Fiber vs Cable vs DSL: What Each Actually Delivers

A gigabit plan on cable and a gigabit plan on fiber are the same sentence in a sales script. In the last federal measurement of both, the advertised upload speeds behind those two words sat thirty-two times apart.

No retention agent can fix that, and no tier change will either, because the number is a property of the wire rather than of your account. This is the part that goes missing when a shopping guide sorts providers by price. If you are already subscribed and trying to work out why your line behaves the way it does, the useful question is not which technology is best. It is which one you actually have, what its ceilings are, and which of your complaints sit above those ceilings.

The upstream number belongs to the plant, not to your plan

Start with the measurement that separates the technologies most sharply.

The FCC's Thirteenth Measuring Broadband America report, built on data collected in September and October 2022, computed the weighted mean advertised upload speed among the top 80 percent of service tiers, by technology: 528 Mbps for fiber, 16.6 Mbps for cable, 9.5 Mbps for DSL. The report gives the reason in one sentence. Fiber has enough capacity to be engineered symmetrically, while "for other technologies with more limited capacity, higher capacity is usually allocated to download speeds than to upload speeds, typically in ratios ranging from 5:1 to 10:1."

On coax that allocation is physical. Upstream and downstream are different frequency bands on the same cable, separated by filters inside every amplifier. CableLabs describes the North American options in its explainer on band splits: in a sub-split, 5 to 42 MHz carries upstream traffic while 54 MHz through 1.2 or 1.8 GHz carries downstream. Mid-split moves the upstream edge to 85 MHz. High-split takes it to 204 MHz. Sub-splits are still the common case in North America, and moving past one means the operator swaps amplifiers and legacy gear across the plant.

A cable upstream ceiling, then, is a capital decision somebody made about your neighborhood, and your account is downstream of it in every sense. Ask which split serves your node before you ask about faster tiers. If the answer is sub-split, every tier they can sell you is drawing its upload from the same 37 MHz.

You can often settle the question without asking anyone. A cable modem's own diagnostic page, reachable at 192.168.100.1 on most units, lists the frequency each upstream channel is locked to. Channels sitting at 22 or 35 MHz are inside the 5 to 42 MHz band. Anything above 42 MHz means the plant in front of you has already been moved up. That reading tells you which ceiling applies before a sales conversation starts, and it does not depend on what a retention agent is willing to say.

Nine codes decide what the map calls your line

When a provider reports availability at your address, it also reports how the connection reaches the premises. The 2026 Section 706 Report notes that the Broadband Data Collection offers nine mutually exclusive technology codes for fixed service. The BDC help center table (updated 28 February 2025, read 28 August 2026) defines them:

Three of those definitions do real work.

Code 50 excludes fiber to the curb, which means a network running fiber to a cabinet and copper for the last stretch files as code 10 whatever the plan is named. Code 10 then lumps that in with plain DSL and with ethernet over copper, so the entry cannot tell you which of the three is at your wall. And code 71 puts a national carrier's 5G home product in the same bucket as a regional wireless ISP on licensed spectrum, two businesses whose capacity behaves nothing alike.

The distances behind the wireline codes are in the rule itself. Under 47 CFR 1.7004(c)(1)(iii), in the eCFR issue of 1 August 2026 read on 28 August 2026, DSL coverage at or above 25/3 Mbps may not be reported beyond 6,600 route feet from the DSLAM, hybrid fiber coax beyond 12,000 route feet from the aggregation point, and fiber to the premises beyond 196,000 route feet from the optical line termination point. The rule bolts carve-outs onto each of those figures, for a current or previous subscriber beyond the bound and for a build obligation already underway, which is one of several reasons a polygon edge is not a wire. They are reporting limits rather than promises about performance, and the spread is still the whole story of copper: one technology may not be claimed past a few thousand feet, the other is allowed out to 37 miles. If your DSL is slower than a neighbor's on the same plan and the same price, the difference is most likely measured in feet of cable.

Latency is where the technologies stop substituting for each other

Throughput ranges overlap between these media. Latency ranges do not.

The Thirteenth report measured idle latency by technology: 8 to 14 ms for fiber, 13 to 22 ms for cable, 20 to 61 ms for DSL. It also published latency under load for the first time and found it runs significantly higher than idle across every technology, "with a more pronounced difference for DSL subscribers." That second measurement is the one that matches the lived experience of a call collapsing the moment somebody in the house starts an upload.

Satellite sits in another regime. The Eighth report, on September 2017 data, put median satellite latency at 594 to 612 ms against 12 to 37 ms for terrestrial services, and explained the cause: packets travel approximately 44,500 miles from an earth station to the satellite and back. Geostationary orbit produces that number and no amount of provisioning removes it, which is why the filing rules separate satellite into codes 60 and 61 at all.

Here is the part that matters if you have been treating the map as a shopping tool. It publishes none of these figures. Section 1.7004(c)(1)(ii) requires a fixed provider only to indicate whether round-trip latency for each reported speed combination is at or below 100 ms, at the 95th percentile of its measurements. One bit. Fiber at 9 ms and a wireless link scraping 95 ms on a calm afternoon are the same value in that field. The threshold is the Commission's own: the 2026 Section 706 Report (FCC 26-55, adopted 13 August 2026) confirms in a footnote that it treats low-latency connectivity in this context as less than or equal to 100 ms.

For the wireless technologies there is nothing federal of this kind to look up. The Eighth report is the last one that publishes satellite results: the Ninth mentions satellite once, to describe a test method, and the Thirteenth does not mention it at all. Its ten participating ISPs and twelve ISP-technology configurations are entirely wireline, and no fixed report has been published since it, on data now collected four years ago. For fixed wireless and satellite, the only per-technology numbers that describe your address are the ones you produce yourself, which is the argument for a method you can repeat rather than a screenshot of one good test.

Everything is shared; the question is where

All five of these technologies pool capacity somewhere, and knowing where tells you what an evening slowdown actually means.

On cable the pool is the upstream band on your node, divided among the homes in that service group. On fixed wireless it is the sector on the tower, and the carrier rations it with a qualification gate rather than a price, which is why a covered address can still be refused service. On satellite it is beam capacity over your region. Fiber shares as well, since a passive optical network splits one feed among a group of premises, but the pool is deep enough that it is rarely the thing you notice.

The 2026 Section 706 Report puts numbers on the distance between where a service is offered and where people take it. As of 30 June 2025 an estimated 99.7 percent of the U.S. population had access to a 100/20 Mbps satellite service, while satellite accounted for 2.3 percent of fixed broadband connections. Fixed wireless reached about 56 percent of the population and made up 8.5 percent of connections. On satellite the Commission adds a caveat drawn from the operator's own published specifications: reliability may decline during periods of congestion, and actual speeds depend on location, time of day and service plan. Availability and adoption measure different things, and the second one carries information the first one hides.

The label is where a provider commits to a number

Typical download speed, typical upload speed and typical latency are required fields on the broadband label. Those three are what your own logs get held against, and the second and third are the ones this article is about.

Get them out of the spreadsheet rather than the sales page. Every provider still has to publish its label contents in a file at one dedicated URL under 47 CFR 8.1(a)(3), and FCC 26-48, adopted 22 July 2026, deletes that paragraph — except that the Federal Register notice delays "instruction 3 (Sec. 8.1(a))" indefinitely, so the duty is live today, 28 August 2026, with a removal order already written against it. What that order changes about the fee lines is a longer story. For this purpose one step is enough: download the file, keep the row for your plan with the date you fetched it, and write the two numbers where you can see them next to your own.

Reading the three numbers against each other

Put them side by side: the technology your provider filed for your address, the typical upload and latency from your plan's label with the date you pulled it, and your own medians from a week of wired runs. The comparison sorts almost every complaint into one of three piles.

Measurements near the label, and a label near what the medium can do, means the line is behaving and the fault is on your side of the modem. That is a wiring or Wi-Fi question with a twenty-minute test attached to it. Measurements well under the label means you have a service complaint with a documented baseline, which is a different conversation from a general one about slowness. And a label already sitting at the ceiling of the technology means no upgrade on the same medium buys you what you are missing, whatever the tier is called.

Only the third case calls for a different line rather than a better plan. It is also the case no salesperson will name for you, because the technology code is the one item on the whole record that cannot be renegotiated over the phone.