16 Data is constantly moving. Users upload files, applications exchange information, backups are written, AI models read training datasets, and customers download content. Every one of those transfers has a direction. That’s all ingress and egress describe. Ingress is data moving into a system. Egress is data moving out. The definitions are straightforward, but the consequences are not. For most organizations, the direction data moves determines much more than network traffic. It influences cloud costs, recovery planning, security policies, AI infrastructure, and even how difficult it becomes to move between providers. If you only need to remember one thing, it’s this: Ingress is usually inexpensive. Egress is where the long-term operational and financial impact begins. This guide explains the difference between ingress and egress, why cloud providers price them differently, and why understanding egress has become increasingly important as organizations build larger data platforms. What is Ingress? Ingress is data entering a system. That system might be a cloud environment, an object storage platform, a Kubernetes cluster, a corporate network, or even a single application. If information is arriving, it is ingress. Examples of Ingress Uploading files into object storage Backup software writing nightly backups IoT devices sending telemetry into a data lake Applications receiving API requests Replicating data into another storage cluster Copying a training dataset into a GPU environment Most cloud providers charge little or nothing for ingress. Making it easy to bring data into a platform reduces friction and encourages customers to store more data there. What is Egress? Egress is data leaving a system. Unlike ingress, egress usually happens when data is being used rather than simply stored. Examples of Egress Downloading files Restoring backups Serving images, video, or web content Reading objects from cloud storage Moving data to another cloud provider Running analytics jobs against remote datasets AI training and inference repeatedly reading stored data This distinction matters because organizations typically write data once but read it many times. A backup may be uploaded only once, but it could eventually be restored during a disaster. A dataset may be copied into cloud storage once, but an AI pipeline might read it hundreds of times during model development. Storage is often a one-time event. Reading data is continuous. That difference explains why egress has a much greater influence on long-term operating costs. Egress vs. Ingress at a Glance FeatureIngressEgressDirectionInto a systemOut of a systemTypical OperationsUploads, writes, replication inDownloads, restores, reads, replication outNetwork TrafficIncomingOutgoingObject StoragePUT requestsGET requestsTypical PricingUsually free or very low costFrequently charged per GBUsage PatternOften occurs onceCan occur repeatedly Why Cloud Providers Charge for Egress Many people assume cloud providers charge for egress because moving data is expensive. Network infrastructure certainly has a cost, but the pricing model also reflects how cloud platforms are used. Making ingress inexpensive removes barriers to adoption. Organizations can migrate backups, archives, analytics datasets, and application data without worrying about large transfer charges. Egress works differently. Once large datasets are inside a platform, moving them elsewhere—or repeatedly reading them from another location—creates ongoing transfer costs. Those charges increase alongside usage rather than storage capacity. In other words, the more valuable your data becomes, the more frequently you access it. As access grows, egress costs can begin to outweigh storage costs. This is one reason egress fees receive so much attention during cloud architecture discussions. Why Egress Matters More Than Ingress For small workloads, the difference between ingress and egress is usually insignificant. At enterprise scale, it becomes an architectural decision. Backup and Disaster Recovery Cloud storage can be an economical place to keep backup copies because writing backups generally incurs little or no transfer cost. Restoring them is different. A large-scale recovery requires reading that data back, often under severe time pressure. Restoring hundreds of terabytes or several petabytes generates egress traffic precisely when an organization can least afford delays or unexpected costs. Recovery planning therefore needs to account for both restore performance and data transfer economics. AI and Analytics Modern AI workloads are extremely read-intensive. Training models, fine-tuning models, vector indexing, retrieval-augmented generation (RAG), and large-scale analytics all read the same datasets repeatedly. Unlike archival storage, where data may remain untouched for months, AI pipelines continuously retrieve information. For organizations building AI infrastructure, the cost of repeatedly reading data can become just as important as the cost of storing it. Media Delivery Streaming platforms, content distribution systems, and software downloads generate constant outbound traffic. The larger the audience, the more frequently stored content is retrieved. In these environments, egress scales directly with customer demand. Multi-Cloud Environments Organizations increasingly distribute workloads across multiple cloud providers. That flexibility improves resilience and reduces dependence on a single platform, but it also increases the amount of data crossing cloud boundaries. Each transfer may seem small in isolation. Across months or years, those transfers become a recurring operational expense. Egress is Always Relative One of the most common sources of confusion is forgetting that ingress and egress depend on perspective. Uploading a file from your laptop into cloud storage is: Egress from your laptop Ingress for the cloud Downloading the same file later reverses the direction: Egress from the cloud Ingress for your computer The same transfer is both ingress and egress depending on which system you’re describing. The same principle applies inside cloud environments. Data moving between regions, availability zones, or storage clusters may still be considered egress from one system and ingress into another, even if it never crosses the public internet. Didn’t Cloud Providers Remove Egress Fees? Major cloud providers introduced programs that reduce or waive certain data transfer charges for customers permanently leaving their platforms. These changes were designed to comply with regulatory updates like the EU Data Act, making switching providers easier and reducing barriers to migration. What did not change is the pricing for routine operational egress. Organizations still generate billable outbound traffic when they: Restore backups Download stored objects Run analytics against cloud data Train AI models Serve content to users Transfer data between environments beyond the applicable free allowances The changes made cloud migration easier. They did not eliminate the ongoing cost of reading data during normal operations. Egress and Security Ingress and egress are also fundamental security concepts. Ingress controls determine what is allowed into an environment. Egress controls determine what is allowed to leave. Firewalls, zero-trust architectures, Kubernetes network policies, and cloud security platforms all distinguish between inbound and outbound traffic because the risks are different. Many cyberattacks focus on gaining access through ingress. Conversely, many data breaches ultimately succeed through uncontrolled egress, when sensitive information leaves the environment without authorization. Understanding the difference is therefore critical for both networking and cybersecurity. Egress in Object Storage Object storage makes the distinction especially clear. Writing objects into an S3 bucket is ingress (PUT requests). Reading those objects is egress (GET requests). For workloads that primarily archive data, outbound transfers may remain relatively low. For workloads involving AI, analytics, backup restores, or media distribution, reads occur continuously. That changes the economics. Organizations evaluating object storage should therefore look beyond raw capacity and consider how frequently data will be accessed throughout its lifecycle. A workload that stores data once but retrieves it hundreds of times has very different cost characteristics than one that simply archives information for long-term retention. This is one reason many organizations deploy private or on-premises object storage for read-intensive workloads. Keeping frequently accessed data close to the applications, analytics platforms, or GPU clusters that consume it avoids repeatedly transferring information across metered cloud boundaries while still providing S3-compatible access. How to Reduce Egress Costs Organizations cannot eliminate egress entirely, but they can reduce both the frequency and cost of outbound data transfers through smart architectural decisions. Keep compute close to data: AI training, analytics, and high-performance applications become significantly more expensive when every read crosses a metered network boundary. Running compute in the same environment as frequently accessed data minimizes both latency and egress charges. Cache frequently accessed content: Applications that repeatedly serve the same files or API responses benefit from caching. CDNs, edge caches, and application-level caching reduce the number of requests that reach the underlying storage platform, lowering outbound traffic while improving performance. Avoid unnecessary cross-region transfers: Replication between regions improves resilience, but applications that continuously read data across regions or cloud providers can generate substantial transfer costs. Where possible, keep workloads local to the data they consume. Match storage architecture to workload: The public cloud per-gigabyte egress model makes sense for data that is written once and rarely read. However, for read-intensive operations like continuous AI model development or large-scale media processing, deploying private or on-premises object storage provides predictable economics while maintaining the S3-compatible access modern applications require. Frequently Asked Questions Is uploading data ingress or egress? Both. Uploading is egress from the source system (like your local server) and ingress for the destination receiving the data (like a cloud storage bucket). Why is egress usually more expensive? Ingress generally happens once when data is written. Egress often occurs every time data is accessed, making it an ongoing operational cost rather than a one-time event. Does ingress usually cost money? Most cloud providers either do not charge for ingress or charge very little compared to outbound transfers to lower the friction of adoption. Did cloud providers eliminate egress fees entirely? No. They introduced programs that help customers permanently migrate away from their platforms to comply with new regulations, but routine operational egress is still generally billed beyond applicable free allowances. Which workloads generate the most egress? AI training, inference, analytics, media streaming, content delivery, and large backup restores all generate significant outbound traffic because they repeatedly read stored data. The Takeaway Ingress and egress describe the direction data moves. Ingress brings data into a system. Egress sends it out. The definitions are simple, but their impact reaches far beyond networking terminology. Egress influences cloud costs, disaster recovery planning, AI infrastructure, security policies, and long-term architectural flexibility. Organizations that understand how often their data will be read—not just how much they will store—are better positioned to choose the right storage architecture, predict operating costs, and avoid surprises as workloads continue to grow. Related reading What is the EU Data Act?https://www.solved.scality.com/what-is-the-eu-data-act/ How to avoid cloud lock-in while still benefiting from cloud serviceshttps://www.solved.scality.com/how-to-avoid-cloud-lock-in-while-still-benefiting-from-cloud-services/ AI unstructured data processing: Key techniques explainedhttps://www.solved.scality.com/ai-unstructured-data-processing/ Object storage vs. block storage vs. file storage: What’s the difference?https://www.solved.scality.com/object-storage-vs-block-storage-vs-file-storage/
