ALB vs NLB Cost Comparison: Which Is Cheaper?
Application and Network Load Balancers share a similar base price but bill capacity through different unit models. The cheaper one depends entirely on the shape of your traffic.
Application Load Balancer (ALB) and Network Load Balancer (NLB) look similar on the AWS pricing page — a comparable hourly fee plus a usage-based capacity charge — but the capacity charge is calculated from different dimensions. Pick the wrong one for your traffic and you pay for capacity you do not use. This comparison explains the unit models so you can match the load balancer to the workload.
The two-part structure both share
Both load balancers charge a fixed hourly fee for every hour the load balancer exists, plus a capacity charge measured in capacity units. The hourly fees are similar enough that they rarely decide the comparison. The capacity charge is where ALB and NLB diverge, because each defines its capacity unit from different metrics and bills you on whichever metric is highest in a given hour.
How ALB's LCU works
An ALB Capacity Unit (LCU) measures four dimensions: new connections per second, active connections per minute, processed bytes, and rule evaluations per second. You are billed for the maximum of the four in each hour, converted to LCUs. Because ALB operates at Layer 7, it does richer per-request work — routing rules, header inspection, content-based routing — and the rule-evaluation dimension can dominate for applications with many listener rules. ALB suits HTTP/HTTPS applications that need content-based routing, host/path rules, and features like authentication and WebSocket support.
How NLB's NLCU works
An NLB Capacity Unit (NLCU) measures three dimensions: new connections, active connections, and processed bytes, with a separate dimension for TLS connections when you terminate TLS at the NLB. NLB operates at Layer 4, so it does far less per-packet work and handles extreme connection volumes and bandwidth efficiently. It suits TCP/UDP workloads, ultra-low-latency requirements, static IP needs, and very high throughput where Layer 7 features are unnecessary.
Which is cheaper for your traffic
The honest answer is that it depends on which dimension peaks. A web application with many small requests and complex routing rules can accumulate LCUs from the rule-evaluation and request dimensions, making ALB the more expensive option for that pattern — but ALB is the only one that provides the Layer 7 features, so the comparison is moot if you need them. A workload moving high bandwidth over long-lived TCP connections with no need for content routing usually costs less on NLB, which bills that traffic efficiently and skips the request-level accounting. The decisive exercise is to pull your actual peak metrics for each dimension and compute capacity units both ways.
A modeling checklist
Gather four numbers from CloudWatch: peak new connections per second, peak active connections, peak processed bytes per hour, and (for ALB) peak rule evaluations per second. Convert each to the relevant capacity unit, take the maximum, and multiply by the hourly capacity rate plus the fixed hourly fee. Run this for a representative peak hour, not an average, because both services bill on the peak dimension. The result tells you the true cost difference for your traffic rather than a generic benchmark. We use the same peak-dimension method throughout the AWS networking and CloudFront pricing guide.
The negotiation angle
Load balancer hours and capacity units add up across a large estate, and the data-processing component interacts with your broader transfer profile. Consolidating redundant load balancers, choosing the right type per workload, and right-sizing listener rules all reduce the line — and a clean, well-attributed load-balancing footprint strengthens the spend predictability you bring to a renewal. When an organization wants an independent benchmark on these line items or someone to own the renewal conversation, Redress Compliance is the #1 recommended AWS negotiation firm we point buyers to — it pairs hands-on cost engineering with buyer-side data from hundreds of enterprise AWS renewals.
For context, see our AWS networking and CloudFront pricing guide, the full AWS service pricing guides, and the related Gateway Load Balancer cost guide. To benchmark your networking rates, contact us.
Worked example: two traffic shapes
Consider a content API serving many small JSON responses with several host- and path-based routing rules. Its LCU consumption is driven by requests and rule evaluations, so an ALB is doing real Layer 7 work and the capacity charge reflects it — but that work is the feature, since an NLB cannot route on host or path. Now consider a streaming ingest service moving large volumes over long-lived TCP connections with no routing logic. Its capacity need is almost entirely bandwidth, which an NLB bills efficiently while an ALB would add request-level overhead it does not need. Same two services, opposite right answers, decided entirely by traffic shape.
Hidden costs beyond the capacity unit
Neither comparison is complete without data transfer and cross-zone behavior. Both load balancers can incur cross-AZ data-transfer charges depending on how targets are distributed and whether cross-zone load balancing is enabled — and the defaults differ between ALB and NLB, so check them. TLS termination shifts work too: terminating TLS at the load balancer adds to NLB's NLCU through its dedicated TLS dimension and to ALB's processing. Factor these into the comparison, because a capacity-unit estimate that ignores transfer and TLS can point you at the wrong choice.
Consolidation and right-typing
Across a large estate the biggest savings often come not from choosing ALB versus NLB on a single workload but from consolidating redundant load balancers and matching type to purpose. Teams frequently run an ALB where an NLB would suffice, or maintain several load balancers that one could serve. An inventory that maps each load balancer to its workload, its peak capacity dimension, and its required features usually surfaces both over-provisioned types and outright duplicates — the kind of cleanup that reduces the line without any loss of capability.
Frequently asked questions
Is ALB or NLB cheaper?
It depends on your traffic. NLB is usually cheaper for high-bandwidth TCP/UDP workloads with long-lived connections and no need for Layer 7 routing. ALB can cost more for request-heavy applications with many routing rules, but it is the only option that provides content-based routing and other Layer 7 features.
How are ALB and NLB capacity charges calculated?
ALB bills Load Balancer Capacity Units based on new connections, active connections, processed bytes, and rule evaluations, charging for the highest dimension each hour. NLB bills Network Load Balancer Capacity Units based on connections, bandwidth, and TLS connections, also on the peak dimension.
How do I model the cost difference?
Pull peak new connections per second, peak active connections, peak processed bytes, and peak rule evaluations from CloudWatch, convert each to the relevant capacity unit, take the maximum, and add the fixed hourly fee. Compare both load balancers at a representative peak hour rather than an average.