In 2020, a streaming startup signed a five-year peering deal with a mid-tier CDN, locking in volume discounts for a projected audience that never materialized. The service shut down eighteen months later. The contract, however, kept running. Every month, the CDN invoiced a company that no longer existed, and the network reserved capacity that no one was using. Such scenarios are common in an industry where infrastructure commitments routinely outlive the businesses that made them.

The Contract Outlives the Streamer

Peering agreements between content delivery networks and their clients typically run five to ten years. CDNs invest heavily in colocation, transit, and hardware, and they need predictable revenue to justify that capital. But the average streaming startup does not survive that long. A 2022 study by the Streaming Observatory found that 62% of streaming services launched between 2015 and 2020 had either shut down or been acquired within three years. The contracts they signed, however, do not expire with them.

When a client vanishes, the CDN does not immediately reclaim the capacity. The ports, the router tables, and the reserved bandwidth remain in place, often for months or years, because unwinding a peering arrangement is manual and slow. The network operator has to verify that the client is truly gone, renegotiate or terminate the agreement, and then physically reconfigure the infrastructure. In practice, this work gets deprioritized. The capacity sits idle, a ghost that still consumes resources.

The financial impact is real. The CDN still pays for the underlying transit and colocation, even if the bandwidth is not being used. The startup's bankruptcy means the invoice goes unpaid, but the cost does not disappear. It gets absorbed into the CDN's operating expenses, diluted across the remaining customers. In some cases, the CDN is contractually obligated to keep the capacity warm for a notice period, even after the client stops paying. The result is a sunk cost that lingers on the balance sheet.

This dynamic is not confined to tiny players. Large streaming services have shut down or been acquired, leaving behind peering obligations that their successors either inherit or abandon. For example, when the short-form video service Quibi shut down in December 2020, its CDN contracts with multiple providers were left stranded. The CDNs had to manually renegotiate or absorb the costs, a process that took months.

Why CDNs Sign Long-Term Deals

CDNs are capital-intensive businesses. Building a network edge requires upfront spending on servers, bandwidth, and data center space. To amortize those costs, they need guaranteed revenue streams. Long-term contracts provide that predictability. A five-year commitment lets a CDN plan its capacity expansion with confidence, knowing that a certain amount of traffic is locked in. This is the core reason why these deals exist, and it is not a malicious design. It is a rational response to an expensive infrastructure.

For startups, the appeal is different. A long-term peering deal with a reputable CDN can be a credibility signal. It tells investors and partners that the company is serious about scaling. It also buys better pricing: CDNs offer steep discounts for volume guarantees, which can be tempting for a young company projecting rapid growth. The trade-off is that the startup is committing to a level of traffic it may never achieve, and the penalty for under-utilization is baked into the contract.

Peering wars have made this worse. As CDNs compete for market share, they have become more aggressive in their negotiations, offering lower rates and more favorable terms to win large accounts. For instance, in 2019, Netflix's deal with a major CDN was rumored to be renegotiated at a significantly lower rate, putting pressure on smaller CDNs to match. But the underlying economics have not changed. The CDN still needs to recover its fixed costs, and if a client under-delivers, the margin erodes. In some cases, the CDN is willing to absorb the risk because it expects the client to grow. When the client does not, the loss is real.

There is also a psychological factor. Startups are optimistic by nature. They believe their user base will expand, and the contract terms reflect that optimism. The salesperson on the CDN side knows the odds, but the commission structure rewards closing the deal. So both parties sign, each hoping the other's assumptions hold. When they do not, the contract becomes a liability.

The Ghost Traffic Problem

Ghost traffic is the term some engineers use for the packets that flow through a CDN but are not tied to any active customer. This can happen when a client goes dark but the routing rules remain. The CDN's edge servers still advertise the prefixes, and the router tables still contain the entries. The traffic does not disappear; it just becomes orphaned. The network keeps processing it, but no one is paying for it.

Re-routing that traffic requires manual intervention. An engineer has to identify the dead routes, verify that the client is truly defunct, and then update the configuration. This is tedious work, and it often gets deprioritized in favor of more pressing operational tasks. As a result, ghost routes can persist for months. They consume memory and processing cycles, and they can even cause subtle routing anomalies that affect other customers.

Forecasting errors compound the problem. When a CDN plans its capacity, it uses historical data and growth projections. If a large client churns, the forecast is suddenly off. The CDN may have built out capacity that is now underutilized, and it takes time to adjust. In the interim, the network operates with excess capacity, which drags down utilization rates and margins.

Operational teams inherit these orphan contracts. A new engineer joining a CDN might discover a peering agreement with a defunct company that has been running for years. The contract is still in the billing system, the capacity is still reserved, and no one remembers why it was set up. This is not a failure of individual diligence; it is a structural issue. The systems are not designed to detect and retire dead agreements automatically.

Lessons from the Streaming Bust

The streaming bust of the early 2020s offers a stark example. Quibi, the short-form video service, launched in April 2020 and shut down in December of the same year. It had signed CDN and peering deals that were meant to support a massive audience. When the service collapsed, the infrastructure was left stranded. Some reports suggest that Quibi's CDN partners were able to repurpose the capacity, but not without significant effort.

Smaller CDNs absorbed the shock more acutely. They did not have the scale to absorb the loss of a major client, and the stranded capacity hurt their bottom line. Some were forced to renegotiate terms with other customers or eat the costs. Hyperscalers like AWS and Google Cloud had more flexibility. They could shift the capacity to other workloads, but even they had to manually reconfigure their networks.

Resale markets for dark fiber emerged as a partial solution. When a streaming service went under, its fiber leases could be sold to other companies, often at a discount. But this is not a clean process. The original contract terms may restrict transfer, and the buyer has to take over the remaining obligations. It is a messy secondary market, and it does not always work.

The lack of a standard exit clause for bandwidth is a glaring gap. Most contracts include provisions for termination, but they are often punitive. A startup that wants to exit early faces penalties that can be worse than paying for unused capacity. This asymmetry is a key reason why stranded infrastructure persists.

The Economics of Stranded Infrastructure

When a CDN loses a major client, its fixed costs do not disappear. The colocation fees, the transit costs, and the hardware depreciation all continue. The CDN must spread those costs over a smaller customer base, which means either raising prices or accepting lower margins. In a competitive market, raising prices is not always possible. The result is a slow erosion of profitability.

Secondary markets for unused capacity have emerged, but they are inefficient. A CDN might sell excess bandwidth to a smaller provider at a discount, but the price is often below the original cost. This is a rational response to sunk costs, but it distorts the market. It also creates a perverse incentive: customers learn that they can wait for a distressed sale rather than pay full price.

Renegotiation favors the surviving party. When a contract is renegotiated, the CDN has more leverage than the startup, because the startup needs the infrastructure and has few alternatives. But when the startup is failing, the leverage shifts. The CDN may offer concessions to keep the contract alive, even if it means accepting lower rates. This dynamic is not unique to CDNs, but it is particularly acute in a capital-intensive industry.

Balance sheets hide the liability. Stranded capacity is often not visible in financial statements. It shows up as lower utilization or higher operating costs, but it is not a line item. This makes it easy for management to ignore, and it can persist for years without being addressed.

How Operators Can Hedge Against Churn

There are practical steps that CDNs can take to reduce the impact of client churn. One is to include break clauses tied to revenue. Instead of a flat commitment, the contract can include provisions that allow for termination if the client's revenue falls below a certain threshold. This gives the CDN an exit if the client is not generating the expected traffic.

Auditing usage quarterly, not annually, is another option. By tracking actual traffic against projections, the CDN can identify under-performing contracts early and take action before the situation becomes critical. This requires investment in monitoring tools and a willingness to act on the data.

Building flexibility into peering ports is also important. Instead of dedicating a fixed amount of capacity to a single client, the CDN can use software-defined networking to dynamically allocate resources. This allows the CDN to shift capacity to other customers as needed, reducing the impact of a single client's departure.

Diversifying across content types is a longer-term strategy. A CDN that serves only streaming video is more exposed to the fortunes of that market. A CDN that also serves web content, software downloads, and enterprise applications can absorb the shock of a streaming startup failing. Finally, tracking the financial health of startup clients publicly can give early warning signs. If a client is struggling to raise funding or is cutting costs, it may be a signal that the contract is at risk.

The New Normal: Short Leases, Flexible Peering

The industry is slowly moving toward shorter leases and more flexible peering arrangements. Monthly contracts are gaining traction, especially among smaller CDNs and startups. These contracts are less attractive from a planning perspective, but they reduce the risk of stranded assets. Usage-based pricing is also becoming more common, replacing flat-rate commitments with models that charge based on actual traffic.

Software-defined exchanges are enabling this shift. These systems allow CDNs to provision and de-provision peering connections on demand, without the manual effort that was previously required. This makes it easier to respond to churn and to repurpose capacity quickly. Machine learning is also being used to predict churn, analyzing usage patterns and financial signals to identify at-risk contracts.

Contracts are adapting to the streaming lifecycle. Instead of assuming that a client will grow forever, they are incorporating milestones and checkpoints. If a client does not hit certain targets, the contract can be renegotiated or terminated. This is a more realistic approach, but it requires both parties to be honest about the uncertainties.

The shift is not without trade-offs. Short-term contracts mean less revenue predictability for CDNs, which can make it harder to justify capital investment. Usage-based pricing can be more complex to administer, and it may not suit every customer. Some network engineers argue that the industry overcorrected after the streaming bust, and that long-term deals still have a place for established players with predictable traffic. For example, a major live sports streaming service might benefit from a five-year contract that locks in capacity for major events, even if it means absorbing some risk. The key is to build in flexibility, such as volume adjustments or escape hatches, rather than abandoning long-term agreements altogether. But the alternative, clinging to rigid long-term deals that outlive their usefulness, is a recipe for stranded infrastructure. The industry is learning that the only way to avoid ghost traffic is to build flexibility into the system from the start.