Internet Speeds Explained: Mbps, MB/s, and What Your Plan Actually Delivers

Internet bandwidth is measured in bits per second (bps), with the most common consumer-facing units being megabits per second (Mbps) for residential broadband and gigabits per second (Gbps) for higher-tier fibre and enterprise circuits. The bit unit traces back to the earliest telegraph and modem signalling, where the underlying serial transmission of binary symbols gave the bit-rate its natural physical unit. Modern Ethernet, fibre-optic, and wireless networking all preserve the bit-rate convention because the underlying physical layer is still bit-by-bit transmission, even though end-user file-transfer experiences are typically described in bytes per second (MB/s) on download progress bars. The 8-bits-per-byte conversion at the boundary is the most common source of consumer-facing speed confusion.

A 1 Gbps internet plan provides a 1,000,000,000 bits-per-second signalling rate at the physical layer. Because there are 8 bits in a byte, the theoretical maximum byte throughput is 1,000,000,000 ÷ 8 = 125,000,000 bytes per second, or 125 MB/s. After Ethernet, IP, and TCP protocol overhead — frame headers, packet headers, segment headers, inter-frame gaps — the real-world usable throughput is typically 940–960 Mbps, or about 117–120 MB/s in byte terms. Speed-test apps like Ookla and Fast.com measure this real-world figure rather than the gross link rate, which is why a "gigabit fibre" customer typically sees 940 Mbps on a clean speed test rather than the advertised 1000 Mbps.

The distinction between bits-per-second and bytes-per-second is the single most common cause of internet-speed confusion. Mbps (megabits per second, lowercase b) is what ISPs advertise; MB/s (megabytes per second, uppercase B) is often what download progress bars and file-transfer applications display. The two differ by a factor of 8: a 100 Mbps plan delivers 12.5 MB/s at theoretical maximum, and a 1 Gbps plan delivers 125 MB/s. A "1 GB file on a 100 Mbps connection" takes 1,000,000,000 × 8 ÷ 100,000,000 = 80 seconds at theoretical maximum, not 10 seconds (the wrong answer that comes from conflating bits with bytes). The 8× factor is preserved in every bandwidth-to-throughput conversion and matters at every download-time estimate.

Modern internet protocols carry substantial overhead. An Ethernet frame adds 18 bytes of frame header, footer, and inter-frame gap to every packet payload. An IP header adds 20 bytes (IPv4) or 40 bytes (IPv6) per packet. A TCP header adds another 20 bytes. With a typical 1500-byte Ethernet payload, the protocol headers consume about 4% of the link capacity for IPv4-over-TCP traffic, leaving 96% of the bit rate for application data. UDP-based traffic (some streaming, gaming, DNS) saves a small amount of TCP overhead but still loses 3–4% to lower-layer headers. The 940 Mbps real-world figure on a 1 Gbps link is the protocol-overhead-adjusted maximum, achievable on a clean local-area network with healthy TCP flows.

Video streaming bitrates set the bandwidth requirements for popular services. Netflix HD (1080p) streams at about 5 Mbps; Netflix 4K streams at 15–25 Mbps depending on the encoding and HDR mode. YouTube HD ranges from 2.5 to 8 Mbps depending on resolution and codec. Disney+, HBO Max, and Apple TV+ cluster in similar ranges. A 25 Mbps internet plan handles a single 4K stream comfortably; a 100 Mbps plan handles four simultaneous 4K streams plus background video calls and casual browsing; a 1 Gbps plan supports far more concurrent users than any typical home generates and is bandwidth-overkill for almost all residential streaming use cases.

Internet plans are typically asymmetric, with the upload speed substantially lower than the download speed. A "1 Gbps down, 50 Mbps up" plan is common because consumer traffic is mostly download-heavy (web browsing, video streaming, app downloads). Symmetric plans matter for video conferencing (Zoom, Teams, Google Meet need 1.5 to 4 Mbps upload for HD video), cloud-backup services (iCloud Photo Library, Google Photos, Backblaze need sustained upload to keep up with new content), and remote work involving file sharing or screen sharing. Fibre-to-the-home connections often offer symmetric Gbps service; cable internet is typically asymmetric with download speeds far higher than upload.

Speed-test results depend on more than just the line rate. The end-to-end path quality matters: a 1 Gbps fibre connection might be limited by a 100 Mbps server-side bottleneck during peak hours. Wi-Fi performance is usually the limiting factor in residential settings, with even modern Wi-Fi 6 typically delivering 200–500 Mbps to a phone or laptop despite a 1 Gbps wired connection. Router-and-switch performance, congestion at the ISP's peering points, and the geographic distance to the test server all affect the measured throughput. A 1 Gbps plan rarely delivers 940 Mbps to every device on the home network simultaneously; it delivers that throughput to a single wired device under good conditions.

Match the plan to the household's peak concurrent demand rather than aspirational use cases. A single-occupant home with moderate streaming and remote work needs about 100–200 Mbps; a four-person family with multiple 4K streams, video calls, gaming, and cloud backup typically wants 300–500 Mbps; a high-bandwidth household with multiple work-from-home users, content creators uploading large files, and frequent large-file downloads benefits from 1 Gbps. Anything above 1 Gbps is rarely useful for residential use because most servers and content sources cannot deliver content faster than 1 Gbps to a single connection regardless of the local link rate.

A speed-test result reports three numbers: download speed in Mbps, upload speed in Mbps, and ping (or latency) in milliseconds. The Mbps figures show the real-world throughput your connection actually delivers under the test conditions. The ping figure shows the round-trip time to the test server, with values under 30 ms typical for fibre, 30–50 ms typical for cable, and over 100 ms suggesting a problematic connection or distant server. To convert a Mbps speed-test number to MB/s for download-time estimation, divide by 8: a 940 Mbps test result means about 117 MB/s of real-world byte throughput.