Megabits per second to Gigabits per second (Mbps to Gbps)

One megabit per second (Mbps) equals 1,000,000 bits transmitted per second under the SI decimal convention used universally by network engineers, ISPs, and standards bodies — not the binary 2²⁰ that the storage-prefix convention implies for bytes. The IEC and IEEE both treat "Mbps" as 10⁶ bps in IEEE 802.3 (Ethernet), 802.11 (Wi-Fi), 3GPP cellular standards, ITU-T G.9961 powerline, DOCSIS cable, and the relevant ITU-R radio-spectrum recommendations. The Mbps-to-MB/s conversion is exact: 1 Mbps ÷ 8 bits/byte = 0.125 MB/s, so a 100 Mbps connection delivers a maximum of 12.5 MB/s before TCP/IP, Ethernet, and link-layer protocol overhead reduces effective throughput by 5–15% to roughly 11.0–11.9 MB/s. The symbol distinguishes case carefully: Mbps (uppercase M, lowercase b, lowercase ps) is megabits per second; MB/s (uppercase M, uppercase B, slash s) is megabytes per second; the 8:1 ratio between the two is identical to the bit/byte distinction, and is the most consequential unit-conversion in consumer technology.

Megabits per second became the consumer-facing unit of internet speed in the late 1990s and has remained the dominant marketing-and-regulatory unit through every successive broadband generation. Cable modem rollouts under the DOCSIS 1.0 specification published by CableLabs in March 1997 advertised peak shared-segment capacity of 38 Mbps downstream, with per-subscriber tiers initially marketed at 1.5 Mbps and 3 Mbps; ADSL deployments by US RBOCs and European incumbents through the same period advertised 1.5–8 Mbps downstream. The unit was inherited from earlier wide-area networking — the T1 carrier specification standardised by AT&T in the 1960s ran at 1.544 Mbps, and the European E1 carrier at 2.048 Mbps — and from the IEEE 802.3 Ethernet original 10 Mbps shared-coaxial specification of 1983. The Wi-Fi family carried the convention forward: 802.11b (1999) at 11 Mbps, 802.11g (2003) at 54 Mbps, 802.11n (2009) at 150–600 Mbps, 802.11ac (2013) at 433–6,933 Mbps, 802.11ax / Wi-Fi 6 (2019) at up to 9.6 Gbps aggregate, 802.11be / Wi-Fi 7 (2024) at 46 Gbps theoretical peak. Cellular followed the same path: 3G HSPA at 14–42 Mbps, 4G LTE Cat 4 at 150 Mbps and Cat 16 at 1 Gbps, 5G NR sub-6 GHz at hundreds of Mbps, 5G mmWave at theoretical multi-Gbps peaks. Regulatory definitions of "broadband" have tracked the consumer-marketing tier with a lag — the FCC defined broadband as 4/1 Mbps in 2010, raised it to 25/3 Mbps in 2015, and to 100/20 Mbps in March 2024, the threshold below which a connection is no longer counted toward the National Broadband Map's "served" status.

Residential broadband plans denominate every advertised tier in Mbps or Gbps. Comcast Xfinity 2026 retail tiers run from 75 Mbps Connect through 1,200 Mbps Gigabit and 2,000 Mbps Gigabit X2; Spectrum runs 300/500/1,000 Mbps tiers; Verizon Fios runs 300/500/1,000/2,000 Mbps fibre-to-the-home tiers; AT&T Fiber runs 300/500/1,000/2,000/5,000 Mbps tiers. The marketed downstream rate is what every ISP's sales material leads with, the regulatory upload-rate disclosure (under the FCC Broadband Nutrition Label rules effective April 2024) is the second-line figure, and the typical-during-peak-hours figure is the third-line disclosure that the FCC requires in 8-point or larger type. Comparison shopping at the household level — Cox versus Spectrum versus T-Mobile Home Internet — is conducted entirely in Mbps until the gigabit-tier transition. Streaming-service minimum-bandwidth requirements published by Netflix, YouTube, Disney+, Apple TV+, Hulu, and Max all denominate in Mbps. Netflix recommends 3 Mbps for SD, 5 Mbps for 1080p HD, 15 Mbps for 4K UHD; YouTube recommends 3 Mbps for 1080p, 20 Mbps for 4K HDR; Apple TV+ 4K Dolby Vision streams at sustained ~25 Mbps peaks. The "minimum required Mbps" figure published in every help-centre article is the speed-test threshold consumers run against to diagnose buffering — the typical Speedtest.net or Fast.com result expressed in Mbps download / Mbps upload — and the household-Wi-Fi troubleshooting workflow (router placement, channel selection, mesh-node sizing) is conducted entirely in Mbps observed at the device. Wi-Fi router and mesh-system marketing collapses all per-band capacities into a single aggregate-Mbps figure (AX5400, AX6000, BE9300, BE19000) that nominally sums the 2.4 GHz, 5 GHz, and 6 GHz radio capacities — a marketing-aggregate that no single device ever achieves because connections use one band at a time. The actual per-device sustained Mbps is determined by client radio capability (Wi-Fi 5 vs 6 vs 6E vs 7), spatial stream count, channel-bandwidth selection (20/40/80/160/320 MHz), and signal-strength-dependent modulation/coding rate. The gap between the printed-on-the-box AX5400 and the device-facing 600 Mbps is roughly 9× and is one of the larger consumer-marketing inflation factors in any product category. Cellular speed-tier marketing follows the same Mbps convention. T-Mobile, Verizon, and AT&T all publish "typical 5G download" Mbps ranges by market, and the FCC's Mobile Broadband Performance Report quarterly publishes nationwide Mbps medians. 5G NR sub-6 GHz typical user-plane throughput in 2026 is 150–500 Mbps in dense-urban deployments and 50–150 Mbps in rural; 5G mmWave urban hot-spot delivery exceeds 1,000 Mbps; LTE-Advanced legacy delivery sits at 25–100 Mbps where re-farmed for 5G overlap.

One gigabit per second (Gbps) equals 1,000,000,000 bits transmitted per second under the SI decimal convention used universally by IEEE, ITU-T, 3GPP, and OIF standards. The Gbps-to-GB/s conversion is exact: 1 Gbps ÷ 8 bits/byte = 0.125 GB/s = 125 MB/s of theoretical maximum file-transfer throughput, with TCP/IP, Ethernet, and link-layer overhead reducing practical effective throughput by 5–15% to roughly 110–120 MB/s under favourable conditions. The symbol Gbps (uppercase G, lowercase b, lowercase ps) is distinct from GB/s (uppercase G, uppercase B, slash s) by the same factor of 8 that separates bits from bytes at every prefix tier. Gbps is the standard unit for Ethernet-and-Fibre-Channel switching-fabric capacity (IEEE 802.3 family), for PCIe-link bandwidth in modern host bus adapters and NICs, for cellular peak-data-rate quotes (3GPP Release 16 5G mmWave at 20 Gbps theoretical peak), and for the per-wavelength capacity of coherent DWDM optical transponders. At the multi-Tbps scale of modern undersea-cable systems, Gbps is the unit of single-wavelength capacity from which aggregate per-pair and per-cable Tbps figures are summed.

Gigabits per second became the working unit of enterprise and data-centre networking with the IEEE 802.3z Gigabit Ethernet standard ratified in June 1998 — a 1 Gbps full-duplex link over fibre or copper that displaced 100BASE-TX (Fast Ethernet) as the LAN-server-uplink baseline through the early 2000s and as the desktop wired-Ethernet baseline by 2010. The 10 Gigabit Ethernet standard (IEEE 802.3ae) followed in June 2002, then 40 GbE and 100 GbE (IEEE 802.3ba, 2010), 25 GbE (IEEE 802.3by, 2016), 200 GbE and 400 GbE (IEEE 802.3bs, 2017), and 800 GbE (IEEE 802.3df, 2024) — the rate ladder that hyperscaler data-centre fabrics from AWS, Azure, Google Cloud, and Meta have followed for the leaf-and-spine switching topologies underneath every cloud workload. The cultural inflection that pushed gigabit out of enterprise specification sheets and into consumer expectation came on 26 July 2012, when Google Fiber went live in Kansas City, Kansas with a 1 Gbps symmetric residential plan at $70/month — at a time when the median US residential broadband connection was 6 Mbps and incumbent ISPs offered "extreme" tiers at 50–100 Mbps. Google Fiber expanded to a handful of additional metros (Austin, Provo, Atlanta, Charlotte, Nashville) before construction was paused in 2016, but the competitive pressure forced AT&T, Comcast, Verizon Fios, and Cox to roll out their own gigabit residential tiers across most major US markets between 2014 and 2020, and 1 Gbps residential FTTH had become the default new-build target by 2024 with 2/5/10 Gbps consumer tiers now available from AT&T Fiber, Frontier, Ziply, and Sonic. The defining gigabit-per-second platform of the 2020s is the long-haul DWDM fibre-optic network. Coherent 400 Gbps and 800 Gbps single-wavelength transponders introduced by Ciena, Infinera, and Nokia between 2020 and 2024 multiplexed 80–96 wavelengths per fibre pair to deliver 32–76.8 Tbps per pair, the capacity that anchors the modern global undersea-cable buildout. The MAREA Microsoft–Facebook–Telxius transatlantic cable went live in 2018 at 160 Tbps and was upgraded to 224 Tbps in 2022; the Google-funded Equiano cable from Portugal to South Africa launched in 2022 at 144 Tbps; the Meta-funded 2Africa cable, the longest in the world at 45,000 km, completed its first segment in 2023 at 180 Tbps. 5G NR theoretical peaks under the 3GPP Release 16 specification reach 20 Gbps in mmWave deployments under ideal conditions, with Release 18 (5G-Advanced, 2024) raising the peak to 50 Gbps.

Data-centre fabric switching is the gigabit-per-second unit's most active 2026 domain. Hyperscaler leaf-and-spine fabrics at AWS, Azure, Google Cloud, and Meta are now built around 400 Gbps and 800 Gbps switch-to-switch trunks, with 200 Gbps and 100 Gbps to-server downlinks on top-of-rack switches. Arista, Cisco Nexus, NVIDIA Spectrum and Juniper QFX product lines all spec their port densities and aggregate switching capacity in Gbps and Tbps; a typical 51.2 Tbps switch ASIC (Broadcom Tomahawk 5, NVIDIA Spectrum-4) drives 64× 800 Gbps ports or 128× 400 Gbps ports at line rate. Cloud-customer-facing virtual-NIC bandwidth tiers, AWS placement-group inter-AZ throughput limits, and Azure ExpressRoute / Google Cloud Interconnect dedicated-circuit options all denominate in Gbps tiers (1, 10, 40, 100 Gbps). GPU-accelerator interconnects sit at the high end of the Gbps tier. NVIDIA NVLink generation 4 delivers 900 GB/s aggregate per H100 GPU across 18× 50 Gbps lanes (450 Gbps unidirectional or 900 GB/s bidirectional in NVLink-language); the InfiniBand NDR (400 Gbps) and XDR (800 Gbps) standards from the InfiniBand Trade Association anchor AI-training-cluster east-west traffic; the AMD Infinity Fabric on MI300X-class accelerators sustains 5.3 TB/s of HBM3 bandwidth per package. The 2024–2026 AI-training data-centre buildout has driven Gbps-tier per-port pricing down by an order of magnitude relative to general-purpose-cloud networking. Long-haul fibre-optic backbones, undersea cables, and metro-DWDM systems are denominated in per-wavelength Gbps and per-cable Tbps. Coherent 400 Gbps and 800 Gbps single-wavelength DWDM transponders from Ciena, Infinera, Nokia, and Cisco-acquired Acacia Communications now dominate new long-haul deployments; legacy 100 Gbps wavelengths from the 2010s remain in service across regional and metro networks. Telecommunications carriers (Lumen, Zayo, GTT, Telia Carrier, Tata Communications) sell wavelength-services in 100, 400, and 800 Gbps tiers, IP-transit in 10 and 100 Gbps port-rate tiers, and Ethernet Private Line services across the same tier ladder. Cellular peak-data-rate marketing has crossed into Gbps territory under 5G NR. T-Mobile, Verizon, and AT&T all publish "5G Ultra Wideband" or "5G+" mmWave-deployment Gbps peaks in dense-urban hot spots, with verified consumer-device achievements in the 1–4 Gbps range under line-of-sight conditions. The 3GPP Release 16 specification fixes the 5G NR theoretical peak at 20 Gbps download / 10 Gbps upload; Release 18 ("5G-Advanced", finalised June 2024) raises the peak to 50 Gbps with new spectrum bands and 8×8 MIMO configurations. Storage and peripheral interface standards have converged on the Gbps tier alongside networking. USB4 (announced August 2019, USB-IF) runs at 20 Gbps or 40 Gbps full-duplex per port over USB-C connectors and is the basis for Thunderbolt 4, which the Intel-led Thunderbolt programme standardised at 40 Gbps with mandatory PCIe-tunnelling and DisplayPort-tunnelling capabilities; Thunderbolt 5, announced September 2023 and shipping on Apple M4 Pro and M4 Max devices in 2024, runs at 80 Gbps symmetric and 120 Gbps asymmetric ("Bandwidth Boost" mode) for high-resolution external-display workloads. NVMe SSD interfaces have followed PCIe generation: a PCIe 4.0 ×4 NVMe drive runs at 64 Gbps raw signalling (~7 GB/s practical sequential reads); a PCIe 5.0 ×4 NVMe drive runs at 128 Gbps raw (~14 GB/s); a PCIe 6.0 ×4 NVMe drive at 256 Gbps raw. The convergence has eliminated the historical separation between "networking-fast" and "storage-fast" — both categories now sit in the 40–800 Gbps band and are bottlenecked by the same physical-layer signalling generations (NRZ vs PAM4, retimer-and-redriver count, copper-vs-fibre PHY). Consumer-grade Wi-Fi marketing has crossed the Gbps boundary on every premium router. Wi-Fi 6 (802.11ax) AX6000 and AX11000 routers advertise 6 Gbps and 11 Gbps aggregate radio capacities; Wi-Fi 7 (802.11be) BE19000 and BE33000 routers advertise 19 Gbps and 33 Gbps aggregate; the per-device sustained throughput at the client end is one band at a time and tops out at 2–3 Gbps even on the highest-tier 320 MHz Wi-Fi 7 channels — the same aggregate-vs-sustained gap that mbps-tier routers show, scaled up.