Application load balancer LCU cost is the capacity-unit charge on a load balancer that is already serving traffic. You also pay for each hour or partial hour the balancer is running. The LCU is the traffic meter. AWS charges the dimension with the highest usage, averaged over the hour, and ignores the other three for that hour. A busy API almost always has this line. A balancer with no traffic is a different cleanup, covered in idle load balancer cost.
Rutagon treats LCUUsage as its own bill line, next to LoadBalancerUsage for the running hour. AWS cloud infrastructure reviews start from the Elastic Load Balancing pricing page for the Region on the invoice. The worked examples below are the US East (N. Virginia) examples on that page. Another Region has its own hourly and LCU rates.
Where application load balancer LCU cost is priced
The pricing page defines an LCU as four dimensions. New connections are 25 per second. Active connections are 3,000 per minute, or 1,500 per minute when mutual TLS is in use. Processed bytes are 1 GB per hour for EC2 instances, containers, and IP targets, and 0.4 GB per hour for Lambda targets. Rule evaluations are 1,000 per second. You pay for the largest of those four. The running charge is separate: each partial hour the balancer exists is billed as a full hour.
In the page's first US East example, the application takes one new connection per second, each lasting two minutes, five requests per second, and 300 KB per second of processed bytes, with 60 rules. New connections are 0.04 LCU. Active connections are 120 per minute, also 0.04 LCU. Processed bytes are 1.08 GB per hour, or 1.08 LCU. Rule evaluations, after the 10 free rules, are 250 per second, or 0.25 LCU. The hour is priced on 1.08 LCU. At $0.008 per LCU that is $0.00864 per hour, $6.22 across a 30-day month. The running charge in that example is $0.0225 per hour. Together they are $0.03114 per hour, or $22.42 per month. Bytes won. The other three dimensions were real and were not billed.
usage: region-LCUUsage is the LCUs consumed
usage: region-LoadBalancerUsage is the running hour, partial hours round up to a full hour
usage: region-ReservedLCUUsage and region-IdleProvisionedLBCapacity when a reservation is set
price_us_east_example: $0.008 per LCU, $0.0225 per running hour
winner: max of new connections, active connections, processed bytes, rule evaluations
lambda_bytes: 0.4 GB per LCU, and that share adds to the EC2 byte LCUs Rule math is request rate * (rules processed - 10). The page says that if 10 or fewer rules are configured, the rule dimension is ignored. The CloudWatch metric RuleEvaluations still includes those free evaluations, and it does not count the default rule. Summing the metric and dividing by 1,000 will overstate the billed LCUs. Subtract the free 10 per request before you call rules the winner. In the example, 60 rules were not the winner. A listener with a host rule per tenant and little body traffic can flip that. We do the subtraction before we delete rules that are not the line.
Lambda targets change the byte dimension. One LCU covers 0.4 GB per hour toward Lambda and 1 GB per hour toward EC2, containers, or IP addresses. The page's mixed example adds those LCU counts together. They do not share a single gigabyte allowance. LambdaTargetProcessedBytes is the Lambda share of ProcessedBytes. A large download through a Lambda target spends LCUs at the 0.4 GB rate. Moving that response to an EC2 or container target, or keeping the object off the balancer entirely, is the change that cuts the byte LCU. Shrinking the instance behind the balancer does nothing to this line.
A reservation is a second LCU charge. You pay for the LCUs you reserved and for any minute you go past them. In the page's US East reservation example, 200 reserved LCUs and 240 actually used cost $1.60 for the reserved LCUs and $0.32 for the 40 extra, plus the $0.0225 running hour, $1.9425 for that hour. ConsumedLCUs still reports what you used. With a reservation you pay the reserved amount plus the excess. ReservedLCUs is the reserved amount. Unused reserved capacity is IdleProvisionedLBCapacity. A reservation set for a launch that never arrived keeps billing the reserved LCUs on a balancer whose consumed LCUs are quiet.
New accounts see a free tier on the pricing page: 750 hours per month shared between Classic and Application load balancers, and 15 LCUs for Application Load Balancers. We read that section against the account's age. A balancer that has been up for a long time is not inside it. We do not subtract 15 LCUs from a production LCUUsage line unless the bill shows the credit.
Which dimension to change
ConsumedLCUs is the number the hour will use when no reservation sits above it. NewConnectionCount, ActiveConnectionCount, ProcessedBytes, and RuleEvaluations tell you which dimension is able to win. We estimate each one with the pricing-page divisors: 25 new connections per second, 3,000 active connections per minute, 1 GB or 0.4 GB of bytes per hour, 1,000 rule evaluations per second after the free 10. The largest estimate is the one we change. The others can stay.
Connection churn is the new-connection dimension. Clients that open a TCP connection per request, instead of reusing one, spend LCUs at 25 per second per LCU even when the body is small. Keep-alive on the client is the fix when that estimate is the max. Active connections are the sockets held open. A fleet of idle WebSocket clients counts here at 3,000 per minute per LCU, and mutual TLS cuts that unit to 1,500. The bytes those sockets move are still the byte dimension. Holding a quiet socket and pushing a large body are different knobs.
Rules are the cheap dimension until they are not. The first 10 processed rules are free, and 10 or fewer configured rules drop the dimension out of the LCU calculation entirely. Past that, every extra rule multiplies by the request rate. Consolidating host or path rules, or moving a long list off the balancer and into the application where it is not evaluated on every request, is the fix when the subtracted rule math is the max. We do not delete the health-check path and call that a cost change. We delete rules that the request rate is paying for.
Reservations get set from a measured peak of ConsumedLCUs, or they get removed. A reservation larger than the peak is IdleProvisionedLBCapacity by definition. The pricing example is the shape: you pay $0.008 in that Region for reserved LCUs you did not consume, and you still pay $0.008 for the ones you did consume above the reservation. The reservation is capacity you asked to have waiting. It is not a discount. Public IPv4 and data transfer sit on other lines. The pricing page bills both separately from the LCU.
What we leave on the balancer
We leave the LCU alone when the winning dimension is the traffic you meant to serve and the worked example, scaled to your counts, matches the bill. The page's $6.22 LCU month was 1.08 LCU of bytes, not waste. Waste is a reservation above consumption, a Lambda target carrying downloads, a rule list evaluated on every call, or a connection opened and thrown away per request.
Rutagon runs this as a short bill review: LoadBalancerUsage, LCUUsage, any reserved and idle-provisioned LCU lines, then the four dimensions against the pricing-page units. The US East examples are the source for $0.008, $0.0225, $6.22, $22.42, and the $1.9425 reservation hour. Your Region's row replaces those rates. The dimension math does not.
Frequently Asked Questions
What is the LCU line on the bill?
LCUUsage is the capacity units consumed. LoadBalancerUsage is the running hour, and a partial hour bills as a full hour. If you reserved capacity, ReservedLCUUsage and IdleProvisionedLBCapacity are the reserved units and the reserved units you did not consume. ConsumedLCUs in CloudWatch is the usage, including when a reservation is higher.
Why do Lambda targets cost more LCUs for the same bytes?
An LCU includes 1 GB per hour toward EC2, container, and IP targets, and 0.4 GB per hour toward Lambda. The pricing page adds those LCU counts together when a balancer has both. A heavy body through Lambda is the byte dimension at the 0.4 GB unit. Moving that body off the Lambda target changes the unit.
Do extra listener rules always raise the bill?
The first 10 processed rules are free, and 10 or fewer configured rules drop the dimension out of the calculation. Above that, evaluations are the request rate times the rules past 10. RuleEvaluations includes the free ones, so the metric is higher than the billed count. Rules raise the bill when that subtracted number is the largest of the four dimensions.
Does a reservation cap the LCU charge?
A reservation is a floor you pay and a line you can still exceed. The US East example reserves 200 LCUs, uses 240, and bills $1.60 for the reserved units plus $0.32 for the extra 40, at $0.008 per LCU, plus the $0.0225 running hour. Unused reserved units still cost. We set the reservation from measured ConsumedLCUs or we remove it.
Ask for an AWS bill review that names which LCU dimension is winning on each live balancer. Contact Rutagon at rutagon.com/contact, 907-841-8407, or contact@rutagon.com.