United States v. SunGard Data Systems, Inc.

REDACTED MEMORANDUM OPINION HUVELLE, District Judge. Plaintiff, the United States of America, has filed suit pursuant to Section 15 of the Clayton Act, 15 U.S.C. § 25, to enjoin Sun-Gard Data Systems, Inc. (“SunGard”) from acquiring the disaster recovery solutions *174 assets of Comdisco, Inc. (“Comdisco”) on the grounds that the acquisition would substantially lessen competition in the market for shared hotsite disaster recovery services for mainframe and midrange computers, in violation of Section 7 of the Clayton Act, 15 U.S.C. § 18. Defendants have set forth a series of arguments in opposition to the government’s complaint. First, defendants dispute the government’s narrow definition of the relevant product market. Contending that the appropriate market is not simply shared hotsite services, but the entire continuum of disaster recovery services, defendants assert that the proposed acquisition will not substantially lessen competition in this broader product market. Second, defendants argue that, even assuming that the government has established a presumption that the acquisition would violate the Clayton Act, the transaction will not actually produce an anti-competitive effect. Defendants offer six theories in support of this argument: 1) plaintiffs statistics regarding the product market are unreliable because they do not reflect the rapidly changing technologies in the disaster recovery industry; 2) price discrimination would not be profitable for the new entity under a Critical Loss analysis, because of the risk of losing too many existing customers; 3) SunGard and Comdisco are not in competition for the vast majority of customers, and therefore, then merger will have only a minimal anti-competitive effect; 4) there are a host of domestic and foreign companies that are poised to enter the market due to the low barriers to entry; 5) defendants’ knowledgeable and sophisticated customers would impede the exercise of market power by the new entity; and 6) the efficiencies resulting from the transaction will actually cause prices to drop and service to improve. The proposed acquisition has been postponed by agreement of the parties pending the Court’s decision. After thorough consideration of the parties’ briefs; the exhibits, testimony, and arguments presented by the parties at an expedited trial on November 8 and 9, 2001; and the proposed findings of fact and conclusions of law submitted by the parties; and for the reasons set forth herein, the Court will deny the plaintiffs request for permanent injunctive relief. This Memorandum Opinion constitutes the Court’s findings of fact and conclusions of law. BACKGROUND I. The Disaster Recovery Industry To state the obvious, today’s society relies extensively on computer systems in order to function effectively. Many entities, both private and public, run applications on IBM mainframe and other high-end computing platforms because the operations they perform require the high levels of performance and reliability provided by these systems. The functions that these computer platforms typically perform include processing and storing transaction information, maintaining customer accounts, controlling production resources, inventory and shipping, and maintaining financial and administrative records. As the tragic events of September 11, 2001 demonstrate, the possibility of a disaster that destroys or disables an entity’s computer capacity is a contingency for which a prudent business should be prepared. For a computer system, however, disasters need not be on the grand scale of a terrorist attack, fire, or earthquake to cause serious disruptions; network, hardware, or operational failures can also irreparably damage a company’s computer system. Because of the essential role that computer applications play in the operation of any business, many companies have *175 come to rely on a disaster recovery plan to reduce the potentially devastating impact of a disaster on their computer system. Disaster recovery vendors fill this need by selling a variety of services that enable the restoration of computer applications at another location if a natural disaster, major power outage, or other event causes their customer’s primary data centers to become unavailable. Computer applications vary both in terms of the types of computer platforms on which they run 1 and the degree to which they are mission-critical. In particular, these applications have different “recovery time objectives” (“RTOs”). Some applications are so critical that they require virtually instantaneous recovery; for some it is sufficient that they be restored within a few days; and for others restoration within a week or more will suffice. Accordingly, different types of disaster recovery services exist to meet these varying needs. A. Hotsite Services 1. Shared Hotsites Shared hotsite services are a widely-used disaster recovery system sold by vendors to companies that depend on mainframes and other high-end platforms. Because hotsites are shared by multiple clients, they provide cost-effective disaster recovery for large companies. 2 Three vendors in North America provide the vast majority of shared hotsite services to companies that use large-scale mainframe and midrange data processing centers' — defendants SunGard and Comdisco, and IBM. Approximately 7,500 North American customers currently use external shared hot-site services provided by defendants. {See Def. Ex. 151, Declaration of James Simmons (“Simmons Decl.”) ¶ 13; Gov. Ex. 106, Sullivan Dep. at 74.) Shared hotsite services are remote facilities that have a wide variety of computer systems and communication facilities that are needed for a client to recover its business applications should its own data center become unavailable. A shared hotsite enables a customer to replicate its own computer center at a separate location, thereby avoiding the risk that the hotsite will also be disabled by a regional disaster. Most shared hotsites service business applications with RTOs ranging from 24 to 96 hours. Shared hotsite services rely on back-up tapes of a client’s data center, which are recorded and maintained at a third location. When disaster strikes a .client’s data center, those back-up tapes are taken from the third location to the hotsite. Once delivered, the client’s personnel and hot-site technicians load the software onto the computers at the hotsite, transfer the back-up tapes onto the hotsite’s computer storage systems, and commence operations. This process generally requires between 24 and 96 hours, and hotsite vendors typically allow customers to use the facility for up to six weeks. 3 *176 Because only a small number of customers are likely to experience a disaster at any given time, hotsite vendors sell the same physical assets and services to many customers, which “share” the hotsite, thereby reducing the cost. If multiple simultaneous disasters occur, hotsite vendors either make their facilities available on a first-come, first-served basis, or they allocate their capacity among customers. Clients can also use alternate hotsites run by the same vendor if that customer’s primary hotsite is occupied. 2. Internal Hotsites An internal hotsite performs the same functions and satisfies the same RTO as a shared hotsite, but the hotsite facility is owned by the company itself, rather than by an external vendor. This solution typically involves a company sending daily back-up tapes to a remote location— whether to a second office in another city, or to a computer records management company, such as Iron Mountain, which stores about 100 million reels of magnetic tape and has approximately 40 percent of the outsourced records management business in the United States. (Def.Ex. 96.) Internal hotsites require a business to own a second computer processing system, although this can be part of a business’ infrastructure. B. Other Disaster Recovery Services The parties have identified five other forms of business continuity services— quick-ship services, coldsites, work area recovery, mobile hotsite recovery, and high availability services (also known as dedicated recovery services) — that are relevant to this proceeding. 1. Quick-Ship Services Quick-ship services ship computer equipment to locations designated by the customer within a specified time, but do not set up or provide support for the equipment, instead, the client retains these responsibilities. RTOs for quick-ship services vary dramatically, depending on the vendor and computer equipment: the Hewlett-Packard Company (“HP”) has a policy of quick-shipping systems within six hours (Def. Ex. 143, Deposition of Brian Fowler (“Fowler Dep.”) at 25), while quick-shipment of large systems may take up to one week. Plaintiff acknowledges that some small midrange computers can be shipped and installed to meet the same RTO as a shared hotsite. (Complaint ¶ 17.) 2. Coldsites Coldsites are computer-ready facilities that contain the temperature control and communication links suitable for a data center, but do not contain any computer hardware. Unlike hotsites, coldsites require clients to supply, install, and configure the computer equipment necessary to replicate their data center. The process of shipping and configuring the necessary computer systems is time-consuming. As a result, coldsites are often used by customers for long-term data recovery after the initial six-week, post-disaster hotsite period. 3. Work Area Recovery A work area recovery system is a mobile or fixed-location facility in which employee workstations are configured with desktop computers and local area networks, thus enabling employees to continue basic business operations in the event of a disaster or an interruption of business. Because the work areas do not contain the large computers or communications networks required to replicate or rebuild a large data center, however, they serve a different function than hotsites. *177 4. Mobile Hotsite Recovery Mobile hotsites are trailers configured for use as small data centers. Because of their size limitations, mobile hotsites cannot be used as a substitute for many shared hotsite disaster recovery services. 5. High Availability/Dedicated Recovery Services High availability services enable customers to reduce their recovery time to less than twenty-four hours, and in some cases, the recovery time can be instantaneous. Defendants’ expert witness, Michael Keat-ing, the Business Continuity Planning Leader for the Midwest Region of Marsh Risk Consulting, testified that there are essentially three types of high availability solutions: 1) “data vaulting, where rather than backup daily on magnetic tape, backups are done directly to dedicated hard disk at an internally controlled or commercial hotsite; [2) ] data logging, where data is transferred over telecommunications lines at periodic (generally hourly) increments; and [3) ] data mirroring, where dual sets of data exist in real time at separate locations.” (Gov. Ex. 277, Report of Michael Keating (“Keating Report”) ¶ 34.) In order to provide such rapid back-up functions, high availability solutions require hard disks or data processing equipment that is dedicated to a single client. As a result, high availability systems are much more expensive than shared hotsite services. For instance, Paul Sullivan, a Senior Vice President at Comdisco, testified that high availability approaches cost at least 50 percent more than shared hotsites, and could be priced up to ten times more than shared hotsites, depending on the particular dedicated solution that is selected. (Gov. Ex. 106, Sullivan Dep. at 19.) Dedicated processing recovery solutions may be internal systems run entirely by the business, or they may be provided by a commercial vendor. Both alternatives can be expensive. For an external solution, the cost saving that comes from sharing a hotsite with other companies is eliminated since the remote location is dedicated to a single customer. This difference in price can be dramatic because of the number of companies that typically share an external hotsite. (See Gov. Ex. 58, at M00072 (“SunGard believes it has the lowest ratio of subscribers to Midrange hotsites in the industry. Right now an approximate 22:1 ratio exists (average 22 clients per hot-site).”).) Internal high availability is also extremely expensive, as SunGard’s own documents emphasize. (See, e.g., Gov. Ex. 62 (discussing the high costs of hardware and communications for dedicated disaster recovery).) The Keating Report suggests, however, that depending on a customer’s needs, the downtime and associated costs experienced by a company may offset the expense of installing and contracting for a dedicated availability solution. (Gov. Ex. 277, Keating Report Tab 7.) II. The Parties Defendant SunGard is a Delaware corporation with its principal place of business in Wayne, Pennsylvania. SunGard is a major supplier of information technology, including investment support systems, trade processing, risk and asset management, and disaster recovery services. SunGard is a large provider of shared hotsite disaster recovery services, earning approximately $340 million in revenue in North America from this business in 2000. (Def.Ex. 169.) Unlike Comdisco, Sun-Gard’s North American revenue derives primarily from shared hotsite disaster recovery services; SunGard earned only [redacted] million from high availability services in 2000. (Def.Ex. 169.) According to SunGard’s market analysis (see Def. Ex. 151, Simmons Decl. ¶ 14), the demand for high availability services is rapidly increas *178 ing, especially in light of the events of September 11, 2001, and as defense counsel admitted during argument (Nov. 9 Trial Transcript at 17), it is for this reason that SunGard is anxious to acquire Com-disco, which already has a substantial presence in the high availability market. 4 In contrast to Comdisco, SunGard services more companies with small and mi-drange systems. According to James Simmons, the Chief Executive Officer of SunGard Recovery Services LP, SunGard has 2,000 customers that pay monthly fees to SunGard of less than $1,000 and more than 4,000 customers (out of some 5,500 total customers) that pay monthly fees of under $10,000. (Def. Ex. 151, Simmons Decl. ¶ 66.) There has also been a noticeable shift in its business from mainframe to midrange systems. For instance, in 1998 SunGard’s mainframe hotsite revenue was $105 million, or 41 percent of Sun-Gard’s North American availability solutions revenue. (Def.Ex. 169.) By 2000 this figure had dropped to [redacted] million, or 30 percent of SunGard’s North American availability solutions revenue. (Id.) In contrast, in 1998 SunGard’s mi-drange hotsite revenue was $147 million, representing 57 percent of SunGard’s North American availability solutions revenue, but by 2000 this had increased to [redacted] million, or 68 percent. (Id.) Only 887 SunGard customers are mainframe users. (Gov.Ex. 99.) Defendant Comdisco is a Delaware corporation headquartered in Rosemont, Illinois. Comdisco is also a major supplier of computer services, including shared hotsite systems and high availability services. Comdisco’s shared hotsite business generated approximately [redacted] million in revenue in North America in 2000. (Def.Ex. 170.) Unlike SunGard, a greater proportion of Comdisco’s business involves shared hotsite support for mainframe systems, rather than smaller processors. Over the past four years, approximately one-half of Comdisco’s revenue has come from shared hotsites for mainframes, and in 2000, 45 percent of its North American hotsite revenue ([redacted] million) was attributable to mainframe computers. (Id.) Comdisco’s hotsite customers number approximately 2,200, as compared to Sun-Gard’s 5,500 (Gov. Ex. 106, Sullivan Dep. at 74; Def. Ex. 151, Simmons Decl. ¶ 13), and 640 of these customers are mainframe users. (Gov.Ex. 101.) Nonetheless, it is readily apparent from the testimony of customers and Comdisco executives, as well as the internal documents of both companies, that the client bases of SunGard and Comdisco substantially overlap, and that the two companies regularly compete, along with IBM, with respect to new and existing customers. (See, e.g., Gov. Ex. 2.) III. Procedural History On July 15, 2001, HP entered into an acquisition agreement to purchase substantially all of the assets of Comdisco’s disaster recovery business for approximately $610 million. The next day, Com-disco filed a voluntary Chapter 11 bankruptcy petition with the U.S. Bankruptcy Court for the Northern District of Illinois, Eastern Division (Case No. 01-24795). Pursuant to an August 9, 2001 Order of the Bankruptcy Court, Comdisco’s assets were to be sold at an auction on October 11, 2001. Two bids were submitted at the auction. SunGard bid $825 million, and HP bid $700 million. SunGard was therefore selected as the successful bidder, and its acquisition *179 proposal was to be presented to the Bankruptcy Court for approval at the Sale Hearing on October 23, 2001. The day before the sale was to take place, the Antitrust Division of the United States Department of Justice filed its complaint in this Court, and moved for a temporary restraining order to prohibit defendants from consummating the proposed transaction. On the morning of October 23 — -just minutes before the Bankruptcy Court was to approve the acquisition — this Court entered a stipulated order by which the parties agreed to preserve the status quo until the earlier of 1) the Court’s ruling on plaintiffs request for permanent injunctive relief or 2) November 15, 2001. 5 All parties agreed, pursuant to Fed.R.Civ.P. 65(a)(2), to consolidate plaintiffs motion for a preliminary injunction and its request for permanent injunctive relief, and to have a trial on the merits before this Court. In addition, the parties proposed an “expedited” discovery and briefing-schedule: defendants answered the complaint five days after it was filed, all parties provided the reports of their experts one week after the filing of the complaint, fact discovery closed eleven days after the suit was instituted, and proposed findings of fact and conclusions of law were filed just two weeks after the filing of the complaint. 6 On November 8, the Court held a ten-hour evidentiary hearing. At the hearing, the United States called one expert witness, Peter Bronsteen, the President and Founder of the Princeton Economics Group, Inc. and a former in-house antitrust economist at Skadden, Arps, Slate, Meagher & Flom. 7 The defendants offered testimony from two expert witnesses, Barry C. Harris, the former chief economist in the Antitrust Division of the Department of Justice, and an industry expert in disaster recovery, Michael Keating. In addition to these three live witnesses, plaintiff submitted the declarations of eight fact witnesses, 8 and defendants filed declarations from five fact witnesses. The Court also received numerous exhibits — 285 from the government and 172 from defendants — that had been culled from the many boxes of documents that had been produced during discovery. On the morning of November 9, the Court heard closing arguments of counsel. It has now been only slightly more than three weeks since the initial complaint and motion for a TRO were filed in this action. During this extraordinarily brief period of time, the parties have completed the entire litigation process involving complicated legal issues and a highly sophisticated and technical industry. Despite the extremely expedited nature of the process, both sides submitted thorough and well-crafted pleadings and have assisted the Court at every juncture. The Court wishes to commend all parties for their professionalism, *180 and in particular, the Court is appreciative of the excellent work done by all counsel at the trial last week. LEGAL ANALYSIS I. Standard of Review Section 7 of the Clayton Act, 15 U.S.C. § 18, prohibits a corporation from acquiring “the whole or any part of the assets of another [corporation] engaged also in commerce or in any activity affecting commerce, where in any line of commerce or in any activity affecting commerce in any section of the country, the effect of such acquisition may be substantially to lessen competition, or to tend to create a monopoly.” The United States is authorized by Section 15 of the Clayton Act to seek an injunction to block a pending acquisition. 15 U.S.C. § 25. The United States has the ultimate burden of proving a Section 7 violation by a preponderance of the evidence. To establish a Section 7 violation, plaintiff must show that a pending acquisition is reasonably likely to cause anticom-petitive effects. See United States v. Penn-Olin Chem. Co., 378 U.S. 158, 171, 84 S.Ct. 1710, 12 L.Ed.2d 775 (1964) (noting that a Section 7 violation is established when “the ‘reasonable likelihood’ of a substantial lessening of competition in the relevant market is shown”). “The ‘Congress used the words “may be substantially to lessen competition” (emphasis supplied), to indicate that its concern was with probabilities, not certainties.’ ” FTC v. H.J. Heinz Co., 246 F.3d 708, 713 (D.C.Cir.2001) (quoting Brown Shoe Co. v. United States, 370 U.S. 294, 323, 82 S.Ct. 1502, 8 L.Ed.2d 510 (1962)). “Section 7 does not require proof that a merger or other acquisition [will] cause higher prices in the affected market. All that is necessary is that the merger create an appreciable danger of such consequences in the future.” Hospital Corp. of Am. v. FTC, 807 F.2d 1381, 1389 (7th Cir.1986). As this Circuit explained in Heinz, 246 F.3d at 715, the decision in United States v. Baker Hughes Inc., 908 F.2d 981 (D.C.Cir.1990), sets forth the analytical approach for establishing a Section 7 violation. “The basic outline of a section 7 horizontal acquisition case is familiar. By showing that a transaction will lead to undue concentration in the market for a particular product in a particular geographic area, the government establishes a presumption that the transaction will substantially lessen competition.” Baker Hughes, 908 F.2d at 982; see also United States v. Citizens & Southern Nat’l Bank, 422 U.S. 86, 120-22, 95 S.Ct. 2099, 45 L.Ed.2d 41 (1975). In other words, the government establishes a prima facie case of a Section 7 violation by “showing] that the merger would produce ‘a firm controlling an undue percentage share of the relevant market, and [would] result [ ] in a significant increase in the concentration of firms in that market.’ ” Heinz, 246 F.3d at 715 (quoting United States v. Philadelphia Nat’l Bank, 374 U.S. 321, 363, 83 S.Ct. 1715, 10 L.Ed.2d 915 (1963)). “Such a showing establishes a ‘presumption’ that the merger will substantially lessen competition.” Id.; see Baker Hughes, 908 F.2d at 982. To rebut this presumption, defendant must “show that the market-share statistics give an inaccurate account of the merger’s probable effects on competition in the relevant market.” Heinz, 246 F.3d at 715 (internal quotation omitted). “ ‘If the defendant successfully rebuts the presumption [of illegality], the burden of producing additional evidence of anticom-petitive effect shifts to the government, and merges with the ultimate burden of persuasion, which remains with the government at all times.’ ” Id. (quoting Baker Hughes, 908 F.2d at 983). *181 An application of the burden-shifting approach requires the Court to determine (1) the “line of commerce” or product market in which to assess the transaction; (2) the “section of the country” or geographic market in which to assess the transaction; 9 and (3) the transaction’s probable effect on competition in the product and geographic markets. See United States v. Marine Bancorporation, Inc., 418 U.S. 602, 618-23, 94 S.Ct. 2856, 41 L.Ed.2d 978 (1974); FTC v. Harbour Group Investments, L.P., 1990 WL 198819, 1990-2 Trade Cas. (CCH) ¶ 69,247, at 64,914 n.3 (D.D.C.1990); see also FTC v. Swedish Match, 131 F.Supp.2d 151, 156 (D.D.C.2000); FTC v. Cardinal Health, Inc., 12 F.Supp.2d 34, 45 (D.D.C.1998); FTC v. Staples, Inc., 970 F.Supp. 1066, 1072 (D.D.C.1997). II. Prima Facie Case As noted, the first step in determining whether a transaction will substantially lessen competition is to define the relevant product market. Swedish Match, 131 F.Supp.2d at 156; see Brown Shoe, 370 U.S. at 324, 82 S.Ct. 1502. A. Relevant Product Market “Defining the relevant market is critical in an antitrust case because the legality of the proposed mergers in question almost always depends upon the market power of the parties involved.” Cardinal Health, 12 F.Supp.2d at 45. Not only is the proper definition of the relevant product market the first step in this case, it is also the key to the ultimate resolution of this type of case, since the scope of the market will necessarily impact any analysis of the anticompetitive effects of the transaction. See, e.g., Swedish Match, 131 F.Supp.2d at 156; Staples, 970 F.Supp. at 1073. Plaintiff carries the burdens of proof and persuasion regarding market definition. Baker Hughes, 908 F.2d at 982-83; United States v. Engelhard Corp., 970 F.Supp. 1463, 1466 (M.D.Ga.1997), aff'd, 126 F.3d 1302 (11th Cir.1997). Plaintiff argues that the appropriate product market is the provision of shared hotsite services for customers with mainframe and midrange computer processing centers. 10 This market includes services for customers with mainframe and mi-drange data centers and equipment, but excludes customers that seek hotsite services for small computer systems. 11 According to the government, under this view of the market, the proposed acquisition would create a duopoly in which the merged firm would control approximately 71 percent of the market, and IBM would control almost all of the remainder. (Gov. Ex. 257, Report of Peter Bronsteen (“Bronsteen Report”) at Ex. 4.) Defendants advocate a much broader definition of the relevant product market that would encompass a number of other disaster recovery systems — -most notably high availability solutions, internal hotsites, and quick-ship services. *182 1. Legal Standard The Supreme Court has articulated the general rule for determining a relevant product market: “The outer boundaries of a product market are determined by the reasonable interchangeability of use [by consumers] or the cross-elasticity of demand between the product itself and substitutes for it.” Brown Shoe, 370 U.S. at 325, 82 S.Ct. 1502; see also United States v. E.I. du Pont de Nemours & Co., 351 U.S. 377, 395, 76 S.Ct. 994, 100 L.Ed. 1264 (1956). “Interchangeability of use and cross-elasticity of demand look to [1) ] the availability of products that are similar in character or use to the product in question and [2) ] the degree to which buyers are willing to substitute those similar products for the product.” Swedish Match, 131 F.Supp.2d at 157; see E.I. du Pont de Nemours, 351 U.S. at 393, 76 S.Ct. 994; see also Hayden Pub. Co. v. Cox Broadcasting Corp., 730 F.2d 64, 71 (2d Cir.1984) (framing the question as whether two products “can be used for the same purpose,” and, if so, whether and to what extent purchasers are willing “to substitute one for the other”). The 1992 U.S. Department of Justice and Federal Trade Commission’s Horizontal Merger Guidelines (1997 rev.) [hereinafter Merger Guidelines] incorporate this same approach. The Merger Guidelines take the smallest possible group of competing products and ask whether “a hypothetical monopolist over that group of products would profitably impose at least a ‘small but significant and nontransitory’ [price] increase [(“SSNIP”)].” Merger Guidelines § 1.11. This SSNIP is usually an “increase of five percent lasting for the foreseeable future.” Id. This methodology has been adopted by the courts, as well as the parties in this action. See, e.g., Swedish Match, 131 F.Supp.2d at 160; Staples, 970 F.Supp. at 1076. In addition to the cross-elasticity of demand and supply, the Supreme Court in Brown Shoe articulated several “practical indicia” as guidelines for defining the relevant market. These include “industry or public recognition of the submarket as a separate economic entity, the product’s peculiar characteristics and uses, unique production facilities, distinct consumers, distinct prices, sensitivity to price changes, and specialized vendors.” 370 U.S. at 325, 82 S.Ct. 1502. As observed by one court in this Circuit, “the determination of the relevant market in the end is ‘a matter of business reality — [] how the market is perceived by those who strive for profit in it.’ ” Cardinal Health, 12 F.Supp.2d at 46 (quoting FTC v. Coca-Cola Co., 641 F.Supp. 1128, 1132 (D.D.C.1986), vacated as moot, 829 F.2d 191 (D.C.Cir.1987)); see Rothery Storage & Van Co. v. Atlas Van Lines, Inc., 792 F.2d 210, 219 n. 4 (D.C.Cir.1986) (“The industry or public recognition of the submarket as a separate economic unit matters because we assume that economic actors usually have accurate perceptions of economic realities.”). Based on a thorough review of the law and the evidence, the Court finds that the government has not met its burden of establishing that the relevant product market is limited to shared hotsite services for mainframe and midrange computer processing centers. Two factors compel this conclusion. First, the evidence presented demonstrates that the government’s market contains an extremely heterogeneous group of customers, particularly in terms of their needs and their computer equipment. Given the rapid changes in computer capabilities and the reduced costs of both hardware and communications, the evidence does not permit the exclusion of either internal hotsites or quick-ship services from any market that includes shared hotsites. Second, the striking het *183 erogeneity of the market, particularly as reflected by the conflicting evidence relating to customer perceptions and practices, further undercuts plaintiffs product market definition. The difficulty in attempting to decipher any conclusions about defendants’ approximately 7,500 customers was obviously exacerbated by the abbreviated discovery schedule in this case. For instance, several customers who were interviewed by one party then changed their position when interviewed by the opposing party. (Compare Gov. Ex. 234 ivith Def. Ex. 3 (Ciuzio Statements); Gov. Ex. 236 vñth Def. Ex. 3 (Mobley Statements); Gov. Ex. HOB with Def. Ex. 3 (Wade Statements); and Gov. Ex. 110K with Def. Ex. 3 (McMiehael Statements).) Customer responses were also often vague and confused, perhaps due to the complexity of the issues and the difficulty in framing specific questions regarding the financial viability of switching from an external hotsite service to an internal solution. In particular, it was consistently unclear what definition of “internal” hotsite or “internal” solution was being discussed, so one could not know if a customer was comparing an external hotsite solution to a high availability solution or to an internal hotsite solution. Based on this equivocal evidence, the Court is unable to determine whether a SSNIP would cause “a significant number of users” to switch to alternative forms of disaster recovery, FTC v. Owens-Illinois, Inc., 681 F.Supp. 27, 36 (D.D.C.1988), and as a result, plaintiff cannot sustain its burden. 2. Product Lines in the Relevant Market As noted, defendants argue that the relevant product market is the entire continuum of disaster recovery services, including coldsites, mobile and work area recovery, quick-ship services, shared hot-sites, internal hotsites, and high availability solutions' — both internal and external. Plaintiff focuses on just a portion of the continuum — shared hotsites for mainframe and midrange systems. The Court agrees that a relevant product market may be defined as a submarket within a larger continuum, see United States v. Gillette Co., 828 F.Supp. 78, 83 (D.D.C.1993) (finding the relevant product market to be “all premium writing instruments ... with SRPs from $40 to $400”), and based on the evidence, it finds that plaintiff has proven that coldsites, work area and mobile recovery services, and high availability systems should be excluded from the relevant market. Plaintiff has failed to provide sufficient evidence, however, to justify the exclusion of quick-ship services and internal hotsites from the market. a. Coldsites, Work Area Recovery, and Mobile Recovery Although they suggest that the product market includes the entire business continuity/disaster recovery (“BC/DR”) continuum, defendants have failed to cast any doubt on plaintiffs initial contention that coldsites, work area recovery, and mobile recovery services are not substitutes for shared hotsites. Coldsites are often used as a complement to shared hotsites, and typically have an RTO greater than 72 hours. (Gov. Ex. 218, Declaration of David Krohmal (“Krohmal Deck”) ¶ 13; Gov. Ex. 219, Declaration of Thomas Carroll (“Carroll Deck”) ¶9; Gov. Ex. 267, Deposition of Barry Harris (“Harris Dep.”) at 59.) Mobile recovery systems have a similarly long RTO, and SunGard does not currently have a single contract for mobile recovery for a mainframe computer. (Gov. Ex. 272, Deposition of Jim Simmons (“Simmons Dep.”) at 76-77; Gov. Ex. 56, at 201630; Gov. Ex. 107, Deposition of John Jackson (“Jackson Dep”) at 16.) Work area recovery systems are used to provide work place recovery for employees, not data recovery, and are not config *184 ured to support mainframe and midrange computer systems. (Gov Ex. 219, Carroll Decl. ¶ 10.) The Court therefore finds that significant numbers of customers would not switch from shared hotsites to coldsites, mobile recovery, or work area recovery systems in response to a SSNIP for shared hotsites. b. High Availability Any high availability system that is a substitute for a shared hotsite must include data processing and not just data storage. Both external and internal high availability systems are significantly more expensive than shared hotsites. Even Comdisco executives note that the cost differential ranges from 50 percent more costly up to twenty times more expensive. (Gov. Ex. 106, Sullivan Dep. at 19 (“as little as 50% [more to] ... 5 to 10 times as costly”); Gov. Ex. 107, Jackson Dep. at 30-31 (“5 times to 20 times more expensive”); Gov. Ex. 218, Krohmal Decl. ¶ 17; Gov. Ex. 221, Declaration of Karl Ross (“Ross Decl.”) ¶ 11.) The evidence based on cost alone suggests that a shared hot-site customer would not switch to high availability in response to a SSNIP. (See Gov. Ex. 277, Keating Report at Tab 7 (noting that the hardware, software, communications, and miscellaneous costs of an internal high availability solution are approximately 14 times higher than those of a shared hotsite)). Moreover, the record before the Court demonstrates that those customers that do switch from shared hotsites to high availability often do so for reasons entirely unrelated to cost. In particular, customers switch to a dedicated recovery system because they require RTOs significantly faster than the 24- to 72-hour window provided by hotsites, better security for their data, and more control over their recovery system. (See, e.g., Gov. Ex. 106, Sullivan. Dep. at 44-45; Gov. Ex. 107, Jackson Dep. at 29.) Comdisco executives were unaware of any customer that had switched to high availability because of cost. (Gov. Ex. 106, Sullivan Dep. at 44-46; Gov. Ex. 107, Jackson Dep., at 10-13.) The Court is therefore persuaded that customers would not switch to high availability in response to a SSNIP of shared hot-sites, and that dedicated recovery solutions are not in the same product market as shared hotsites. c. Quick-Ship Services Quick-ship services are used for smaller systems, including midrange computers, but are not currently a viable alternative for mainframes, which are too large to ship quickly. 12 (Gov. Ex. 257, Bronsteen Report ¶ 18.) Plaintiff argues that, despite the fact that many customers do choose to quick-ship midrange systems, quick-ship is not a substitute for shared hotsites with regard to those systems. The Court finds that the evidence does not support this contention. Plaintiff has defined the product market to include shared hotsite services for both mainframe and midrange processors. The four most significant examples of the latter are the AS/400, RS/6000, HP 9000, and Sun processors. The record reflects that shared hotsite revenue for both SunGard and Comdisco from mainframe systems is declining, while proceeds from midrange systems are rising. (Def. Exs. 169-70 (SunGard revenue from midranges has climbed from $147 million in 1998, to $189 million in 1999, to [redacted] million in 2000, to a projected [redacted] million in 2001; Comdisco revenue from midranges has increased from 21 percent of overall *185 revenue in 1998, to 24 percent in 1999, 27 percent in 2000, and a projected 33 percent in 2001).) 13 Midrange servers already comprise the majority of the market for shared hotsite services — and this percentage is rising significantly each year. Plaintiffs expert conceded that customers could quick-ship AS/400s. (Def. Ex. 141, Deposition of Peter Bronsteen (“Bronsteen Dep.”) at 44-45; Nov. 8 Trial Transcript, Vol. I, at 22.) In 2000 Sun-Gard’s shared hotsite revenue from AS/ 400s approached $40 million. (Gov.Ex. 97.) SunGard also estimates that more than 50 percent of their customers have contracts that allow them to choose between quick-ship, mobile, and shared hotsite recovery at the time of a disaster, which indicates that at the time of contracting, quick-ship and hotsite services are interchangeable for many customers. (Def. Ex. 151, Simmons Decl. ¶ 12.) The unrebutted testimony of defendants’ industry expert also indicated that both AS/400 and RS/6000 systems can be quick-shipped. (Nov. 8 Trial Transcript, Vol. II, at 33.) Keating testified, for example, that a used AS/400 can cost as little as $1,500, and that a used RS/6000 can cost as little as $10,000. (Id. at 48-49.) In short, the record demonstrates that quick-ship service is a viable substitute for a shared hotsite for at least some customers with midrange systems, and as noted, midrange systems comprise a substantial segment of the government’s proposed product market both in terms of revenue and numbers of customers. In response, plaintiff argues that a quick-ship service generally has a longer recovery window than a hotsite service, and that at least according to the testimony of one SunGard customer (see Gov. Ex. 273, Deposition of Douglas Varner (“Var-ner Dep.”) at 110-11), a quick-ship solution “would not work in our environment” because of the length of time it would take to get the system up and running. Based on this rather scant evidence, the government asks the Court to conclude that quick-ship is not a viable option for defendants’ customers. This cannot be done given defendants’ evidence regarding the viability of quick-shipping midrange computers, especially entry-level systems. (Nov. 8 Trial Transcript, Vol. I, at 22.) In particular, it is clearly possible to quick-ship the less expensive midrange computer systems, as opposed to mainframe computers or large midrange systems. While we know that a significant portion of the government’s product market includes midrange computers, it is impossible to extrapolate from the data how much of the market represents entry-level midrange systems. In its proposed findings of fact, the government points to Douglas Varner and to one other witness who also testified that quick-ship services were not an acceptable option. But both witnesses cited examples that involved companies with multiplat-form systems, including mainframes. (Gov. Proposed Findings of Fact ¶¶ 28-29 (citing Gov. Exs. 244, 273.)) Moreover, plaintiff faults the defendants for their inability to quantify the number of firms for which quick-ship is a good alternative to shared hotsites. (Gov. Proposed Findings of Fact ¶ 29.) However, any gap in the evidence is a flaw in plaintiff’s case — not defendants’. For instance, plaintiffs expert did not know how many of SunGard’s customers had a single UNIX-based server — for which a quick-ship would likely be a satisfactory substitute for a shared hot-site. (Nov. 8 Trial Transcript, Vol. I, at 33.) Plaintiff also failed to offer any evidence to suggest that the cost differential between quick-ship and an internal hotsite *186 would deter a customer from switching in response to a 5- to 10-percent increase in the price of' a hotsite. Instead, when asked about the price of these options, plaintiffs counsel candidly admitted that “[i]t varies depending on the equipment being recovered, as does quick-ship depend on what’s being recovered.” (Nov. 9 Trial Transcript at 79.) Since the evidence does not permit one to infer that quick-ship services are not an acceptable substitute for at least some mi-drange servers, the Court is unable to exclude them from the relevant product market definition. d. Internal Hotsites Plaintiff also contends that internal hot-sites do not fall within the same relevant product market as shared hotsites, because not “enough” customers would switch to internal hotsites in response to a SSNIP of shared hotsites. (Gov. Proposed Findings of Fact ¶ 55.) Here again, the Court finds that plaintiff has not offered sufficient evidence to satisfy its burden. As a matter of law, “[cjourts have generally recognized that when a customer can replace the services of [an external product] with an internally-created [] system, this ‘captive output’ (i.e. the self-production of all or part of the relevant product) should be included in the same market.” Cardinal Health, 12 F.Supp.2d at 48; see also Spectrofuge Corp. v. Beckman Instruments, Inc., 575 F.2d 256, 278 (5th Cir.1978); United States v. ALCOA, 148 F.2d 416, 424 (2d Cir.1945); IIA Areeda et al., Antitrust Law ¶ 535e (1995). 14 Under the Merger Guidelines, captive production can be considered to the extent that “such inclusion reflects [its] competitive significance in the relevant market prior to the merger.” Merger Guidelines § 1.31. In response, plaintiff argues that an internal alternative should not be included in its definition of a purely external market, noting that in Heinz, the Court did not consider homemade baby food within the market that was defined as “jarred baby food.” Heinz, however, is inapposite, because the parties stipulated to the relevant product market. FTC v. H.J. Heinz Co., 116 F.Supp.2d 190, 195 (D.D.C.2000), rev’d on other grounds, 246 F.3d 708 (D.C.Cir.2001). 15 Similarly, plaintiffs expert — while *187 conceding that internal hotsites are “a good alternative” for shared hotsites — contends that they should not be included in the relevant market because “when you go to assess the likely competitive effects of the merger, you’d find that [the internal hotsite users] are irrelevant because they don’t have the capacity to expand and serve other people. Therefore, they don’t constrain the prices charged by the merging firms.” (Nov. 8 Trial Transcript, Vol. I, at 92-93.) This analysis, however, misconstrues the Merger Guidelines and the case law: what is significant is not whether the companies that currently use internal solutions have the capacity to enter the market as vendors for others, but whether the customers that currently use shared hotsites would switch to an internal hotsite in response to a SSNIP. Defendants contend that if the product market is broadened to include both external and internal hotsites for mainframe and midrange computers, SunGard’s market share drops to 20 percent, Comdisco’s falls to 15 percent, and IBM’s sinks to 13 percent. (Nov. 9 Trial Transcript at 20-21.) According to defendants’ statistics, internal hotsites — excluding systems with dedicated storage and including only those that have an RTO of 24 hours or greater— account for 50 percent of this hotsite market. Id. at 24. This is, admittedly, an imprecise estimate of the internal hotsite market based on a rough extrapolation from limited data. For example, defendants rely on a Wall Street Journal article which notes that half of Iron Mountain’s 94 customers that were affected by the September 11 disaster use internal back-up systems, while the other half use recovery solutions operated by SunGard, Comdisco, and IBM. (Def.Ex. 96.) Defendants also rely on a survey by CIO, a leading research firm in the field of disaster recovery, which revealed that 50.9 percent of firms either currently have or intend to develop “[o]ffsite locations, immediately ready for use” in case of disaster. (Def. Ex. 127, at 4.) As plaintiff notes, however, this was a survey of only 258 companies which were self-selected — i.e., it included only businesses that took the initiative to respond to an on-line poll. (Nov. 9 Trial Transcript at 97-98.) Moreover, the 50.9 percent includes an unidentified number of firms that merely “intend” to establish internal disaster recovery systems, and thus, the percentage that currently use such a solution is not identified. Defendants’ industry expert hypothesized that 50 percent of companies with disaster recovery systems managed them internally — but his support for that estimate was also the statistically-insignificant CIO survey. (Gov. Ex. 277, Keating Report ¶ 33.) The research of defendants’ expert economist revealed that of the Fortune 1000 companies, only 360 use SunGard or Comdisco for BC/DR — leading to the conclusion that some portion of the remaining 640 companies manage their recovery solutions internally. (Gov. Ex. 282, Harris Report at Tab 8.) However, this rather crude comparison does not identify the number of companies that have an internal hotsite system, as opposed to some other type of internal recovery system. Yet, even if one cannot quantify the extent of the internal hotsite market, it is clear from the record that internal hotsites do, in fact, compete with the external shared hotsite business. A November 6, 2001 internet publication written by a member of the computer industry notes that shared and internal hotsites are the “two practical alternatives” for most companies. (Def. Ex. 15, at 2.) A SunGard document states, “Our primary competitor across all product lines is the Internal Solution. To win the business our challenge is to provide a higher value (the combination of benefits and price) than companies can provide internally.” (Def. Ex. 5, at 2.) A Comdisco document notes, *188 “Our largest competitor remains the internal solution.” (Def. Ex. 101; see also Gov. Ex. 81 (SunGard estimates “based on research that ... 71% of [midrange] and server sites use an in-house solution.”); Def. Ex. 113, at 30,494 (HP reports that “[mjost of the competition emanates from the customers themselves.... For most customers, once the decision is taken to implement a recovery solution, the next step is to decide whether to go for an in-house solution or to use an outside source provider”).) To some extent, defendants’ statistics also bear this out: although the retention rates of both companies is approximately 95 percent, both SunGard and Comdisco lose more customers to internal solutions each year than to all other external vendors combined. (Def. Ex. 100; Gov. Ex. 107, Jackson Dep. at 39-42.) In 2000, for example, SunGard lost 89 customers to internal solutions; this year, the company projects losing 94 customers to internal recovery systems. Approximately half of the lost revenue came from clients with mainframe processors; the other half resulted from those with midrange systems. (Def.Ex. 100.) In fact, over the past three years, SunGard has lost more customers and more revenue each year to internal solutions than it did the prior year. (Def. Ex. 100; Gov. Ex. 281, Harris Report ¶ 20; Def. Ex. 151, Simmons Decl. ¶¶ 34-38.) SunGard has also lost more bids to internal solutions than to any other competitor. (Def. Ex. 151, Simmons Decl. ¶ 57; Gov. Ex. 281, Harris Report ¶ 24.) And defendants have provided approximately 68 letters from former SunGard clients that switched from a shared hotsite to an internal solution. (Def. Exs. 6-14; 16-74.) 16 It is also apparent from the evidence that in addition to those customers that have switched to an internal solution or that have, in the first instance, implemented an internal solution, there are also customers that have threatened to switch to an internal hotsite solution. (See, e.g., Def Ex. 109; Gov. Exs. 90, 91, 111E-N; Gov. Ex. 281, Harris Report ¶ 26; Gov. Ex. 273, Varner Dep. at 132-33; Gov. Ex. 107, Jackson Dep. at 52, 60, 95.) Thus, although plaintiff has frequently attempted to classify the product market as an oligopoly that the proposed acquisition would shrink to a duopoly, the record leaves little doubt that SunGard and Com-disco consider internal solutions, including internal hotsites, as their main competitive threat and that, in fact, there is increasing evidence that their perception is fully justified in view of the decreasing cost and changing nature of the technology. 17 See United States v. Microsoft Corp., 253 F.3d *189 34, 49 (D.C.Cir.2001) (“Rapid technoldgical change leads to markets in which firms compete through innovation for temporary market dominance, from which they may be displaced by the next wave of product advancements.”) (internal quotations omitted); FTC v. B.R. Donnelley & Sons Co., 1990 WL 193674, at *4 (D.D.C.1990) (finding that the market definition should be expanded because the ability of a substitute product to compete “will be enhanced in the future because of further technological and market developments”). Thus, the Brown Shoe factors — especially industry recognition and the peculiar characteristics and uses of the product — support a finding that internal hotsites fall within the same product market as shared hotsite services. In response, the government contends that internal hotsites are not a substitute for shared hotsite services because of their cost. In terms of a pure cost comparison, the Court is hard-pressed to reach any conclusion regarding the feasibility of a switch from external to internal hotsites given the scant data as to the relative cost of the two products. Although plaintiff points to several documents and the deposition testimony of Comdisco executives to argue that the cost of an internal hotsite is exponentially higher than that of shared hotsite services, some of these exhibits do not differentiate between internal high availability and an internal hotsite. (See, e.g., Gov. Ex. 107, Jackson Dep. at 30-31; Gov. Ex. 62, at 2.) 18 The government has also relied on a limited number of declarations from Comdisco customers to argue that a substantial number of customers cannot afford to switch in the face of a 5-to 10-percent price increase. (See, e.g., Gov._ Exs. 218-23, 227, 244). Again, a careful reading of these exhibits demonstrates that these customers have mixed-platform needs, including mainframes, and it is impossible to generalize from these statements with respect to the feasibility of other customers switching to an internal solution. Moreover, as noted above, defendants have introduced a similar number of customer statements, including documents from customers that have switched to an internal solution, to rebut plaintiffs showing. At best, this conflicting evidence defies categorization, but only highlights the difficulty in this case — any generalizations regarding customer behavior cannot be arrived at with any certainty, since it depends on a host of factors, including the type of equipment a customer must duplicate, the particular circumstances and needs of the customer, and in some cases, the size of the customer’s operations. Similarly, the government’s reliance on its Exhibit 58 is insufficient evidence, even when coupled with the Comdisco customer declarations and letters, to satisfy its burden. This SunGard document states that “[i]n general, the cost of maintaining a separate equipment resource exceeds the cost of a hotsite subscription by a factor of between 5 and 15, dependent upon the type of platform being utilized.” (Gov. Ex. 58, at M00069.) While this document cites a cost differential for equipment of somewhere between 5 and 15, it acknowledges that the cost will vary depending on the equipment, and since it is impossible to determine what segment or what proportion of the defendants’ customer base would be vulnerable to such a cost differential, 19 it is difficult to use this document to conclude, as argued by the government, *190 that a substantial number of customers will not switch to an internal hotsite solution if the price of an external system is increased by 5 to 10 percent. 20 3. Customer Base In order to determine the relevant market, the critical question for the Court is whether a hypothetical monopolist could profitably raise price. Merger Guidelines § 1.11. In other words, there must be a significant number of customers that will not switch to a substitute product in response to a SSNIP of shared hotsites. As plaintiff notes, “[e]ven if some segment of customers of a relevant product market could turn to alternative products in the face of a price increase, it does not immunize merging parties from Section 7 if there remain a substantial number of customers for whom there are no competitive alternatives.” (PL Reply to Def. Proposed Findings of Fact at 4.) Courts often determine whether this “substantial number of customérs” exists by analyzing a group of customers that is representative of the entire client base for the product. See Cardinal Health, 12 F.Supp.2d at 48-50 (prescription drug wholesalers constitute the relevant product market, even though certain customers could perform this service for themselves, because “a majority of [the defendants’] customers,” such as hospitals, independent pharmacies, and smaller retail chains, relied on that service); Owens-Illinois, 681 F.Supp. at 37 (“This analysis serves to confirm the conclusion that those few end-use segments proven to be inelastic are not significant enough, in and of themselves, to constitute a relevant product market and are not representative of the glass container market as a whole.”) Determining whether there is sufficient evidence to reach this conclusion regarding defendants’ 7,500 customers has been complicated given the conflicting evidence from the parties’ economists, as well as the conflicting customer statements submitted by the parties 21 *191 In Engelhard, the Eleventh Circuit found that the government had not established a prima facie case because it could not determine whether the customers who testified at trial were representative of the entire product market that the government sought to define. [I]t is possible for only a few customers who switch to alternatives to make the price increase unprofitable, thereby protecting a larger number of customers who would have acquiesced in higher [ ] prices. To evaluate such possibilities, the Government should have ascertained the size of the [product] market in its different end-use applications. However, the Government’s expert, Dr. Bo-disch, could not identify the number of companies using [the product] in many of its end-use applications. This undermines the Government’s entire case. No matter how many customers in each end-use industry the Government may have interviewed, those results cannot be predictive of the entire market if those customers are not representative of the market. Without knowing the size of the market, we cannot know if the customers interviewed are representative of that market. In short, under the circumstances of this case, evidence on the size of the [product] market was essential, and its absence casts a shadow over the reliability of all Dr. Bodisch’s conclusions. Engelhard, 126 F.3d at 1806 (internal citation omitted). The instant action presents these same difficulties. Although both parties have submitted numerous declarations and letters to bolster their respective positions, these exhibits do not allow any reliable conclusion as to whether a substantial number of shared hotsite customers would switch to a substitute disaster recovery service in response to a SSNIP. Plaintiffs case rests largely on ¶43 of the Bronsteen Report. “Even if there were some customers who would substitute [] internal solutions in'response to a small hotsite price increase, this would not change my conclusion that hotsite services are a relevant market. The evidence suggests strongly that too few customers would shift away from hotsite services to internal solutions to warrant their inclusion in the relevant market.” To support that assertion, the government has submitted more than 50 statements from shared hotsite customers that indicate that they could not switch in response to a SSNIP primarily because of cost concerns. {See, e.g., Gov. Exs. 110, 218, 219, 221-22, 226-28, 233-37.) In response, defendants have submitted their own flood of more than 90 statements from customers, attesting to their intention to change their form of disaster recovery if the price of shared hotsites were to go up. {See, e.g., Def. Exs. 3-4, 92-94,125,150.) 22 From this evidence, one can only surmise that there are some customers that cannot switch to an internal solution in response to a SSNIP of shared hotsites, nor can they make a credible threat to switch in order to keep the merged company from raising prices. The government *192 has failed, however, to show whether this captive group is substantial enough that a hypothetical monopolist would find it profitable to impose such an increase in price. The sampling of customer statements before the Court is minuscule when compared with the entire universe of defendants’ shared hotsite customers. Although the government has submitted approximately 50 statements from customers stating that they either would not or could not switch from shared hotsites, there are more than 7,500 customers that currently use defendants’ shared hotsites. Without, more information, the Court simply cannot determine whether these 50 declarations are representative of the shared hotsite client base. 23 In addition, the record does not indicate whether the customers cited by plaintiff are representative of the entire universe of shared hotsite clients, especially given the significant differences among customers in terms of their size, the equipment that they use, and their business needs. For example, absorbing the SSNIP might be the only solution for companies of a certain size — or within a certain industry — while other entities of a different size, or with different business needs and goals, may be able to switch products in response to a price increase. Similarly, plaintiff has not shown whether the customers it sampled are representative of the majority of customers in terms of the hardware they use. For instance, in light of the evidence that quick-ship is a substitute for many customers with midrange systems, plaintiff needed to prove that its sampled customers represent this midrange computer segment of the shared hotsite market with respect to equipment. Instead of fine-tuning its presentation to account for significant differences among defendants’ customers, the government lumped all customers together. As a result, this Court is unable to determine with any degree of certainty whether those companies that claim they would not switch in response to a SSNIP are representative of shared hot-site customers in terms of their business structure. See Cardinal Health, 12 F.Supp.2d at 46-49 (examination of the drug wholesale market by specifically focusing on the needs and abilities of the different groups of customers of drug wholesalers); Owens-Illinois, 681 F.Supp. at 36-46 (exhaustive survey of eleven end-use segments for the product in question to determine cross-elasticity). 24 In the absence of such a determination, the government’s definition of the relevant product *193 market cannot be sustained. See Engel-hard, 126 F.3d at 1306. 25 CONCLUSION While time does not permit a more exhaustive discussion of the volumes of evidence that have been presented during the past two weeks, the Court has attempted to review thoroughly all of the record and the extraordinarily well-done pleadings submitted by the parties. As is clear, the demand of some customers for shared hot-site services is inelastic. The Court cannot, however, find that this is a substantial number given the availability of quick-ship solutions — -.especially for some midrange users — and the rapidly increasing availability of internal hotsite solutions for certain types of customers, depending on their size, their needs, and the computer equipment that they use. In light of the decreasing costs of equipment and telecommunications and the rapidly evolving computer technology, the Court cannot accept the government’s overly narrow and static definition of the product market. The defendants’ customers, as well as their computer systems, are simply too varied and too dissimilar to support any generalizations. Therefore, the central premise of the government’s case — that there are “a substantial number of customers for whom there are no competitive alternatives” (Gov. Reply at 4) — has not been proven. Accordingly, this Court will not enjoin the proposed transaction. ORDER Upon consideration of the entire record before the Court, it is hereby ORDERED that plaintiffs request for a permanent injunction is DENIED; and it is FURTHER ORDERED that in view of the public’s interest in this case and the need to publish this Memorandum Opinion as soon as possible, the parties are to submit, by Monday, November 19, 2001, a joint request regarding the redactions to the Opinion which are necessary to protect the confidential information of the parties and any non-parties; and it is FURTHER ORDERED that the complaint is DISMISSED WITH PREJUDICE. SO ORDERED. Footnotes [1] A variety of computing platforms are currently used by North American businesses. Some of the most common are IBM mainframes, the IBM AS/400 and RS/6000, DEC VMX and UNIX, Hewlett-Packard/9000, NCR PRIMOS, Stratus VOS, Sun UNIX, and Uni-sys MCP, OS, and UNIX. (See Gov. Ex. 106, Deposition of Paul Sullivan ("Sullivan Dep.") at 24; Gov. Exs. 97, 102.) [2] Entities that currently use shared hotsites include, inter alia, the United States Customs Service, Blue Cross/Blue Shield, Cingular Wireless, Honda, Siemens, Walgreens, Duke Energy, Pillsbury, Coca-Cola, Xerox, and numerous financial institutions. (See Gov. Exs. 1-2, 220, 244.) [3] When a hotsite is not being used by a client, it is used for continuous testing of customer contingency plans. [4] In 2000, Comdisco had [redacted] million in revenue (or 23 percent of its total revenue) from its high availability sales. (Def.Ex. 170.) [5] On November 15, 2001, Comdisco will pay its annual employee bonuses. If the future of the company remains undecided at that time, defendants believe that many Comdisco employees will leave the firm, thus substantially reducing its value. The parties therefore requested that the case be decided by that date. [6] The discovery process was made even more challenging by the vast amount of confidential material that had to be obtained from a variety of sources in the disaster recovery industry. Yet, the parties and affected non-parties, working' with the able assistance of Magistrate Judge John M. Facciola, were able to resolve these issues in a timely and civil manner. [7] This firm is currently representing defendant Comdisco in this matter. [8] A ninth fact witness, Peter MacLean of Guardian iT pic, a European business-continuity company, was omitted because the parties were unable to locate Mr. MacLean, who was on vacation, in order to depose him. [9] It is undisputed that the relevant geographic market is North America. (See Complaint ¶ 23; SunGard Answer ¶ 23; Comdisco Answer ¶ 23.) [10] The government has defined midrange computers to include all UNIX-based systems that are not PC LAN systems. Under this definition, the covered midranges are: the AS/400, DEC, HP/9000, Filenet, NCR, RS/ 6000 & SP, Sequent, Stratus, Sun, Tandem, and Unisys platforms. (Gov. Ex. 102, at 3; Nov. 8 Trial Transcript, Vol. I, at 28-30.) [11] Examples of customers excluded under this definition are those who use Intel-based computers and Windows and NT software. Plaintiff contends that these customers should be excluded from the relevant product market because they have a broader range of viable disaster recovery options than do those businesses with larger computing systems. [12] Keating testified that one new company, Mainline, now offers a quick-ship service for mainframes, although this product has an RTO of up to five days, rather than the three- or four-day maximum for shared hotsites. (Nov. 8 Trial Transcript, Vol. II, at 87-88.) [13] According to the plaintiff's calculations, only 640 out of some 2,400 Comdisco customers use mainframe equipment (Gov.Ex. 101), and just 887 of SunGard's 5,500 customers use mainframe equipment. (Gov.Exs.80, 99.) [14] Areeda has explained the concept of captive output in the context of iron ore. If iron ore is the relevant market and if shares are best measured there by sales, then internally used ore — so-called captive output — is part of the ore market even though it is not sold as such. In measuring the market power of a defendant selling iron ore, the ore used internally by other firms constrains the defendant’s ability to profit by raising ore prices to monopoly levels. The higher ore price may induce an integrated firm to expand its ore production — to supply others in direct competition with the alleged monopolist or to expand its own steel production and thereby reduce the demand of other steel makers for ore, or both. Hence, captive output constrains the defendant regardless of whether integrated firms sell their ore to other steel makers previously purchasing from the defendant. In sum, the integrated firm's ore output belongs in the market. Areeda ¶ 535e (citing the Merger Guidelines § 1.31). [15] The Heinz district court did note that homemade baby food was not a substitute under the antitrust laws for jarred baby food because "the Supreme Court’s interchangeability test refers to products." 116 F.Supp.2d at 195 (emphasis and internal quotations omitted). Unlike an internal disaster recovery system, however, homemade baby food is not an aspect of vertical integration; by definition, individual consumers cannot vertically integrate by producing a product that they would otherwise have to purchase. Section 1.31 of the Merger Guidelines also recognizes that vertical integration is relevant to the product market where that integration substitutes for a product that would otherwise have to be obtained from an external vendor. Merger Guidelines § 1.31. [16] Plaintiff has ably demonstrated that for some of these customers, the switch to internal either has not happened, or if it has, it was not the result of a business decision related to the price of the external hotsite solution. (See Gov. Exs. 11 IB, D-I, L-O, Q-R, V.) However, the fact remains that more customers have left to pursue an internal solution than have left to go with a competitor and that this trend is continuing. [17] Defendants have introduced evidence that internal hotsites are becoming an even more viable competitive threat as technology develops. The trend toward smaller, distributed servers has enabled companies to more easily implement internal solutions. (See, e.g., Def. Ex. 72, Letter from General Mills; Def. Ex. 54; Def. Ex. 151, Simmons Deck ¶¶ 8, 27, 33-34, 40.) The ability of companies to buy IBM's new "MIPS-on-demand,” which allows them to expand the capacity of their data processors in case of a disaster, also permits customers to move to an internal solution. (Def. Ex. 151, Simmons Deck 118; Gov. Ex. 98.) MIPS — for "millions of instructions per second” — is a measurement of computer processing capability, and the new IBM product allows a customer to partition the computer into smaller units that can be expanded in the event of a disaster, and then retracted afterward. (Id.) Finally, the declining price of computer hardware also permits customers to switch to an internal back-up more readily. (Def. Exs. 15, 151, Simmons Deck ¶ 34.) [18] In fact, Gov. Ex. 62, which is a SunGard memo entitled "The Hidden Costs of In-House Disaster Recovery," appears to focus on internal high availability because it discusses the high cost of data communications between the primary processor and the backup site. Internal hotsites do not require such telecommunications, relying instead on backup tapes for use only in the event of a disaster. [19] Given the fact that the majority of Sun-Gard’s customers use a distributed UNIX or midrange system, it is especially difficult to argue that a substantial proportion of Sun-Gard’s customers will be unable to switch to *190 an internal solution in light of the fact that a UNIX customer can recover on an internal basis more inexpensively than a mainframe customer. (See Def. Ex. 151, Simmons Decl. ¶ 27.) [20] The evidence also suggests that many combinations of the other forms of disaster recovery may be a substitute for shared hotsites. Unlike most antitrust cases, in which the product at issue is a discrete item — such as the pen in Gillette, the hard-walled containers in Owens-Illinois, the baby food in Heinz, and the loose-leaf tobacco in Swedish Match — disaster recovery services encompass a variety of solutions that often complement each other, rather than a discrete product. While the government has proven that some forms of disaster recovery, when used alone, are not substitutes for a shared hotsite contract, there may well be combinations of these services that can be substituted for the traditional shared hotsite. In fact, most companies employ a variety of business continuity systems to meet their varying RTOs for different types of data, and as noted, more than half of SunGard's clients have contracts that allow them to determine their preferred recovery method at the time of the disaster. (Def. Ex. 151, Simmons Decl. ¶ 11-12.) If the price of shared hotsites were to increase, the evidence suggests that some combination of, for example, work-area recovery and quick-ship, or internal hotsites and mobile recovery, may well be a viable substitute. As Simmons notes, "customers do not demand hotsites— they demand data protection in the most economical manner possible.” {Id. ¶ 7.) Unlike a pen, a hard-walled container, baby food, or chewing tobacco, shared hotsites themselves may not be the "product” that is sought and purchased by disaster recovery customers. Rather, clients are buying the economical protection of data with a specific RTO— shared hotsites are just one particular way to achieve that goal. It is therefore difficult to pigeonhole this somewhat nebulous “product” into traditional antitrust principles. [21] Defendants' economist hypothesized that SunGard could not profitably afford to lose more than 5 percent of its customers in response to a SSNIP because of the relatively high profit derived from each additional cus *191 tomer. Because SunGard has insufficient information as to whether its customers would switch in response to a SSNIP, Harris testified that the company could not take the risk of raising prices. (Gov. Ex. 281, Harris Report ¶¶ 13-22.) In response, plaintiff has demonstrated that shared hotsite providers invest a great deal of time and money in gathering information about their customers and are typically aware of those clients that could switch to an alternative solution. {E.g., Gov. Ex. 107, Jackson Dep. at 53-56.) [22] In addition, defendants have submitted letters from some 68 customers who have switched (Def. Exs. 6-14; 16-74), and other documents reflecting that their customers are considering such a switch. {See, e.g., Def. Exs. 92-94.) [23] On the contrary, since defendants have submitted an equal or greater number of conflicting statements, one can only conclude that the statements submitted by both parties prove very little, if anything at all. Moreover, several of the declarations of fact witnesses deserve little, if any, weight. For instance, defendants’ declaration of Charles Hollis, a vice-president of EMC, which is a huge computer outsourcer, is entitled to limited weight because EMC appears to have a vested interest in the outcome of this litigation. EMC had agreed to provide some financing for SunGard's acquisition of Comdisco, and thus, it may stand to gain substantial business if the transaction goes through. (Gov.Ex. 260.) Similarly, plaintiff's declaration from Brian Fowler of HP is also of little value because HP stands to win the auction for Comdisco if the SunGard purchase is successfully blocked. Finally, several customers submitted conflicting statements, telling the defendants one thing and the government another. See supra p. 183. [24] There is no question that time pressures may have prevented any thorough analysis of the many customers that will be affected by the proposed acquisition. It is, however, beyond dispute that the government must offer sufficient evidence to satisfy its burden of proof, and given the government’s product market definition that includes a wide variety of computer equipment and customer types, a more detailed analysis was needed before one could draw any generalizations about the customers. [25] Because the Court has found that plaintiff has failed to meet its burden of establishing the relevant product market, it need not address the remaining disputed issues in the case — the probable effect of the transaction on competition, the extent of concentration in the market, and defendants’ arguments regarding the unreliability of plaintiff’s market statistics, the likelihood of price discrimination, barriers to entry into the market, the effect of knowledgeable and sophisticated customers, and efficiencies resulting from the transaction.