A live-streaming platform at Twitch's scale is sizing the bandwidth bill for a new regional tier. Sustained outbound is 40 Gbps with peaks near 120 Gbps. Estimate the monthly cost on cloud list egress against a committed CDN contract, and say which assumption dominates the error.

The assumptions, stated

Sustained 40 Gbps, 30-day month, and the traffic is viewer-facing video rather than API responses. Peak-to-sustained ratio is 3:1, which matters for the commit level but not for the byte total. Origin-to-edge traffic is separate and small if the CDN offload ratio is high.

The arithmetic

40 Gbps is 5 GB per second. Over 86,400 seconds that is 432 TB a day, so roughly 13 PB a month. Hold that number: everything else is a price applied to it.

Cloud list egress. AWS list pricing in 2026 charges internet data transfer out from US regions at roughly \(0.09/GB for the first 10 TB a month, falling through tiers to about \)0.05/GB above 150 TB. At 13 million GB almost all of the volume sits in the cheapest tier, so **the bill lands near \(650,000 a month**, or about \)8M a year.

Committed CDN. Negotiated CDN contracts at multi-petabyte monthly commitments are commonly quoted an order of magnitude below cloud list rates. Using that as a range rather than a quote, the same 13 PB lands somewhere in the tens of thousands of dollars a month. The gap is not a rounding difference; it is the difference between a viable and a non-viable business at this bitrate.

Which assumption dominates the error

Not the price. The bytes dominate, and the bytes are set by the encoding ladder. Viewer-hours multiplied by average delivered bitrate is the whole model, and average bitrate is a product decision: how many rungs the ladder has, what the default rung is on a cold start, and whether the player climbs to 1080p60 for a viewer who left the tab open. A 20% reduction in average bitrate through better per-title encoding saves more than a 20% price negotiation, and it is available without a contract.

Second in line is the offload ratio. If the edge serves 98% of segments from cache, origin egress is roughly 2% of delivered bytes. If a long tail of low-concurrency channels drops offload to 85%, origin egress multiplies sevenfold and appears as a cloud egress line nobody modelled.

What the number rules in or out

It rules out serving video directly from compute instances at any scale. It rules in a committed CDN contract before launch rather than after the first invoice, and it makes per-title encoding and ladder tuning an infrastructure-cost project rather than a video-quality project.

When this is the wrong answer

Below roughly 100 TB a month you have no negotiating position, contract minimums exceed the list-price bill, and the operational cost of a multi-CDN setup outweighs the saving. Choose list pricing and spend the effort on cache-hit ratio instead, and the decision flips only when sustained volume passes roughly a petabyte a month, because that is where a commit becomes cheaper than the minimum that buys it.

Common weak answers

• "Use a multi-CDN setup to play vendors off each other." Real at volume, but it splits the commit across contracts and each one prices worse because each commitment is smaller. It is a second-year move, not a first one.
• "Compress harder." Video is already compressed; the lever is the ladder and the default rung, not the codec setting alone, and a codec change costs transcode compute that has to be netted off.
• "Model peak, not average." Peak sets the commit tier and the capacity plan, and it is the wrong input for the invoice. Sizing the bill from 120 Gbps overstates it threefold, which fails in the direction nobody catches: the design gets rejected for a cost it does not have.