How to Transfer Terabytes of Data Over the Internet (2026 Guide)

On paper, a 1 Gbps line moves 10 TB in about 23 hours. Over an 80 ms route with 0.01% packet loss, a modeled single TCP stream would need roughly 40 days. When teams transfer terabytes of data between sites, clouds or continents, the bottleneck is rarely the bandwidth they pay for. It is how the transfer protocol reacts to latency and loss, which is why purpose-built enterprise file transfer software exists.

Speed is only half the job. Terabyte workloads tend to repeat: nightly render outputs, sequencing runs, backup replication, training datasets handed to AI teams. Keeping those files accessible and synchronized across teams is the domain of Enterprise file sync and share, while automating their delivery with scheduling, retries, integrity checks and audit trails falls under Managed File Transfer. This guide gives you calculated transfer times, the five realistic ways to move the data, and a plan for choosing.

Add LawBhoomi as your preferred source on Google.

Add Now →

TL;DR

• At line rate, 10 TB takes about 23.4 hours on 1 Gbps and 2.3 hours on 10 Gbps (decimal TB, standard Ethernet overhead).
• A single TCP stream is capped by latency and loss, not bandwidth: at 80 ms and 0.01% loss, the RFC 9438 CUBIC model gives about 23 Mbps on any link.
• Parallel streams and UDP-based acceleration are the two ways to fill a long, lossy pipe.
• Ship hardware only when the network math loses. AWS Snowball Edge closed to new customers on November 7, 2025; Azure Data Box and Google Transfer Appliance are still offered.
• Check storage too: 10 Gbps at line rate needs about 1.19 GB/s of sustained reads and writes.

How long does it take to transfer 10 TB over the internet?

At full utilization, 10 TB takes about 23.4 hours over 1 Gbps, 2.3 hours over 10 Gbps and almost 10 days over 100 Mbps. Those figures assume the transfer actually fills the link, which a single TCP stream on a long route usually does not.

How we calculated it: time = 8 × 10¹³ bits ÷ effective throughput. 10 TB is decimal (10¹³ bytes); 10 TiB is about 10% larger. The 94.9% ceiling is TCP payload on Ethernet with a 1500-byte MTU (1,460 of 1,538 bytes on the wire). “Sustained 80%” is an assumption. TCP columns use the CUBIC response function in RFC 9438 with a 1,460-byte segment size and 0.01% loss. These are models, not measured benchmarks.

Look at the TCP columns: they do not change with link speed. At 80 ms, the modeled stream uses about 0.2% of a 10 Gbps line.

Explore LawBhoomi's free law library for students and lawyers.

Visit Library →

What are the realistic ways to transfer terabytes of data?

There are five practical routes: single-stream tools, parallel cloud upload tools, managed cloud migration services, UDP-based accelerated transfer software, and physical transfer appliances. The right one depends on route latency, how often the job repeats, and where the data lands.

First, separate two jobs. Automated, repeatable pipelines belong in transfer or MFT tooling, while Enterprise File Sync and Share platforms are built for people syncing, editing and sharing files together.

AWS says a single DataSync task can fully utilize a 10 Gbps link between on-premises storage and AWS. Accelerated tools move the same bytes over UDP and add their own reliability layer: IBM’s documentation shows Aspera’s FASP transfers on UDP port 33001 by default, and Signiant describes replacing TCP with its UDP-based protocol.

Physical shipping is shrinking, not gone. AWS no longer offers Snowball Edge to new customers and points them to DataSync or Data Transfer Terminals: facilities in US metros, London, Munich, Paris, Tokyo and Sydney where you connect your own drives over 100G fiber. Microsoft’s next-generation Data Box holds 120 TB or 525 TB usable, and Google’s Transfer Appliance ships in 40 TB to 300 TB configurations.

Does the same math hold for petabyte data transfer?

Yes: transfer time scales linearly with volume, so 1 PB takes about 9.8 days at 10 Gbps line rate and about 23.4 hours at 100 Gbps. At that scale, storage throughput, file counts and the number of parallel transfer nodes matter as much as the protocol. Plan the move as a pipeline across several machines, not as one giant copy job.

Why does latency slow down a single TCP transfer?

Loss-based TCP sends about one window of data per round trip and shrinks that window after each loss event, so its throughput falls as RTT and packet loss rise. For Reno-style TCP, doubling the RTT or quadrupling the loss rate halves throughput.

The classic model from Mathis et al. (1997) is throughput ≈ (MSS ÷ RTT) × (1.22 ÷ √p). CUBIC, now the default in the Linux, Windows and Apple stacks, recovers faster on long paths but still degrades under loss. At 80 ms, the modeled stream falls from 130 Mbps at 0.001% loss to 23 Mbps at 0.01% and 5.7 Mbps at 0.1%.

Real networks show the same shape. On a 90 ms path with 0.0046% loss and 9,000-byte jumbo frames, ESnet measured 490 Mbps in the lossy direction against 8.2 Gbps in the clean one. Window size matters too: filling 1 Gbps at 80 ms requires 10 MB in flight, and 10 Gbps requires 100 MB, so OS buffers must allow it.

Explore LawBhoomi's free law library for students and lawyers.

Visit Library →

Splitting the job across 16 streams cuts the modeled time from 40 days to 2.5 days. UDP-based protocols take a different path: they handle retransmission and rate control themselves, so throughput no longer depends on TCP’s window-per-round-trip limit.

How to transfer terabytes of data: a 6-step plan

1. Measure the path. Record RTT and packet loss over several minutes with tools such as ping and iperf3, not just the contracted bandwidth.
2. Run the math. Compare size ÷ (link × utilization) with the single-stream figures above, or see Raysync’s guide to file transfer time calculations.
3. Check storage and file shape. Both ends must sustain the target rate, and millions of small files add per-file overhead; bundle them or use a tool that pipelines files.
4. Pick the method. Short route, one-off job: rsync or a parallel CLI. Long or lossy route, recurring job: UDP-based acceleration. Hundreds of terabytes behind a slow uplink: a physical appliance.
5. Protect the data. Encrypt in transit, verify checksums, and require checkpoint resume so a dropped session never restarts at zero.
6. Automate repeats. Schedule or trigger jobs on events, alert on failure, and keep logs for audits.

How Raysync Enterprise handles terabyte-scale transfers

When 10 TB must cross an ocean every week, the line is rarely the problem. A single TCP stream leaves most of it idle, and a dropped session without resume restarts the clock. Raysync Enterprise replaces that pattern with proprietary UDP-based WAN acceleration, intelligent UDP/TCP switching and peer-to-peer transfer between endpoints.

Checkpoint resume and transfer verification keep long jobs intact. AES-256 encryption protects data in transit and at rest, with TLS that supports custom and client certificates. Scheduled or event-triggered one-way and two-way sync, centralized administration with role-based access control and audit logs, plus REST API, SDK and CLI access, turn one-off moves into repeatable pipelines. Raysync Enterprise is TPN certified, a DPP member, and HIPAA and ISO 27001 certified. For multi-site moves beyond the terabyte range, see Raysync’s petabyte-scale data transfer solution. Pricing is customized, and a free trial is available.

Frequently asked questions

How long does it take to transfer 1 TB over the internet? One terabyte takes about 2.3 hours at 1 Gbps line rate and about 23 hours at 100 Mbps. Over an 80 ms route with 0.01% loss, a single TCP stream needs roughly four days, based on the RFC 9438 CUBIC model.

What is the fastest way to transfer terabytes of data? The fastest online method is usually one that keeps the link full: UDP-based accelerated software or many parallel streams. For hundreds of terabytes behind a weak uplink, a physical appliance can win. Measure RTT and loss before choosing.

Is it faster to ship hard drives than to upload? It depends on link speed and volume. Moving 100 TB takes about 9.8 days at 1 Gbps line rate but about 23 hours at 10 Gbps. Compare that with the provider’s end-to-end device turnaround, including ordering, shipping and ingest.

Why is my transfer slower than my internet speed? A single TCP stream is limited by round-trip time and packet loss, not just bandwidth. Storage throughput, small-file overhead, VPN encryption and undersized TCP buffers also cap speed. Test the path with iperf3 to locate the bottleneck before changing tools.

Is UDP file transfer reliable and secure? UDP itself guarantees nothing, so accelerated transfer tools add retransmission, ordering and integrity checks on top. Security comes from the application layer, typically AES-256 encryption plus authentication, so judge the product’s implementation rather than the transport protocol.

Conclusion

To transfer terabytes of data quickly, stop treating bandwidth as the only number that matters. Measure latency and loss, do the math, and pick a method that can actually fill the pipe: parallel tools for short hops, UDP-based acceleration for long recurring routes, and appliances only when the network loses. If your terabytes cross regions every week, start a free trial of Raysync Enterprise and test it on your own route.

Explore the latest legal opportunities for law students and lawyers.

Explore Opportunities →