In these appeals, we consider whether the California Public Utilities Commission (“CPUC”) correctly determined the price that Pacific Bell Telephone Company (“Pacific”) may charge its competitors for access to its local telephone network, pursuant to the Telecommunications Act of 1996 (“Act” or “1996 Act”) arid the implementing regulations of the Federal Communications Commission (“FCC”). We conclude that, although the general methodology chosen by the CPUC to calculate a common cost markup was appropriate, the CPUC improperly implemented the methodology by attributing some common costs to wholesale operations that should have been attributed to retail operations. We therefore reverse the decision of the district court with respect to the amount of common costs that Pacific’s competitors must pay for access to Pacific’s network. We- affirm the decision of the district court in all other respects.
I. Background
Because of the expense and difficulty of installing the lines and hardware necessary for local telephone service, the provision of local telephone service was thought for many years to be.a “natural monopoly.” See AT & T Corp. v. Iowa Util. Bd. (Iowa I), 525 U.S. 366, 371, 119 S.Ct. 721, 142 L.Ed.2d 835 (1999). States therefore granted local telephone companies monopolies in .the provision of local telephone service. Until the 1970s, AT & T was the provider of most of the nation’s local telephone service, as well as the provider of long distance service. As a result of an anti-trust suit brought by the federal government, however, AT & T was forced to divest itself of twenty-two Bell Operating Companies that provided local telephone service., See AT & T Corp. v. FCC., 220 F.3d 607, 611 (D.C.Cir.2000). The Operating Companies were forbidden to provide long distance service. During this time, local Operating Companies controlled the provision of local telephone service through state-sponsored monopolies, while other companies competed to provide long distance telephone service.
Congress dramatically altered this structure when it passed the 1996 Act. Pub.L. No. 104-104, 110 Stat. 56 (codified in scattered sections of 47 U.S.C.). Relying on new technological developments that made it possible for other providers to gain access to local telephone companies’ networks, the Act eliminated the monopoly protections granted to the Operating Companies. It further required that local telephone companies, termed Incumbent Local Exchange Carriers (“ILECs”), offer access *898to their local networks, either by selling local telephone service to Competitive Local Exchange Carriers (“CLECs”) at wholesale rates, by leasing parts of their networks, or by allowing competitors to connect to their networks. AT & T Corp., 220 F.3d at 611. In return, ILECs were permitted to enter the long distance telephone and the cable television markets, both of which had been previously forbidden to them.
To determine how much an ILEC may charge CLECs to gain access to its network, the Act allows the parties to negotiate an agreement providing for the terms of access, including the price for such access. See 47 U.S.C. § 252(a)(1). If the parties fail to agree on those terms, the Act requires state commissions such as the CPUC to resolve the dispute by arbitration. Id. § 252(b). As part of that arbitration process, the state commissions are directed to set rates that are “just and reasonable” in light of the cost of providing the various network elements. The Act provides that these rates should be nondiscriminatory and should allow for a reasonable profit. Id. § 252(d)(1).
The FCC promulgated regulations governing the methodology to be used by the state commissions in the determination of the rates to be charged by the ILECs. In re Implementation of the Local Competition Provisions in the Telecommunications Act of 1996, 11 F.C.C.R. 15499, 1996 WL 452885 (1996) (“Local Competition Order”); see also Verizon Communications v. FCC, 535 U.S. 467, 122 S.Ct. 1646, 152 L.Ed.2d 701 (2002) (upholding the FCC regulations). Under this methodology, termed Total Element Long Run Incremental Cost (“TELRIC”), ILECs are entitled to recover “the forward-looking costs directly attributable to the specified element, as well as a reasonable allocation of forward-looking common costs.” Local Competition Order ¶ 682. Thus, the cost for a particular unbundled network element (“UNE”) has two components: the direct cost of providing the element itself and the portion of the ILECs’ common costs attributable to the provision of multiple elements. We discuss these two kinds of costs in turn.
First, direct costs are those that are “directly attributable” to the UNE. These are costs that are “incurred as a direct result of providing the network elements, or [that] can be avoided, in the long run, when the company ceases to provide them.” Id. ¶ 691. The FCC directs state commissions to measure these costs “based on the use of the most efficient telecommunications technology currently available and the lowest cost network configuration, given the existing location of the incumbent’s wire centers.” 47 C.F.R. § 51.5015(c)(1) (2003). While the technology on which the costs are based is not necessarily the technology that the ILEC actually uses, the FCC reasoned that this method of calculating costs would “best replicate[ ], to the extent possible, the condition of a competitive market,” which in turn would “allow[] the [CLEC] to produce efficiently and to compete effectively, which should drive retail prices to their competitive levels.” Local Competition Order ¶ 679.
Second, common costs are those that are common to multiple UNES.1 These are costs that are “incurred in connection with *899the production of multiple products or services, and remain unchanged as the relative proportion of those products or services varies (ie., the salaries of corporate managers).” Id. ¶ 676. The FCC conceived of two types of common costs: those that are specific to a particular subset of UNEs, and those that are common to the entire corporation:
As discussed above, some of these costs are common to only a subset of the elements or services provided by incumbent LECs. Such costs shall be allocated to that subset, and should then be allocated among the individual elements or services in that subset, to the greatest possible extent. Common costs also include costs that are incurred by the firm’s operations as a whole, that are common to all services and elements (e.g., salaries of executives involved in overseeing all activities of the business), although for the purpose of pricing interconnection and access to unbundled elements, which are intermediate products offered to competing carriers, the relevant common costs do not include billing, marketing, and other costs attributable to the provision of retail service. Given these common costs, setting the price of each discrete network element based solely on the forward-looking incremental costs directly attributable to the production of individual elements will not recover the total forward-looking costs of operating the wholesale network.... [A] reasonable measure of such costs shall be included in the prices for interconnection and access to network elements.
Id. ¶ 694.
In calculating common costs, the FCC provided the states some latitude to choose a methodology, so long as the method is “consistent with the pro-competitive goals of the 1996 Act.” Id. ¶ 696. However, the FCC suggested various allocation methods it considered to be proper. One of those methods was adopted by the CPUC in these proceedings. Under this method, common costs are allocated “using a fixed allocator, such as a percentage markup over the directly attributable forward-looking costs.” Id. This markup reflects a particular UNE’s share of the common costs. Thus, if 20% of an ILEC’s costs are what the FCC would consider common, adding 20% to the direct cost of any given UNE would allow the ILEC to recover its total costs associated with providing that UNE. The ultimate goal of such a markup is to spread the costs that are not directly attributable to any particular UNE over all the UNEs that are supported by those costs.
As a result, when a CLEC leases a particular UNE, it is paying not only the direct costs of that UNE, but also the part of the common costs attributable to the production and maintenance of the particular UNE. Depending on the type of common cost, all or part of it will be attributable to UNEs and allocated among them. Thus, for example, all of the cost of salaries of technicians working only to service UNEs will be allocated to the various UNEs they service. By contrast, only part of the cost of the salary of the CEO of the company will be allocated to UNEs because the CEO supports the operations of the entire company, not just those that provide UNEs.
TI. Proceedings Below
A. CPUC Proceedings
In this case, the CPUC determined what Pacific, an ILEC, may charge CLECs such as appellants MCI and AT & T for access to various network elements. Prior to the passage of the 1996 Act; California had passed legislation of its own that was designed to open local telephone service to *900competition. See Cal. Pub. Util.Code § 709.5 (West 2004). As a result, the CPUC had already begun holding hearings to implement the California law when the 1996 Act was passed. See The Open Network to Bottleneck Services and a Framework for Network Architecture Development of Dominant Carrier Networks, D.96-08-021, R.93-04-003, 67 CPUC 2d 221 (“OANAD proceeding”). After passage of the 1996 Act and the adoption of the Local Competition Order by the FCC, the CPUC adopted the TELRIC methodology, finding that it was superior to the methodology it had initially decided to employ. D.98-02-106 at 18 (“First Cost Decision”).2 One of the features of the TEL-RIC methodology that the CPUC found attractive was the requirement that retail costs be removed from consideration when determining the amount of common costs to be borne by the CLECs in gaining access to UNEs. In concluding that the TELRIC methodology was better suited to its purposes, the CPUC relied on two experts who testified that because the CLECs are purchasing access to specific network elements, and not the services that the CLEC will itself sell to customers, TELRIC was a more appropriate method of calculating costs. They explained:
TELRIC is a cost concept that refers to an intermediate level of production, or to goods sold at wholesale. Therefore, TELRIC costs should not include any of the costs of supplying services to end user customers. This is an important difference between the two cost concepts. An entrant using [UNEs] as the inputs for its end user services can compete with the incumbent either on the level of retailing, or on the way it combines those elements to provide services, or both. If the entrant has to incur both its own retailing costs as well as having to pay some of the incumbent’s retailing costs, it faces a barrier to entry.
First Cost Decision at 21-22 (emphasis in original). Thus, the CPUC adopted TELRIC, in part, in order to ensure that the CLECs would not be forced to pay the ILEC for the costs the ILEC incurred in performing its retail operations.
1. Determination of Common Costs
In its First Cost Decision adopting the TELRIC methodology, the CPUC noted that “the costs from retail services should be excluded from the price of a UNE.” Id. at 52. This is so, according to the CPUC, because retail costs “are not attributable to the production of network elements that are offered to interconnecting carriers.” Id. at 62.
In the same decision, the CPUC determined the amount of common costs that would be included in determining the common cost allocator, as well as the amount of common costs that were retail-based and thus would be excluded. See id. at 63 n. 55 (“In this phase ... our task is to determine the retail portion of common costs that are likely to be incurred by Pacific in a forward-looking environment.”). While Pacific claimed before the CPUC that its total common costs were approximately $1.2 billion, MCI and AT & T claimed that this figure included over $200 million in retail-related costs. The CPUC agreed with MCI and AT & T only in part, determining that only $68 million *901of the costs claimed by Pacific as common had a “clear retail component” and thus should be excluded. The district court described these excluded costs as those that, “in a hypothetical forward-looking environment in which Pacific is solely a wholesaler, Pacific would not incur.” AT & T, 228 F.Supp.2d at 1093. In other words, under the CPUC’s analysis, if Pacific were to cease all of its retail operations, the CPUC should consider all of the remaining common costs. Since only $68 million of the additional common costs claimed by Pacific would not exist in a hypothetical world in which Pacific engaged in no retail operations, the CPUC excluded only those costs.
2. Operation Support Systems
In the CPUC’s Second Cost Decision, it set the forward-looking non-recurring costs for Pacific’s Operations Support System (“OSS”) gateways. See D.98-12-079 at 1 (“Second Cost Decision”).3 An OSS gateway is a kind of UNE that allows CLECs to interconnect with the ILEC. Costs associated with OSS gateways “consist of pre-ordering, ordering, provisioning, maintenance and repair, and billing functions supported by an incumbent LEC’s databases and information.” 47 C.F.R. § 51.319; see also Local Competition Order at ¶ 517-18. As noted by the FCC and the CPUC, an OSS gateway is vital to implementation of the Act, since it is only through such a gateway that a CLEC can efficiently gain access to an ILEC’s network to provide service to end-users. See Second Cost Decision at 4-5.
Although the primary purpose of the CPUC was to set the non-recurring costs (“NRCs”) for the OSS gateway, it also received studies from Pacific addressing the need to recover recurring costs for the provision of OSS gateways. NRCs are “one-time expenses associated with initiating or disconnecting a service,” and include the labor necessary to effectuate the interconnection of a CLEC with an ILEC. Recurring costs, by contrast, are those necessary to maintain the OSS gateway. Compare Second Cost Decision at 12 with id. at 42-43.
The NRCs of providing OSS gateways depend, in part, on how much labor is required to effectuate the interconnection. For example, a CLEC might call in or fax an order to the ILEC requesting service, with a CLEC employee then implementing the request. This, of course, would involve high labor costs, increasing the NRC. Or, to provide another example, the CLEC and ILEC might interact electronically, with minimal employee involvement, leading to lower NRCs. See id. at 3-5.
In the proceeding before the CPUC to determine the OSS gateway NRCs, the interested parties submitted different proposed models to the CPUC for setting rates for the OSS gateways. Not surprisingly, MCI and AT & T assumed significant electronic interaction, leading to much lower NRCs. Id. at 19. Pacific, on the other hand, submitted a model with three variations, each with different levels of electronic interaction (and cost). Id. at 22-23. Ultimately, the CPUC adopted Pacific’s model for determining NRCs, though with modifications. Id. at 30. It concluded that Pacific’s model was the only one that included support for all of the UNEs specified by the CPUC and most of those specified by the FCC, and that accounted for variations in the ways that CLECs might choose to connect with the ILEC. Id. at 24-27. As required by the *902FCC, the CPUC required that the costs ultimately adopted be based on the most efficient technology currently available. That is, it selected the model that assumed a high level óf electronic interaction.
With respect to the recurring costs of providing the OSS gateways, the parties were also in disagreement. MCI and AT & T contended that the recurring costs were so low as to be “de minimis” and thus should not be included in the cost of proT yiding the OSS gateways. Id. at 43. Pacific, on the other hand, argued that these costs were significant and should therefore be passed on to the CLECs. Other parties responded that many of these recurring costs also supported Pacific’s own retail operations, since the OSS gateways are also used by the ILEC in providing local telephone service to its own customers. Id. at 45. Based on the costs studies before it, the CPUC concluded that many of Pacific’s proposed costs were retail-based, and further, that Pacific had failed to isolate costs that were unique to providing CLECs access to the OSS gateways:
We find merit in AT & T/MCI’s ... assertions that Pacific’s ... OSS functions provide benefits to[its] own retail operations. While Pacific attempts to identify discrete customer specific costs in its access port cost analysis, its showing fails to reflect that the underlying systems and databases are in place to serve both retail and wholesale customers and thus the costs cannot be attributed solely to CLCs. We find Pacific' has not . demonstrated that these costs should be recovered from competitors.
Id. at 45.
The CPUC concluded that Pacific had provided insufficient evidence to permit the CPUC to separate the retail-related recurring costs from the non-retail-related recurring costs as required by TELRIC. While concluding that the costs identified by Pacific should not be included in the price charged to CLECs, the CPUC did suggest that those costs might qualify as implementation costs, which were being considered in a separate proceeding.4 Id. at 46. ’ Pacific, apparently concluding that those costs were not implementation costs, declined to submit its cost studies at the separate proceeding and did not seek rehearing of the CPUC’s decision in this proceeding.
3. Determination of UNE Costs
In its third and final decision, the CPUC set the rates that Pacific would be permitted to charge for its UNEs. See D.99-11-050 (“OANAD Decision”).5 Included in that decision was a determination of how the common cost “markup” would be calculated. Again, the interested parties offered different methods. Pacific argued that the markup should be determined by dividing the allocated common costs by the total direct costs of the various network elements. Id. at 54. This method creates a fraction, with the numerator being the total common costs, and the denominator being the total direct costs. The CPUC discusses this method in terms of the numerator and denominator, and this terminol*903ogy was adopted by the district court. For ease of discussion, we will also adopt this terminology. Using this methodology, Pacific arrived at a figure of 22% for the common-cost markup. That is, the direct cost of each UNE would be increased by 22% to account for the common costs incurred by Pacific in producing that UNE. MCI and AT & T, however, argued that the denominator should also include retail-related and Category III costs. Category III costs are associated with unregulated aspects of a LEC, such as the provision of Internet access. MCI estimated that this adjustment would increase the denominator by approximately $2.9 billion, yielding a markup of 14%.6
The CPUC adopted, in substantial part, Pacific’s proposal. Although it added $375 million to the denominator to account for direct costs previously not counted, thereby decreasing the markup to 19%, it rejected AT & T and MCI’s position that the denominator should include retail-related and Category III costs. In so doing, it adopted a statement from Pacific’s representative that AT & T and MCI “ignored the fact that all of the shared and common costs that are retail-related have been removed from the shared and common costs in this phase.... It is therefore entirely appropriate and proper to divide the non-retail shared and common costs by the non-retail [direct costs] to obtain the non-retail ... markup for UNEs.” Id. at 64 (emphasis in original).
With respect to the Category III costs, the CPUC noted that “Pacific’s unregulated [Category III] businesses have their own overhead organizations. To the extent they use Pacific’s overhead departments, the costs are directly billed to them under the Commission-ordered transfer mechanisms.” Since these costs, like retail-related costs, are excluded from the common costs claimed by Pacific, the CPUC concluded that they should likewise be excluded from the direct costs in the denominator. Id. at 65-66.
Finally, the CPUC declined to revisit the issue of OSS gateway costs. It noted that “[n]o OSS [recurring] costs were adopted, because the models submitted by Pacific were [ ] found to contain significant flaws.” Id. at 55 n. 54. It once again suggested that Pacific “seek recovery of OSS recurring costs attributable to servicing CLECs” in the proceeding addressing implementation costs.
B. District Court Proceedings
MCI and AT & T brought suit in federal district court challenging the CPUC’s decision. Specifically, they argued that the manner in which the CPUC calculated the markup, and consequently, the rates at which Pacific must lease various UNEs to CLECs, violated the Act and the FCC’s Order. Pacific filed various cross-claims also alleging that the CPUC violated the Act. It argued that the CPUC acted unlawfully in not setting rates for recurring OSS gateway costs, and asserted other claims not relevant to this appeal. In considering cross-motions for summary judgment, the district court affirmed the decision of the CPUC in relevant part.
First, MCI and AT & T challenged the CPUC’s decision to adopt Pacific’s methodology in determining the common-cost markup, in particular its failure to include the direct costs of all of Pacific’s operations in the denominator, including retail and Category III services. AT & T, 228 F.Supp.2d at 1101. MCI and AT & T *904argued, that “by including only direct UNE costs in the denominator, the CPUC effectively allocated all of Pacific’s firm-wide common costs exclusively to the cost of UNEs.” Id. Pacific responded in the same manner as it did before the CPUC, arguing that since retail and Category III costs were not included in the common-costs the numerator), it was inappropriate to include them in the denominator’
The district court agreed with Pacific. Citing ¶.694 of the Local Competition Order, the district court held that “the TEL-RIC methodology calculates an ILEC!s costs in a hypothetical, forward-looking environment in which the ILEC is solely a wholesaler.” Id. at 1102. As a result, according to the district court,
all retail costs [are excluded] from the calculation of common costs and the common cost markup,- because retail costs should not be included in the ■prices that ILECs ■ charge CLECs to lease UNEs. .... In other words, Pacific’s firm-wide common costs are the common costs that Pacific would experience as a wholesale firm. In such a wholesale environment, it would make no sense to divide Pacific’s common ■costs by anything more than Pacific’s direct costs of UNEs. Pacific’s firm-wide, wholesale-only common costs are reasonably divided by only the direct costs of UNEs, and not to Pacific’s retail services or Category III services, which would not exist in a wholesale only environment.
Id. The district court thus found that the CPUC’s determination of the common-cost markup was reasonable, and affirmed the decision of the CPUC.
. The district court also affirmed the CPUC’s refusal to permit Pacific to seek reimbursement for recurring OSS gateway costs. It found that the CPUC had made a decision on the merits with respect to whether Pacific should be reimbursed for OSS gateway recurring costs, and that by referring Pacific to the Local Competition Proceeding, the CPUC was not guaranteeing that it would reimburse Pacific for those costs. Id. at 1105-06. Ultimately, however, the district court affirmed the decision of the CPUC on exhaustion grounds, holding that by failing to petition the CPUC for rehearing on the matter, Pacific was precluded from appealing that issue to the district court. Id. at .1106.
MCI and AT & T appealed. They once again claim, inter alia, that the CPUC and the district court erred in not including retail and Category III costs in the denominator, and that the common-cost markup was thus inflated. Pacific cross-appealed, claiming once again that the CPUC should have set rates for the recurring costs in providing OSS. For the reasons set forth below, we affirm in part, reverse in part, and remand for further proceedings.
III. Standard of Review
We review the district court’s decision granting a motion for summary judgment de novo. Abelein v. United States, 323 F.3d 1210, 1213 (9th Cir.2003). We may affirm the district court on any basis supported by the record. Newton v. Diamond, 349 F.3d 591, 594 (9th Cir.2003). We also consider de novo “whether the agreements comply with the Act and its implementing regulations.” U.S. West Communications, Inc. v. Jennings, 304 F.3d 950, 958 (9th Cir.2002). However, where the CPUC makes factual findings, those findings are reviewed for substantial evidence. MCI Telecomm. Corp. v. U.S. West Communications, 204 F.3d 1262, 1266-67 (9th Cir.2000).
IV. Discussion
A. Common Cost Markup
We turn first to the CPUC’s decision to adopt Pacific’s methodology in *905determining the common-cost markup to be applied to the cost of UNEs, a decision we review de novo. We interpret the 1996 Act and the FCC regulations in light of Congress’s pro-competitive goals. See Local Competition Order ¶ 618.
The Local Competition Order grants state commissions considerable latitude in determining how to allocate common costs. Id. ¶ 696 (“[F]orward-looking common costs shall be allocated among elements and services in a reasonable manner.... ”). The CPUC decided to use a fixed allocator of common costs, a method suggested by the FCC’s Order. Given the consistency of the FCC’s Order with the terms and goals of the Act, we cannot say that the CPUC’s decision to use that methodology violates the Act. Since the common costs are, by definition, costs that are common to all or part of an ILEC’s operation, allowing a standard markup over all UNEs for common costs properly attributable to UNEs should allow an ILEC to recover the full costs of providing UNEs, as contemplated by the Act itself, while still allowing CLECs to compete with the incumbents on an level playing field. That is, the price paid by the CLEC should approximate the cost to the ILEC of providing the UNEs.
MCI’s and AT & T’s central challenge is not to this basic methodology, but rather to the CPUC’s implementation of it. They claim that the denominator should include retail-related and Category III direct costs, not just the direct costs of providing the UNEs. As noted by the CPUC, however, this would mix apples and oranges. See OANAD Decision at 64. The CPUC allocated the common costs of providing UNEs by dividing these common costs by what the CPUC determined were the direct costs of the UNEs. If, as the CPUC claims, the common costs are only those involved in wholesale operations, adding the direct costs of providing retail-related and Category III services to the denominator would reduce the percentage markup and would, to that extent, underestimate the amount of common costs required to produce a given UNE.
Of course, the CPUC could have included all retail-related and Category III common costs in the numerator, and included the direct costs of providing those services in the denominator. This would have produced a markup that could be applied to all services, both those that the ILEC would provide to CLECs (such as UNEs) and retail services that the ILEC provides to its own retail customers. However, that calculation of the markup would be less accurate, because the markup for providing network services might be different from that for retail services. For instance, if the markup for retail services were 30%, and that for network services were 20%, with each comprising 50% of an ILEC’s business, the markup including both retail and wholesale would be 25%. When applied to UNEs, the 25% markup would therefore overestimate the cost or providing UNEs. This inaccuracy, however, might be outweighed by the ease of application and the difficulty of determining which common costs have a retail component, and which do not.
The point here is not to discuss the merits of such an approach, which we decline to do, but simply to point out that, like the CPUC’s approach, such a method would be consistent, comparing apples to apples. To include only wholesale common costs in the numerator and to include all firm-wide direct costs in the denominator, by contrast, would be inconsistent, and would underestimate the common costs that Pacific has a right to recoup when it charges CLECs for access to its networks. This would essentially force Pacific to subsidize MCI and AT & T’s retail operations, *906which the Act does not require. The CPUC was therefore correct in rejecting MCI's and AT & T's methodology.
This does not, l~owever, end our inquiry. The for~going analysis assumes that the CPIJC properly removed all retail-related costs from both the common costs and the direct costs of providing the UNEs, such that the num~rator and denominator included only wholesale costs. That is, while we conclude that the CPUC was correct in concluding that it should consider wholesale costs and not retail costs, it is not necessarily the case that they actually did so. Rather, it appears that the CPUC included some retail-related common costs in the numerator, thereby artificially inflating the markup.
The key to understanding the CPUC's mistake is in the district court's description of the hypothetical world contemplated by the CPIJC. According to the district court, in calculating the common cost markup the CPUC considered which dom-mon costs would remain in a "hypothetical forward-looking environment in which Pacific is solely a wholesaler" and in which Pacific engaged in no retail activity. AT & T, 228 F.Supp.2d at 1093. The CPUC apparently believed this hypothetical exercise to be required by the `FCC's Local Competition Order, ¶ 694. Paragraph 694 requires no such thing, however. While the Order does appear to forbid state commissions from considering retail common costs, it does not require it to imagine such a hypothetical state. Rather~ imagining a hypothetical wholesale-only world inflates the common costs, since in such a world common costs that in actuality support both retail and wholesale activities are assigned solely to the ILEC's wholesale activities.
An example from Pacific's own brief 11-lustrates how this is so. Pacific notes, correctly, that the salary of an executive, such as the company's president, should be considered a common cost. See id. ("Common costs also include costs incurred by the firm's operations as a whole, that are common to all services and elements (e.g., salaries of executives involved in overseeing all activities of the business) ."). Pacific further argues that most of the cost of the company's president should be considered `as a common cost attributable to wholesale operations, since in a wholesale-only environment, the ILE C would still require a president. While the current cost or salary of the ILEC's president might be slightly reduced to account for the fact that the president of a smaller company (i.e., a wholesale-only telephone company) might be paid less, according to Pacific the majority of the president's salary should be counted as a wholesale common cost.
Pacific's example may be used to illustrate the problem with the methodology. We agree with Pacifi~ that in the hypothetical world in which Pacific is a wholesale-only company, the president's salary might be reduced slightly from its present amount, but that it would likely remain close to the original amount. But what if, in the real world, Pacific actually performs half wholesale and half retail operations? In that event, in the real world, the common cost of the president's salary should be allocated half and half between the wholesale and retail operations. To put it another way, the president's salary might remain at 80% of its present level if all retail operations were terminated or if all wholesale operations were terminated. By considering only the first alternative, the CPUC hypothetical would exclude only 20% of the president's pay as being attributable to retail operations. To the extent that the hypothetical salary of Pacific's president in the hypothetical wholesale-only world exceeds one-half of his or her *907present real-world salary, that hypothetical salary overstates the common costs properly attributable to Pacific’s wholesale operations.
Nothing in the Act or the implementing regulations requires the TELRIC methodology to be implemented in the manner adopted by the CPUC. Indeed, this implementation leads to an anti-competitive result. As noted above,¶ 694 nowhere mentions the hypothetical world contemplated by the CPUC. Although the direct costs of the UNE must be “measured based on the use of the most efficient telecommunications technology currently available and the lowest cost network configuration, given the existing location of the incumbent’s wire centers,” this hypothetical-world approach does not extend to the calculation of common costs. See 45 C.F.R. § 51.5015(c)(1). More important, the CPUC’s hypothetical world produces a situation in which a CLEC has to pay its own direct retail costs and some of the common retail costs of the incumbent.
Stated another way, under the methodology adopted by the CPUC, Pacific will not have to pay all of its retail-related common costs, thereby allowing it to charge lower prices for its own retail services than it otherwise would. Conversely, the CLECs must pay some of Pacific’s retail-related costs, thereby increasing the CLECs’ costs of providing telephone service and exerting upward pressure on the prices they charge their customers. Thus, under the CPUC’s approach, the CLECs are essentially subsidizing Pacific’s provision of retail services and, to that extent, increasing their own costs. As previously noted, the very reason why the CPUC adopted TELRIC was to prevent the barrier to entry faced by a CLEC if it has to “incur its own retailing costs as well as ... the incumbent’s retailing costs.” First Cost Decision at 21-22. As the Act, the Local Competition Order, and the CPUC’s decisions themselves make clear, this is an anti-competitive and unreasonable interpretation of the Act.
Our conclusion is different with respect to Category III costs, however. As noted above, -the CPUC found that there was a separate accounting mechanism for Category III costs, such that those services are billed for any overhead they use from Pacific, as required by the FCC. OANAD Decision at 65-66. To support this conclusion, the CPUC relied on the testimony of Richard Scholl, a witness for Pacific. The district court also accepted this conclusion. See AT & T, 228 F.Supp.2d at 1102 n. 16. We hold that this finding is supported by substantial evidence. Unlike retail common costs, which we have concluded have been improperly included in the CPUC’s calculations of common costs, the separate billing structure of Pacific’s Category III services prevents the common costs used to support them from being included in Pacific’s calculation of its common costs. The CPUC was correct in not including Category III costs in the denominator.
We therefore reverse the decision of the district court affirming the CPUC’s determination establishing the common cost markup added to the direct cost of UNEs. On remand, the district court should direct the CPUC to calculate the markup using a common cost figure that properly accounts for the portion of Pacific’s activities that are retail-related.
B. OSS Gateway Recurring Costs
On cross-appeal, Pacific challenges the CPUC’s refusal to set recurring costs for the provision of OSS gateways. At the outset, we agree with Pacific that it did not have to petition the CPUC for a rehearing in order to preserve its right to appeal its decision. Under California state *908law, state courts do not have jurisdiction to hear appeals from a decision of the CPUC unless the party seeking review has petitioned for rehearing within 30 days of the decision. See Cal. Pub. Util.Code § 1731(b). We, however, are not bound by this procedural requirement. In AT & T Communications Systems v. Pacific Bell, 203 F.3d 1183, 1186 (9th Cir.2000), we determined that Congress did not intend that varying state procedural .requirements should act as bars to judicial review of proceedings pursuant to the 1996 Act. Although that decision dealt with a different stage of proceedings, the underlying reasons for the holding of that case— speed and uniformity of interpretation— are equally applicable here. The district court therefore erred in failing to reach the merits of Pacific’s claim.
On the merits, however, we affirm the CPUC’s decision. Because the determination of what Pacific should be permitted to charge CLECs for the provision of OSS gateways is factual in nature, we review it for substantial evidence. As noted by the district court, the CPUC’s determination that Pacific should not be able to recover recurring costs was a substantive decision on the merits. AT & T, 228 F.Supp.2d at 1105. The CPUC concluded that many of the recurring common costs attributed by Pacific actually supported its own retail activity. The FCC regulations and the Local Competition Order direct state commissions to exclude such retail costs when considering what rates to set for UNEs. See Local Competition Order ¶ 691 (“Retailing costs ... are not attributable to the production of network elements that are offered to interconnecting carriers and must not be included in the forward-looking direct costs of an element.”); see also 47 C.F.R. § 51.505(d)(3). Since Pacific did not prove that these direct costs were separate from retail costs, the CPUC felt it was not justified in including them in the rate for OSS gateways. The CPUC did note that these costs might instead be considered implementing costs, and that Pacific could present those costs to the Local Competition Proceeding that was considering such costs. However, the CPUC specified that it was not “predetermining that the [recurring OSS] costs presented [in the Second Cost Decision] are eligible for recovery in that proceeding.” Second Cost Decision at 46. More important, the possibility that the costs might be implementation costs does not affect the CPUC’s determination that they were not recurring OSS gateway costs.
Pacific bears the burden of proving what direct costs should be included. See Local Competition Orde^ 680 (“We note that incumbent LECs have greater access to the cost information necessary to calculate the incremental cost of the unbundled elements of the network. Given this asymmetric access to cost data, we find that incumbent LECs must prove to the state commission the nature and magnitude of any forward-looking costs that it seeks to recover in the prices of interconnection and unbundled network elements.”). Pacific, seizing' on language in ¶ 695 that the ILEC shall have the burden of proof “in the arbitration process” argues that since the CPUC’s decisionmaking process was started before the passage of the 1996 Act, it is not an “arbitration process” and that the Act’s burden of proof requirement is therefore inapplicable. However, ¶ 695 applies to the determination of common costs, whereas ¶ 680, which lacks the limiting language of ¶ 695, applies to the determination of direct costs. Since the recurring costs are a type of direct cost, ¶ 680 applies. Moreover, even if ¶ 695 did apply, the proceeding in any event is better seen as an arbitration process. Congress anticipated that interconnection agreements would be decided either by agreement of *909the parties or through arbitration by a state committee. Although the OANAD proceeding did not go through the normal process by which the parties first attempted to enter an agreement, but rather was initiated by the CPUC, it closely resembles the arbitration process. To the extent they are different from the rules governing agreements between LECs, the arbitration rules should control.
Given the evidence before it, the CPUC acted reasonably in concluding that Pacific had not met its burden of proving that it was entitled to be reimbursed for recurring OSS gateway costs. When the CPUC made this determination, it had Pacific’s cost study before it, as well as that of other ILECs. It also had a declaration filed by MCI and AT & T arguing that the costs were de minimis. On the basis of the record before the CPUC, we find that there was substantial evidence supporting its decision not to include recurring costs. Such a conclusion is well within the authority and expertise of the CPUC, which was charged by Congress and the FCC with implementing the terms of the 1996 Act. We will not disturb that determination here.
Conclusion
The decision of the district court with respect to the determination of common costs is REVERSED and REMANDED to allow a determination of the correct amount of wholesale common costs, excluding retail-related common costs. The determination of the OSS recurring costs is AFFIRMED.
AFFIRMED in part, REVERSED in part, and REMANDED.
Each party shall bear its own costs on appeal.
. "Common costs” are sometimes referred to as "shared and common costs” by the CPUC, referring to costs that are shared by only some network elements and those that are common to all network elements. For simplicity, we refer to them as common costs. See AT & T Communications of Cal. Inc. v. Pac. Bell Tel. Co., 228 F.Supp.2d 1086, 1090 n. 2 (N.D.Cal.2002).
. Interim Decision Adopting Cost Methodology, Evaluating the Hatfield Computer Model, and Deciding Other Issues Related to Cost Studies of Pacific Bell’s System, Rulemaking on the Commission's Own Motion to Govern Open Access to Bottleneck Services and Establish a Framework for Network Architecture Development of Dominant Carrier Networks, D.98-02-106 (Feb. 19, 1998).
. Rulemaking on the Commission’s Own Motion to Govern Open Access to Bottleneck Services and Establish a Framework for Network Architecture Development of Dominant Carrier Networks, D.98-12-079 (Dec. 17, 1998).
. In that proceeding, the CPUC attempted to determine the costs that would be borne by Pacific (and other ILECs) in implementing the terms of the 1996 Act. These costs were recovered by a surcharge on telephone customers in California. See D.99-07-048, R.95-04-043 at 2.
. Interim Decision Setting Final Prices for Network Elements Offered by Pacific Bell, Rulemaking on the Commission’s Own Motion to Govern Open Access to Bottleneck Services and Establish a Framework for Network Architecture Development of Dominant Carrier Networks, D.99-11-050 (Nov. 18, 1999).
. Other markups were offered by other interested parties not a part of this appeal and not relevant here.