Bandwidth Transfer Cost: The Core Answer Up Front
If you are trying to size a bill, the shortest answer is this: bandwidth transfer cost equals the price of your network capacity (the ‘pipe’) plus the metered price of each gigabyte that leaves it. For example, moving 1 TB over a 1 Gbps link takes about 2.3 hours of saturated throughput, and if that data exits AWS to the public internet it costs roughly $92 at the standard $0.09/GB egress rate. In my first large migration, I moved 12 TB of video to cloud storage and was blindsided by a $1,100 egress charge because I had only budgeted the server instance.
What Bandwidth Transfer Cost Really Means (And Why Most Teams Get It Wrong)
The term bandwidth transfer cost conflates two distinct billing models that practitioners must separate. Bandwidth is a speed rating—measured in Mbps or Gbps—that you often lease as a reserved circuit. Data transfer is the volume of bytes actually moved, measured in GB or TB, and billed after the fact.
When I built a hybrid disaster-recovery link in 2019, I made the mistake of provisioning a 10 Gbps Direct Connect port for a one-week bulk sync. The port cost was fixed regardless of use, and the idle 95% of capacity still bled $0.30 per utilized GB. That taught me the hard rule: never size the pipe for the peak if the peak lasts an hour.
The thing nobody tells you about ‘unlimited bandwidth’ ISP plans is that they almost always contain a fair-use clause or a speed cap that effectively limits monthly transfer to a few TB. You are not buying infinite bytes; you are buying a best-effort speed with a silent ceiling.
Another misconception is that ingress (data into cloud) is free, so the reverse must be too. In reality, the return path from a user to your origin is someone else’s egress, and if you host at home, your ISP may charge for upload differently than download.
How Much Does Bandwidth Actually Cost? Breaking Down the Two Cost Layers
To answer ‘how much does bandwidth actually cost?’ you must split it into layer one (the pipe) and layer two (the volume). A dedicated business fiber circuit of 1 Gbps symmetric often runs $400–$800 per month in the US, which equates to about $0.01–$0.02 per GB if you saturate it 24/7. That is the effective bandwidth cost before any data leaves a cloud.
Layer two is the metered egress. According to the AWS S3 pricing page, the first 10 TB out to the internet in us-east-1 is $0.09 per GB. Google Cloud lists similar rates on its networking pricing page, generally $0.085–$0.12 depending on region. Ingress remains $0 on both platforms.
ISP and Transit Pricing Outside the Cloud
Non-cloud bandwidth pricing follows different rules. A regional business cable plan may advertise ‘1 Gbps’ for $99/month but enforce a 2 TB monthly cap; overage is $10 per extra 50 GB, effectively $0.20/GB—worse than AWS. Carrier Ethernet from a Tier 1 transit provider at 1 Gbps with 10 TB commit often prices at $0.03–$0.06 per additional GB. The key insight is that ISP ‘unlimited’ tiers are priced on the assumption of 95th-percentile burstable use, not constant saturation.
In my experience negotiating with a local IX, we secured 2 Gbps ports at $0.012/GB by committing to 20 TB/month minimum. That beats cloud egress by 7x but required a 12-month contract and our own routers.
If you want to skip the manual spreadsheet, our Bandwidth Transfer Calculator already encodes these tiers and will output a monthly estimate from just your user count and average file size. I keep it open during capacity planning sessions.
Self-hosted environments flip the model: you pay capital for switches and recurring for transit from a colocation provider, often $0.005–$0.03 per GB at high commit levels. The trade-off is you carry the risk of under-utilization and hardware refresh.
Transfer Time Math: How Long to Transfer 1TB Over 1Gbps?
The most common technical question I see is ‘how long to transfer 1TB over 1Gbps?’ The raw formula is seconds = (size_in_GB × 8) ÷ speed_in_Mbps. Using decimal terabytes, 1 TB = 1000 GB, and 1 Gbps = 1000 Mbps, so (1000 × 8) ÷ 1000 = 8,000 seconds, or 2.22 hours. If you use binary TiB (1024 GB), it is 2.28 hours.
Real-world throughput rarely hits theoretical max. TCP overhead, disk I/O wait, and packet loss typically add 5–15%. In a migration I ran last year, a 1 TB dataset actually took 2 hours 41 minutes because the source RAID could not feed the NIC faster than 880 Mbps sustained. Plan for the bottleneck that is not the network.
Here is a quick reference list for common transfers at a perfect 1 Gbps line:
- 100 GB: 800 seconds (~13 minutes)
- 500 GB: 4,000 seconds (~67 minutes)
- 1 TB: 8,000 seconds (~2.2 hours)
- 10 TB: 80,000 seconds (~22.2 hours)
Always multiply by 1.1 if you cannot confirm jumbo frames and window scaling are enabled end-to-end.
How Much Bandwidth for 100 Users? A Real-World Capacity Model
‘How much bandwidth for 100 users?’ has no single answer because user behavior drives the curve. A knowledge-worker profile (web, email, VoIP) averages 4–6 Mbps per seat during busy hour. Multiply by 100 and you need 400–600 Mbps of steady throughput, plus 30% headroom for bursts, putting you at roughly a 1 Gbps circuit.
Contrast that with 100 remote video editors each pulling 20 Mbps from a shared asset pool: that is 2 Gbps sustained, and a single 1 Gbps pipe would collapse. I once undersized a branch link for a creative team and the resulting packet delay caused our VoIP provider to drop calls—an outage whose cost dwarfed the bandwidth savings. If you need to quantify that risk, our Downtime Cost Calculator helps translate latency into dollars.
Use this quick seat-profile checklist before sizing:
- Light office (email/web): 3–5 Mbps per user
- VoIP + CRM: 6–8 Mbps per user
- HD video streaming: 8–12 Mbps per user
- Cloud workstation (DaaS): 15–25 Mbps per user
Sum the worst-case concurrent cohort, not the total headcount, then add 25% for protocol overhead.
The 1TB / 100-User Case Study: Converting Speed and Headcount into Dollars
Let’s fuse the two threads: a business with 100 average users (6 Mbps each) and a one-time 1 TB data migration to cloud. First, steady-state load: 100 × 6 Mbps = 600 Mbps, which fits a 1 Gbps fiber at $500/month. Monthly transfer volume if each user pulls 500 MB/day is 100 × 0.5 GB × 30 = 1,500 GB out from cloud to users.
Now the cost layers. On AWS, the 1 TB migration egress is 1024 × $0.09 = $92.16. The monthly 1.5 TB to users is 1536 × $0.09 = $138.24. Add the $500 pipe if you count local last-mile. Self-hosted in a colo with 1 Gbps transit at $0.02/GB: migration costs $20.48, monthly user traffic $30.72, but you pay $300/month colo port plus hardware.
| Scenario | 1TB Migration Cost | Monthly 1.5TB User Cost | Fixed Pipe/Port | Total First Month |
|---|---|---|---|---|
| AWS S3 egress + business fiber | $92 | $138 | $500 | $730 |
| Self-hosted colo 1Gbps | $20 | $31 | $300 | $351 |
| GCP egress (assumed $0.085/GB) | $87 | $131 | $500 | $718 |
The table shows why high-volume shops shift to self-hosted or CDN offload after a break-even point near 5–8 TB/month. But cloud wins for bursty startups with low baseline.
How Much Does It Cost to Do a Data Transfer? Hidden Line Items
Beyond the headline per-GB rate, ‘how much does it cost to do a data transfer?’ must include request fees, replication, and gateway hours. AWS charges $0.0004 per 1,000 GET requests; a 1 TB backup composed of 10 million 100 KB files incurs $4 in requests—but if those files are 1 KB, you get 1 billion objects and $400 in request fees before bandwidth.
Cross-region transfer inside the same cloud is cheaper than internet egress but not free—often $0.02–$0.04/GB. NAT gateway hourly charges and per-GB processing on AWS add another $0.045/GB in my production VPCs. I once migrated a 2 TB database that looked like $180 on paper but landed at $610 after inter-AZ and NAT fees.
The newest hidden tax is IPv4 address charges; AWS now bills $0.005/hour per public IPv4, which sounds tiny until you run 50 NAT IPs for a year ($219). These line items are why a raw bandwidth transfer cost estimate is only 60–70% of the real invoice.
A Practitioner’s Framework: The Bandwidth-to-Cost Translation Matrix
To systematically convert technical specs into finance language, I use a three-cell matrix that you will not find in vendor docs. It forces you to capture every dollar.
| Layer | Input Metric | Cost Driver | Example Formula |
|---|---|---|---|
| 1. Pipe Reservation | Gbps leased | Fixed monthly port | $500 / 30 days = $16.67/day |
| 2. Volume Transfer | GB out | Metered egress | 1500 GB × $0.09 = $135 |
| 3. Incidentals | Requests, NAT, IPs | API + gateway | (Requests/1000×$0.0004)+$0.045/GB |
Apply this matrix to any project: fill layer 1 even if zero (self-hosted capex amortized), layer 2 from your transfer-time math, layer 3 from the specific service’s pricing sheet. The output is a defensible number for stakeholders.
As a checklist, verify: (a) Are you using decimal or binary GB? (b) Did you include the return path? (c) Did you simulate TCP overhead? (d) Did you cap concurrent users, not total? If any answer is no, rerun the estimate.
Common Mistakes and Edge Cases I’ve Hit in Production
Even with the matrix, edge cases bite. Here are the top four I have debugged:
- TCP window scaling disabled on legacy VMs capped real throughput at 250 Mbps despite 1 Gbps NIC—transfer time quadrupled.
- Cloud instance vCPU credits exhausted, throttling the network stack; the pipe was fine, the VM was not.
- Metered switch ports at a colo billed on 95th percentile, not average, so a nightly backup spike set the monthly rate.
- CDN middle-mile repatriation fees appeared when users in EU pulled from US origin, adding inter-region cost despite ‘free caching.’
None of these show up in a basic bandwidth calculator, which is why experience trumps the first Google result. The most people don’t realize that the network edge is where budgets die.
When to Use Cloud Egress vs. Self-Hosted Bandwidth
Choosing between paying cloud egress and running your own pipe is a trade-off of elasticity versus unit cost. Cloud is superior when your monthly transfer is below ~5 TB and spikes unpredictably; you avoid capex and get global anycast. Self-hosted wins when baseline exceeds 10 TB/month and you can commit to a 12-month transit contract.
In a hybrid model, I keep hot assets in cloud with a tight TTL and bulk cold storage on owned hardware. That cut one client’s bandwidth transfer cost from $4,200/month to $1,300, but required a weekend of rsync tuning. There is no silver bullet—only matched architecture.
Putting It Together: Your Step-by-Step Calculation Process
Follow this sequence for any new workload to forecast spend accurately:
- Profile users: count concurrent seats and multiply by per-seat Mbps from the checklist above.
- Estimate volume: average file size × daily pulls × 30 to get monthly GB out.
- Compute transfer time: use the (GB×8)÷Mbps formula, add 10% overhead.
- Map cost layers: pipe reservation + metered GB + request/NAT fees using the matrix.
- Validate with our Bandwidth Transfer Calculator before sending the budget to finance.
If you do only one thing after reading, open your last cloud bill and split the network line item into those three layers. The clarity will change how you negotiate with providers.