Terabit per second to Terabyte per second

Global internet traffic peaks at roughly 1,000–1,500 Tbps (1–1.5 Pbps) as of 2026. This is growing at about 25% per year, driven by video streaming, cloud computing, and AI training data transfers. A single viral live event can spike regional traffic by tens of Tbps.

Internet traffic automatically reroutes through other cables and paths via BGP routing protocols, usually within seconds. Speed may degrade in the affected region but rarely drops entirely. Cable cuts happen more often than people think — about 100 per year globally, mostly from ship anchors and fishing trawlers.

Dense wavelength-division multiplexing (DWDM) sends dozens of different light colors (wavelengths) through a single fiber simultaneously, each carrying its own data stream. A modern cable contains multiple fiber pairs, each carrying 100+ wavelengths, with each wavelength modulated at 400 Gbps or more.

Netflix's library is estimated at around 30–40 petabytes. At 1 Tbps, downloading the entire catalog would take roughly 70–90 hours. At 100 Tbps (a realistic submarine cable capacity), you could theoretically grab all of Netflix in under an hour.

Researchers at Japan's NICT achieved 22.9 Pbps (22,900 Tbps) through a single multicore fiber in 2024. That is enough to transfer the entire Library of Congress in a fraction of a second. These lab records typically reach commercial deployment 5–10 years later.

Large language models have billions of parameters that must be read from memory for every inference pass. An LLM with 70 billion parameters at 16-bit precision needs 140 GB of data read per forward pass. At 3 TB/s, the H100 can perform roughly 20 inference passes per second — bandwidth directly determines tokens-per-second output.

During LLM inference each token requires reading all model weights from memory. A 70-billion-parameter model at 16-bit precision means 140 GB read per forward pass. At 30 tokens per second, that is 4.2 TB/s of memory reads — right at the limit of an H100's HBM3. This is why AI inference is "memory-bound": the GPU's compute cores sit idle waiting for data. Quantising weights to 8-bit or 4-bit halves or quarters the bandwidth demand, directly increasing tokens per second.

The NVIDIA B200 GPU with HBM3e achieves approximately 8 TB/s of memory bandwidth as of 2025. Each generation roughly doubles bandwidth — from 2 TB/s (A100) to 3.35 TB/s (H100) to 4.8 TB/s (H200) to 8 TB/s (B200). The trajectory suggests 16+ TB/s within a few years.

About 16.7 minutes. A petabyte is 1,000 terabytes, so at 1 TB/s, the math is simple division. For context, the Library of Congress contains roughly 10–20 petabytes of data. Transferring it all at 1 TB/s would take about 3–6 hours.

Yes — petabytes per second (PB/s). Experimental optical interconnects and photonic computing architectures are pushing toward PB/s-class bandwidth. Some supercomputer storage systems already aggregate into the PB/s range when all nodes operate simultaneously. It is the next frontier for AI training clusters.