Daniel's been thinking about something that's been bugging me too, actually. He points out that we've talked about how enterprise fiber is sold as bandwidth with guarantees, while consumers get sold speed with a lot of fine print. His question is: if we know fiber optic glass can carry terabits, what actually creates the speed limits we see on our bills? Is it all just artificial tiering so ISPs can upsell us? If they can guarantee bandwidth to enterprises, why not do the same for everyone? And why does this speed-tier pattern show up the same way across coax, DSL, and fiber, as if the delivery technology doesn't even matter?
The pattern is the tell. You see a hundred megabit, three hundred, five hundred, one gig tiers on fiber. Same tiers on cable. Same tiers on DSL back when that was a thing. Different physics, same menu. That's not a coincidence, and it's not purely a technical constraint either.
So what's actually being sold when I pay for five hundred megabits?
Access to a shared pool with a software-enforced peak rate. That's it. The thing on your bill that says "speed" is really a cap — a ceiling on how fast you can pull from a pipe that dozens or hundreds of your neighbors are also pulling from. And the ISP has done a bunch of math to make sure that if everyone pulls at once, the whole thing doesn't fall over.
Which is not how I think most people imagine it. I think most people imagine a dedicated pipe from their house to the internet.
Right. And that dedicated pipe does exist — it's what enterprises buy. It's called Dedicated Internet Access, DIA, and it's a fundamentally different product. Point-to-point fiber from the customer straight to the ISP's router. No splitting, no sharing, no "up to." You buy a gig, you get a gig, and if you don't, the ISP pays you money.
So the same two words — internet connection — describe two completely different things depending on whether you're a person or a corporation.
And ISPs have no incentive to clarify that. The confusion between speed and bandwidth is basically their best marketing tool. Speed is a rate — bits per second, how fast data moves. Bandwidth is capacity — the maximum rate the medium can support. ISPs have spent decades teaching consumers to think in terms of speed tiers, because "buy more speed" is an easy upsell. "Buy a higher cap on your shared pool" doesn't sound as good.
It's like if a gym sold you a membership by saying "you can run at twelve miles per hour," but what they actually mean is the treadmill is capable of twelve miles per hour — you just share it with forty other people and you each get a thirty-second slot.
That's uncomfortably close. And the gym would put "up to twelve miles per hour" in the contract and call it a day.
Let's start with the physical reality, because Daniel's question gets at something important. If the glass can handle terabits, where do the limits actually come from?
So here's the thing about fiber optics. A single strand of single-mode fiber, with the right equipment on both ends, can carry multiple terabits per second. You use something called DWDM — dense wavelength division multiplexing — which basically shoots different colors of laser light down the same glass, each carrying its own data stream. The glass itself is almost never the bottleneck at consumer scale, or even at most enterprise scales. The limitation is everything you attach to the glass.
So the cable coming into my house is not the problem.
The cable is the last thing that's the problem. The limits come from three places. One, the ONT — the optical network terminal, the box on the side of your house where the fiber terminates. That thing has firmware that says "this customer gets five hundred megabits." Two, the PON split ratio. Most consumer fiber is GPON — gigabit passive optical network — and it works by splitting one fiber from the ISP among thirty-two or sixty-four subscribers. The downstream capacity on GPON is two and a half gigabits per second, shared among everyone on that splitter.
Wait. So my "dedicated fiber connection" is shared with up to sixty-three of my neighbors?
Yes. And that's the part most people don't know. Consumer fiber is shared infrastructure, just like cable. The fiber runs from your house to a passive optical splitter in a cabinet somewhere in the neighborhood, and from there a single fiber runs back to the ISP's central office. Everyone on that splitter is contending for the same two and a half gigs.
That seems like a fairly important detail to leave out of the brochure.
The third place limits come from is the OLT — the optical line terminal at the ISP's end — and specifically its upstream port connecting to the ISP's core network. This is where oversubscription is engineered. The ISP might have a ten-gigabit uplink serving two hundred to five hundred subscribers. If everyone tried to use their full provisioned speed simultaneously, that uplink would be overwhelmed instantly.
So the speed tier on my bill is a software cap that prevents me from being the person who ruins it for everyone else.
That's exactly what it is. And that's not purely artificial — it's a statistical multiplexing strategy. The ISP is betting that most users don't use their full speed most of the time. They dimension the network for average usage, not peak usage. The speed tier is how they enforce that bet.
What's a typical oversubscription ratio?
Consumer ISPs typically oversubscribe at ratios between twenty to one and fifty to one on the aggregation network. So a ten-gigabit uplink might serve two hundred to five hundred subscribers who are each sold a one-gig connection. The math only works because usage is bursty — someone streams a video, downloads a file, then goes idle. Multiply that across thousands of subscribers and the peaks smooth out.
And if they don't smooth out?
Then everyone's speed drops and you get angry phone calls. Which is why ISPs monitor those ratios constantly. If a particular PON splitter or neighborhood node starts saturating regularly, they'll either add capacity or... they'll add more tiers to push heavy users into higher-priced plans where the economics justify the bandwidth.
So tiering is partly a traffic management tool and partly a pricing tool.
Right. Now contrast this with enterprise Dedicated Internet Access. DIA doesn't use PON at all. It's active Ethernet — point-to-point fiber from the customer directly to the ISP's router. No splitter, no sharing, no statistical multiplexing. The bandwidth is physically dedicated. The Committed Information Rate — the CIR — is contractually guaranteed, and the contract includes liquidated damages if the ISP fails to deliver.
Which is why they can guarantee it. They've over-provisioned the network to make sure they never miss.
And they charge accordingly. A DIA gigabit connection might cost three hundred to a thousand dollars a month depending on location, while a consumer GPON gigabit connection costs sixty to ninety dollars. Same speed number on paper, completely different product underneath.
So Daniel's question about why ISPs can't just give consumers what enterprises get — the answer is they could. It would just cost three hundred dollars a month.
And most consumers wouldn't pay that. The shared model exists because it's what the market will bear at consumer price points. But — and this is where it gets interesting — the degree of tiering is absolutely a business choice, not a technical necessity.
This is the part I want to dig into. The gap between what the infrastructure can deliver and what's actually sold.
Let's take a concrete example. Openreach in the UK — they run FTTP, fiber to the premisis, using GPON. Same physical fiber, same ONT, same splitter for everyone on that segment. But they sell it in four tiers: a hundred megabits, three hundred, five hundred, and one gig. All four tiers run over identical hardware. The only difference is a number in a config file.
So the fiber can do a gig. The ONT can do a gig. The splitter can do two and a half gigs shared. But they'll sell you a hundred megs and call it a different product.
And charge you less for it. Which is the other side of this — tiering isn't only about extracting more money from power users. It's also about offering lower price points to people who don't need a gig. If the ISP had to provision every customer for a gig, the baseline cost would be higher for everyone.
That's the price discrimination argument. Let people self-sort into tiers based on what they're willing to pay.
And it's economically efficient in that sense. The ISP recovers fixed network costs from the customers who value the service most. But the counterargment is that the cost difference between provisioning a hundred megs and a gig on the same GPON infrastructure is essentially zero. The fiber's already there, the ONT's already there, the splitter's already there. The marginal cost of enabling the higher tier is... nothing.
So why not just give everyone a gig and charge a flat rate?
Because then you lose the revenue from the customers who would have paid for the gig. And you also lose the customers who would have paid for the hundred-meg tier because now the flat rate is higher than what they were paying. Tiering captures both ends of the market.
So it's partly technical necessity — the shared infrastructure imposes real constraints — and partly artificial in the sense that the constraints are configured in software to serve a pricing strategy.
That's the synthesis. And it's not binary — it's a spectrum. Some tiering is about keeping the network stable. Some is about market segmentation. And some is about the fact that ISPs have spent decades training consumers to think about internet in terms of speed numbers, making it very hard to ask for what enterprises get — a guaranteed minimum.
Can we pause on that point about the marginal cost being zero? Because that feels like it should be more scandalous than it is. If I'm Openreach and I've already run the fiber, already installed the ONT, already lit the splitter — what actual cost do I incur when a customer upgrades from a hundred megs to a gig?
The honest answer is: a configuration change. Someone — or more likely some automated provisioning system — changes a rate limiter value in a database. That's it. No truck roll. No hardware swap. No additional power draw to speak of. The electrons don't care what number is in the config file.
So the marginal cost is literally zero.
For that individual upgrade, yes. But the ISP would argue that's not the right way to look at it. They'd say the cost is in the network capacity that the higher tier might consume. If a thousand customers all upgrade from a hundred megs to a gig and actually start using it, the aggregation network might need an upgrade. So the tier pricing is a way of reserving capacity — you pay more, and that money funds the network expansion if enough people actually use what they're paying for.
That's a reasonable argument if the money actually goes to network expansion.
And there's the rub. Whether it does or doesn't is opaque to the consumer. Some ISPs reinvest. Some don't. The tier pricing structure doesn't distinguish between the two.
Daniel also asked why this pattern holds across coax, DSL, and fiber. If the delivery technology changes, why does the tiering stay the same?
Because all three are shared-medium technologies at some level. Coax — DOCSIS — shares the node. DSL shares the DSLAM uplink. PON shares the splitter. The underlying physics are different, but the economic model is identical: you have a shared resource with a finite capacity, and you need to prevent any single user from consuming it all. The mechanism is always a software rate limiter on the customer premisis equipment. The technology changes, the config file doesn't.
The tiering is a function of the business model, not the physical layer.
Yes. And that's why you see the same menu of speeds whether you're on coax or fiber. The fiber could technically deliver more — much more — but the ISP has no incentive to offer it if the coax competition is also selling five hundred megs. The tier is set by the market, not the glass.
Comcast has this fascinating split where they sell two different one-gig products. Gigabit Extra is shared GPON, eighty-nine dollars a month. Gigabit Pro is dedicated fiber with an SLA, two hundred ninety-nine dollars a month. Same speed label, completely different infrastructure.
Most consumers have no idea there's a difference. They see "gigabit" on both and pick the cheaper one. The confusion between speed and bandwidth isn't a bug — it's a feature of how these products are marketed.
Let me ask you something. If GPON can deliver two and a half gigs downstream to the splitter, and NG-PON2 can do forty gigs shared, at what point does the shared capacity become so large that tiering stops mattering?
That's the interesting question. NG-PON2 offers forty gigs downstream shared among thirty-two or sixty-four subscribers. Even at sixty-four-way split, that's over six hundred megs per subscriber if everyone's maxing out simultaneously. At some point the shared pipe is fat enough that congestion becomes rare enough that tiering is harder to justify.
Yet I suspect we'll still see tiers.
Oh, absolutely. Because by then ISPs will have invented new ways to segment the market. Latency tiers, QoS tiers, "premium routing" tiers. The constraint will shift from raw throughput to something else, but the tiering structure will persist because it's too profitable to abandon.
The answer to Daniel's question — are speed tiers artificial constraints created by ISPs to upsell? — is yes and no.
Yes, in the sense that the caps are software-configured and the same hardware could deliver more. No, in the sense that the shared infrastructure model requires some form of usage management to function. The tiering is a real response to a real constraint — but the constraint itself is partly manufactured by the choice to build shared rather than dedicated networks.
That choice was made because dedicated networks would cost more to build, and consumers have shown they won't pay dedicated prices.
Right. The whole system is a negotiation between what's technically possible and what's economically viable. The speed tier on your bill is the point where those two curves cross.
I keep coming back to the language. ISPs sell consumers "speed" and enterprises "bandwidth." They've inverted the actual technical meanings of those words.
Deliberately. Speed is intuitive — faster is better, everyone understands faster. Bandwidth is abstract and sounds like engineering. If consumers started asking for "a committed information rate of five hundred megabits with an SLA," the whole pricing model would unravel.
Because then they'd have to actually deliver it.
Or pay penalties when they don't. Consumer contracts explicitly exclude guarantees. "Up to" does an enormous amount of legal work in two words.
There's something almost elegant about it. The entire consumer internet market is built on a deliberate confusion between two concepts, and everyone just... accepts it.
Because for most people, most of the time, it works well enough. Netflix streams, web pages load, Zoom calls connect. The statistical multiplexing does its job. The problem only becomes visible when the network gets congested — which is exactly when the "up to" fine print activates.
The ISP says "you're getting up to five hundred megs, and right now up to means forty."
They're technically not lying. That's the beauty of it.
Hilbert.
Hilbert: I spent part of two thousand three configuring rate limits on DSLAMs for a regional ISP in Nebraska. We called it PrairieNet. I had a single OC-3 uplink — a hundred fifty-five megabits — for a town of twelve thousand people.
One uplink.
Hilbert: One. My boss told me: set those caps tight or the whole thing collapses at seven PM when everyone gets home from work. We sold three tiers. One and a half megabits, three megabits, six megabits. The six-meg customers paid triple. And ninety-five percent of the time, they got exactly the same service as the one-point-five customers, because the bottleneck was never the last mile. It was that OC-3.
The tiering was basically honest in the sense that you had to cap people to keep the network alive.
Hilbert: It was honest in the sense that I typed real numbers into a real config file to prevent a real outage. But.
But.
Hilbert: The reason we only had one OC-3 was that the company didn't want to pay for a second one. A second OC-3 would have cost about four thousand dollars a month at the time. We had the budget. The CEO drove a Mercedes. So the necessity was manufactured. The caps were real and necessary given the infrastructure we had. The infrastructure was what it was because someone decided not to spend the money.
That's the "manufactured necessity" problem in a nutshell.
Hilbert: It's turtles all the way down. Every constraint is real at the level you're operating at. But the level you're operating at was chosen.
Did anyone ever call and complain that they were paying for six megs and getting two?
Hilbert: Constantly. I had a script. "Sir, your connection is capable of six megabits to our central office. The speed you experience to the broader internet depends on network conditions." I said "network conditions" so many times I started hearing it in my sleep.
Same script they use today, just with bigger numbers.
Hilbert: The numbers change. The script doesn't.
When you hear us talking about artificial versus necessary, where do you land?
Hilbert: I land on: the distinction doesn't matter to the person paying the bill. Whether the cap exists because of physics or because of a spreadsheet, the experience is the same. You pay for a number you don't get. The question that matters is whether there's enough competition that the ISP has to care.
In most of the US, there isn't.
Hilbert: In most of Nebraska in two thousand three, there was us. That was the option. We knew it. The caps reflected it.
The tiering was a function of market power as much as technical capacity.
Hilbert: Always is. I've still got the config templates on a floppy disk somewhere.
Of course you do.
The misconception that drives me craziest is that fiber internet is inherently dedicated. People switch from cable to fiber and think they've escaped the shared pipe. They haven't. GPON is shared. Most consumer fiber is GPON.
The related one — that speed and bandwith are the same thing. They're not. Speed is a rate, bandwith is a capacity. ISPs have spent twenty years teaching everyone to use the wrong word for the thing they're selling, and it's worked so well that even tech journalists get it wrong.
Where does this leave us? Is the tiering model permanent?
I think it's already changing at the edges. You look at municipal fiber networks, Sonic in California, Google Fiber in a few cities — they offer symmetrical gigabit at flat rates with no caps, no tiers, no "up to" games. When there's actual competition, the model shifts from tiering to "here's the pipe, use it."
Because the constraint wasn't technical in the first place.
Not at the level of the fiber. The constraint was always in the business model. Daniel's question gets at something fundamental. The internet is a physical system with real limits, but those limits are shaped by business decisions as much as by physics. The speed tier on your bill is a negotiation between what's possible and what's profitable.
Most of us are negotiating from a position of... not much leverage.
Unless you live in one of the few places with real competition. Then magically, the caps disappear and the speeds go up and the price stays the same. Funny how that works.
Almost like the fiber could do it the whole time.
Almost like that.
This has been My Weird Prompts. Thanks to our producer Hilbert Flumingtop for keeping the levels right and the config files tight. Find us at my weird prompts dot com, or email the show at show at my weird prompts dot com. We'll be back soon.
