The Upload Cap Is Not an Accident: How DOCSIS Baked Asymmetry Into American Broadband

Stefan Schueller’s piece on Swiss broadband focuses, correctly, on the political economy: Switzerland has 25 Gbps residential fiber because it treated infrastructure as a public utility and separated the ownership of physical cables from the business of selling internet. That argument is correct and worth making. But there is a more specific technical story inside it that the framing around free markets and regulatory capture tends to obscure.

The reason most American households get 500 Mbps down and 20 Mbps up is not primarily that ISPs chose to throttle uploads. It is that the infrastructure they built in the 1990s made symmetric service structurally difficult, and nothing has forced them to replace it.

DOCSIS and the Consumption Model

Roughly 65% of US residential broadband runs over DOCSIS cable infrastructure. DOCSIS was designed in the mid-1990s to turn existing cable television networks into two-way data pipes. The dominant application was downloading: web pages, eventually streaming video, software updates. The upstream path, the direction from your home to the internet, was an afterthought. You needed it to send TCP acknowledgments and HTTP requests, not to move data.

The original DOCSIS 1.0 specification reflected this. Coaxial cable carries a wide spectrum of frequencies, roughly 5 MHz to 1 GHz in an HFC (Hybrid Fiber Coaxial) network. The DOCSIS 1.0 split allocated the 5 to 42 MHz band for upstream traffic and the 50 to 750 MHz band for downstream. The upstream got less than 40 MHz of usable spectrum; downstream got 700 MHz. That ratio encoded a specific assumption about how the internet would be used, and that assumption has been wrong for at least fifteen years.

DOCSIS 3.0 and 3.1 improved throughput substantially but preserved the fundamental asymmetry. DOCSIS 3.1 expanded upstream to 5 to 204 MHz and downstream to 258 MHz to 1.218 GHz. The absolute numbers grew, but the ratio stayed roughly similar. A DOCSIS 3.1 node can deliver 1 to 2 Gbps aggregate downstream during off-peak hours; it delivers something closer to 100 to 200 Mbps upstream. These numbers are shared across 250 to 500 homes on the same coaxial node.

The result is what Ookla’s Global Speedtest Index shows: the US median fixed broadband download is around 200 Mbps; the median upload is around 27 Mbps. Switzerland’s median upload is approximately 180 Mbps. That gap is not a business decision layered on top of capable infrastructure. It is the infrastructure.

What DOCSIS 4.0 Actually Addresses

DOCSIS 4.0, finalized in 2020, includes two technical paths that could partially address the upstream problem. The first is Extended Spectrum DOCSIS, which widens the total spectrum to 1.8 GHz and increases upstream bandwidth accordingly. The second is Full Duplex DOCSIS (FDX), which allows the same frequencies to be used for both upstream and downstream simultaneously using active interference cancellation.

Full Duplex DOCSIS is technically ambitious. On a passive coaxial plant, simultaneous transmit and receive on the same spectrum requires measuring and subtracting the echo of your own transmitted signal from the received signal in real time. CableLabs spent years on the specification. Comcast has run limited trials. Deployment at scale requires replacing node equipment throughout an HFC plant.

In theory, FDX DOCSIS could eventually deliver symmetric gigabit service on cable. In practice, deployments as of 2025 are sparse. Charter and Comcast have started Extended Spectrum DOCSIS upgrades in select markets, primarily to expand downstream capacity. Upstream improvements are real but modest. The industry’s own roadmaps project full multi-gigabit symmetric DOCSIS service as a late-2020s outcome in leading markets.

Switzerland’s point-to-point active Ethernet networks upgraded to 25 Gbps by swapping transceivers. A customer plugs an SFP28 module into their switch. Init7 ships the transceiver and the connection runs at 25 Gbps symmetric. The fiber itself, single-mode G.652.D, has the physical capacity for much more. The bottleneck at that point is the customer’s storage subsystem, not the internet connection.

The comparison is not between two different speeds. It is between two different upgrade philosophies. One requires swapping a small optical module; the other requires replacing coaxial distribution plant, interference cancellation hardware, and customer premises equipment across millions of homes.

The Asymmetry as a Developer Tax

For most users, the download-heavy model is tolerable. Streaming video is downstream. Web browsing is downstream. Even most gaming is downstream.

For anyone running infrastructure from home, the math changes completely. A home server hosting anything for external access is bottlenecked by upload. A developer pushing to a self-hosted Gitea instance, running a Minecraft server, operating a Discord bot that calls back to a local webhook endpoint, or self-hosting a Nextcloud installation for remote file access is working against the upload cap constantly.

The asymmetry also affects work-at-home realities that have been common since 2020. A video call on Zoom or Google Meet requires roughly 3 Mbps symmetric. Sharing a screen at 1080p requires more. A household with three people doing simultaneous video calls can saturate a 20 Mbps upstream connection on a mid-tier cable plan. Swiss households on symmetric fiber do not have this calculation to make.

The structural reason for the gap is not that Swiss ISPs are more generous. It is that the physical medium they sell service over has no asymmetric constraint built in. A point-to-point fiber strand carries the same number of photons in both directions. The speed in each direction is determined by the transceivers, not by the medium’s inherent capacity allocation.

Density Was Never the Excuse

The standard defense of American broadband underperformance appeals to population density. Switzerland is small and wealthy; American rural areas are vast and unprofitable. This explains some of the rural access gap but nothing about the urban one.

The US cities with the fastest broadband tend to be cities where a municipal utility built the infrastructure, not where the market delivered it. Chattanooga, Tennessee has EPB, a city-owned electric utility that launched symmetric gigabit service in 2010 and now offers 10 Gbps tiers. Longmont, Colorado has NextLight, the city’s municipal fiber network, offering 1 Gbps symmetric for around $70 per month. These are not dense urban tech corridors.

The density argument also fails when you look at what cable companies have managed to do in dense US cities. Manhattan, San Francisco, and Chicago have had cable infrastructure for decades. Comcast and Charter serve dense apartment buildings with thousands of subscribers per block. The upload speeds in those buildings are not materially better than in suburban Ohio. Density benefits the cable company’s revenue and does not translate into improved service, because there is no competitive pressure to translate it.

In the Netherlands, a country with population density roughly three times Switzerland’s, most urban residents can choose between multiple ISPs all operating over structurally separated open-access fiber infrastructure. The separation means the infrastructure owner profits from having more ISPs on their network, not from limiting them.

The BEAD Program and What It Will Not Fix

The 2021 Infrastructure Investment and Jobs Act allocated $65 billion for broadband, with the bulk flowing through the BEAD program administered by NTIA. This is a significant number. The implementation has significant problems.

Most BEAD funding targets unserved rural areas, which is the right priority, but the program does not require open-access conditions on the infrastructure it funds. A private ISP receiving federal money to build fiber in rural Kentucky can build a closed network and lock competitors out. This replicates the ownership structure that produced the current problem, just in new geography. Some states have tried to attach open-access requirements to their BEAD plans; most have not.

The program also relies on the FCC’s coverage maps to define who is underserved. Those maps were built from ISP self-reporting and have been widely documented as inaccurate. Counties where a cable company reported service availability but never actually connected most homes are classified as served and are therefore ineligible for BEAD funding. The FCC has a map challenge process; it is cumbersome and has not resolved the underlying data quality problem.

The countries that built the infrastructure that makes 25 Gbps symmetric service commercially viable spent their infrastructure investment in the early 2000s, when construction costs were lower and the investment had decades to pay off. Switzerland, South Korea, Japan, and Sweden are collecting the returns now. The US is in 2025 still arguing about whether 100 Mbps qualifies as a broadband baseline.

What Changes the Equation

The technical path to symmetric gigabit service in the US exists. Fiber to the premises, deployed on an open-access wholesale model, would provide it. Some cable operators are deploying fiber overbuilds in their own service areas, threading new fiber alongside their existing coaxial plant. Comcast’s XGS-PON deployments and Charter’s fiber overbuild program represent real movement. The speeds will be better than DOCSIS. Whether those networks will ever be required to provide open access to competing ISPs is a different and currently unresolved question.

The municipal utility model, which produced EPB Chattanooga and NextLight Longmont, demonstrates that symmetric fast fiber is deployable in the US. The policy obstacle is not technical. As MuniNetworks documents, 17 to 19 states have passed laws restricting or prohibiting municipal ISPs, typically drafted with assistance from incumbent carriers. The Sixth Circuit ruled in Tennessee v. FCC (2016) that the FCC lacks authority to preempt those state laws.

The DOCSIS asymmetry was not the only policy path available in the 1990s. It was the path that incumbents chose because it was cheaper to build than symmetric fiber. Once that infrastructure was in the ground, the incumbents had a strong interest in defending the investment rather than obsoleting it. That interest has been more consistently represented in American regulatory proceedings than the interest of developers, home labbers, and anyone else who needs to push data upstream at reasonable speed.

The 25 Gbps gap is not primarily a technological problem. The technology is not difficult. It is a record of thirty years of infrastructure decisions made by entities whose incentives were not aligned with the people who would eventually need symmetric gigabit service.