between CMBS- and ABS-backed data center securitizations, including the same cap rates (see the Market Value Decline section of the criteria). In terms of maturity profiles, the longer-dated legal final maturities often seen in data center ABS increase the exposure to market-value volatility, including in relation to sector evolution and obsolescence risk. In addition, the data center sector has much shorter performance history than other commercial real estate properties and has not experienced any significant downturn due to the expanding demand. Nevertheless, instead of differentiating our approach to property valuations, we will typically cap data center ABS transaction ratings at 'A+' to address the uncertainty regarding any long-term predictions about property value and contractual rates.

In addition, the leases in a CMBS transaction traditionally extend past the loan maturity dates, while in data center ABS, the leases are significantly shorter than the average maturity profile. Moreover, while data center ABS transactions own the assets, the disposition periods are shorter than the hang periods in CMBS transactions, which could increase the potential for volatility in pricing at the point of disposition. At the same time, the special servicer can extend the loan terms in a CMBS transaction as they work out the loan over the long hang period (eight years at 'AAA'). We don't rate to the anticipated repayment date, and we don't view a missed ARD as a default. In contrast to triple-net lease transactions, the typical diversity of such portfolios reduces dependency on individual property performance, thereby allowing for ratings above 'A'. Transactions that achieve a 'AAA' rating are significantly diversified compared with a data center portfolio and aren't exposed to obsolescence risk.

Interconnection

Comment:  A market participant commented that a higher concentration of carriers doesn't necessarily mean that a property is more sought after from a wholesale leasing perspective.

Response:  The framework acknowledges that higher levels of interconnection typically allow for more stable income over time. As a result, properties with high carrier exposure are afforded higher utility scores. The lack of interconnectivity doesn't bias the application of the utility function to wholesale facilities.

Cap rates

Comment:  Multiple market participants commented that the proposed cap rate ranges are more severe than current market cap rates for data centers. In addition, they commented that when considering market-value decline stresses, the cap rates increase significantly. Another market participant believed that the cap rates should account for whether a property caters to hyperscale or retail tenants.

Response:  The methodology is accompanied by an update to the cap rate range, which recognizes recent historical pricing while incorporating flexibility to address specific market and property characteristics. In addition, the cap rates are aligned with our CMBS approach, which targets a market stable cap rate through the cycle. With respect to the market-value decline assumptions, they address property performance in higher stress economic environments, while the 'B' calibrated cap rate captures market volatility within a benign environment.

Recessionary period utilization stress

Comment:  A market participant disagreed with the implementation of the recessionary period utilization stress given that in recent downturns, we have seen an increase in data center activities by top cloud providers and that payment defaults are unlikely given the critical nature of such operations to their business lines. Another market participant commented that while risk of payment delay may be elevated in a recession, data center lease payments should be viewed as ranking senior to debt obligations of in-place tenants. In addition, a market participant commented that the lease term used in CDO Evaluator should be capped at the length of the recessionary period to reflect the likelihood of default during the recessionary period and not over the tenor of the underlying lease contract.

Response:  To address future potential downturns, the proposed framework anticipates the modeling of recessionary periods. The recessionary period utilization stress temporarily affects the level of lease cash flows received by the trust. The higher the economic scenario (liability rating), the more stressful the period. In addition, the approach considers the rating on the in-place tenants such that the impact of the stress scenario is muted for higher-credit-quality tenants. In addition, we would apply the higher of the base utilization stress and the recessionary period stress during recessionary periods.

With respect to the essentiality argument, despite the core nature of the cloud business lines to many in-place tenants, all data centers aren't necessarily strategically important to their future revenues. As a result, while the argument may be made that data center facilities are critical to the overall business of a company, this doesn't equates to guaranteed performance under all lease contracts.

CDO Evaluator simulates default behavior (quantum of default) separate from the application of a default timing curve. In the framework, the CDO Evaluator output reflects the risk associated with actual lease terms, while the recessionary period represents how we allocate the CDOE output.

Capital improvements

Comment:  A couple of market participants commented on the implementation of assumptions regarding capital improvements, which they suggested would be uneconomic for issuers given that our analysis considers significant lease rate and value declines. In addition, they commented that the utility score doesn't correlate with capital improvement needs. Similarly, another market participant commented that higher-cost locations should have higher cost estimates. Another market participant suggested that given that existing transactions already account for maintenance capex in the waterfall, the inclusion of capital improvement assumptions appears to be redundant. The same participant asserted that it's unreasonable to assume that discretionary expenses are paid via a senior expense item, instead suggesting they would be financed independently from the transaction waterfall, namely by an equity contribution from the parent or sponsor.

Response:  The framework looks to account for evolution in the data center sector over time, which could challenge the competitive position of existing facilities, through the implementation of assumptions regarding discretionary capital spending. The current assumptions for maintenance capex don't capture the discretionary spending assumptions proposed under the methodology to address the ongoing competitive positioning of a given property. The capital improvement assumptions are tiered by utility score to account for weaker utility facilities likely requiring more improvements than stronger facilities. In observing the utility sub-scores, the age of build, redundancy, cost of power, and PUE sub-scores can differentiate the potential for capital improvement needs. At the same time, stronger utility scores are typically associated with better location sub-scores. Such sub-scores can indicate market strength and the likelihood of passing through improvement costs to the tenant, thereby contributing to lower cost estimates. In addition, the introduction of capital improvements provides consistency of approach with our property valuations, which include a measurement of capex. Regarding the priority of payments, the methodology doesn't presume that such payments occur in the actual transaction waterfall. Moreover, the proposed approach doesn't account for external financing, which would fall outside of the structure.

Utility score

Comment:  A market participant agreed with the proposal to differentiate properties by way of property-level utility scores but suggested that such scores should be used as a driver of occupancy and lease rate decline rather than as a driver of assumed time to liquidation. This is because facilities with access to large quantities of power will likely remain fully leasable over the next 25-30 years due to projected demand for data storage and computation, scarcity of power availability, and the ongoing maintenance capex dedicated to the data centers backing ABS. Therefore, they suggested that properties that are older, located in weaker markets, or that exhibit other weaknesses are more likely to have falling lease rates and occupancies than to cease operating altogether.

Response:  Evolution in the data center industry over time could leave older facilities in weaker markets exposed to increased volatility in lease rates and property valuations. The utility function addresses such risks by liquidating weaker properties earlier in our analysis. This subjects such properties to additional stress akin to higher lease rate declines and vacancy by cutting off cash flows related to such properties earlier in the transaction lifecycle.

Powered shell facilities

Comment:  A market participant commented that powered shell facilities should carry a higher utility score with stronger modeled re-leasing outcomes given they feature materially lower risk related to operational expenditures, more alternative uses, and stickier retention of tenants upon renewal given a tenant's investment in the property.

Response:  Credit is provided to power shells within the analysis, including potential modifications to the capital improvement stress, lower cap rates to reflect stability of cash flows and alternative use, and consideration of a lower base utilization stress to the extent it is a single-tenanted property. With respect to the utility function, it's not clear that powered shells have a competitive advantage over turnkey over the long term. While power shells can likely be more readily updated to keep pace with newer builds, this investment would fall outside the securitization.

Analytical adjustments

Comment:  A market participant disagreed with considering leverage per rating level and the available cushion relative to peers in determining analytical adjustments based on transaction-specific factors, such as structural subordination or the amortization profile.

Response:  The consideration of leverage per rating level and available cushion relative to peers provides for an additional consideration in determining whether analytical adjustments are necessary. For example, when a transaction is exposed to greater property-value risk resulting from a longer ARD, we will account for the leverage of the transaction when considering whether to apply an analytical adjustment. Similarly, when there is structural subordination, we will consider the relative cushion to peers when determining the appropriate analytical adjustment.

Liquidation timing

Comment:  Multiple market participants believed there's a downward rating bias over time given that the liquidation timing is driven by the earlier of the assumed time to liquidation and the actual maturity of the shortest-dated liability. This results in cash flows being cut off earlier during surveillance of a rating.

Answer:  The framework considers the passage of time in the modeling of amortization, which typically occurs after the ARD. As a result, ratings should not experience significant downward pressure simply from the passage of time. In addition, we may adjust our stress scenarios during surveillance to incorporate actual transaction performance, including how contractual and utilization rates have performed under recessionary conditions or other stresses.