Here's what Daniel wants to know. He's been looking at how internet is sold above the consumer tier, and he noticed something. At the consumer level, you buy a speed — one gig, two gig, whatever the package says. But above that, the product stops being a number on a web form and becomes a contract. He's asking how fibre bandwidth is actually packaged and priced at that level right now, and what a typical service-level agreement on one of those circuits genuinely guarantees. Not the marketing, not the sales pitch — the real commitments in the fine print.
This is one of those topics where the thing you think you're buying and the thing you're actually buying are almost completely different. The contract is the product. The bandwidth is almost a side effect.
Say more.
When you buy consumer fibre, you're buying access to a shared pipe. The provider runs a GPON network — gigabit passive optical network — where one fibre from the exchange gets split passively to serve thirty-two or sixty-four premises. They sell you "one gig" but that's the maximum the port can do, not a guarantee. At three in the morning when nobody's online, you might get it. At eight in the evening when the whole street is streaming, you won't.
And above that tier, you stop sharing.
That's the shift. The product category is called Dedicated Internet Access, DIA. It's a circuit where the bandwidth is yours alone — symmetric, so a hundred meg up and a hundred meg down, all the way to ten gig or more. The fibre from your building runs back to the provider's point of presence and you get a committed information rate — a CIR — that's guaranteed.
So the hundred meg is a floor, not a ceiling.
And that's the first thing people get wrong when they see the price. A hundred-meg DIA circuit can cost more than a one-gig consumer line, and they think they're being ripped off. They're not. They're buying a hundred meg that is always a hundred meg, with an SLA wrapped around it.
Let's get into the packaging then. What does a DIA quote actually look like?
It's built around three things. The committed information rate — the guaranteed minimum bandwidth. The port speed — the physical capacity of the interface, usually higher than the CIR so you can burst above it. And the access tail — the local loop, the fibre from the exchange to your building. The CIR and port speed are priced per meg, and the access tail is priced based on distance and whether fibre already passes the building.
Distance. So a circuit across town costs less than one across the country.
Considerably less. If the provider already has fibre in the ground near you, the build cost is low. If they have to trench new fibre, the quote jumps. I've seen quotes where the monthly recurring charge for the bandwidth was maybe eight hundred dollars, but the access tail added another two thousand a month because the building was three miles from the nearest splice point.
And this is all on a term contract.
One to three years, typically. Early termination fees if you cancel. The provider is amortising the build cost over the term, so they need that commitment. But the contract isn't really about the bandwidth. It's about the SLA. The service-level agreement is what you're paying for. The bandwidth is the delivery mechanism for the SLA.
So walk me through what a typical SLA actually says.
The headline number is always uptime. Ninety-nine point nine percent, ninety-nine point nine nine, sometimes five nines — ninety-nine point nine nine nine percent. And those numbers sound impressive until you do the arithmetic.
Do the arithmetic.
Ninety-nine point nine percent uptime allows for forty-three minutes and forty-nine seconds of downtime per month. Ninety-nine point nine nine percent brings it down to about fifty-two minutes per year. Five nines is five minutes and fifteen seconds per year.
So the difference between three nines and four nines is the difference between "we can be down for most of a lunch break every month" and "we can be down for less than an hour a year."
And the price difference between those two tiers can be enormous. Double, triple. You're paying for the engineering redundancy that makes the higher number possible — diverse fibre paths, backup power, multiple upstream peers.
But the percentage is measured how?
It's measured over a billing period — usually a month or a quarter. Not per incident. So if your circuit goes down for forty minutes on a Tuesday, that's within the ninety-nine point nine percent allowance for the month. The provider hasn't breached anything. The outage gets absorbed by the averaging.
And if it goes down for forty minutes every Tuesday?
Still might not breach, if the other weeks are clean. The monthly total stays under forty-three minutes. You can have a circuit that fails reliably every Friday afternoon for a year and the provider's SLA performance looks perfect on paper.
That seems like a pretty significant gap between what the number implies and what it actually delivers.
It's the gap the whole product is built on. And it gets wider when you look at repair time. The SLA will specify a mean time to repair — MTTR — typically four to eight hours for a business DIA circuit. But the clock starts when you report the fault, not when the circuit goes down.
So if it fails at two in the morning and nobody notices until eight, the clock starts at eight.
And if your monitoring didn't catch it, or the provider's own monitoring didn't flag it, those six hours don't count. The SLA only covers the period between the trouble ticket opening and the provider saying it's fixed. And "fixed" can mean a temporary workaround — they reroute you over a backup path that's slower, or they patch a failing piece of equipment that's going to fail again next week. The SLA doesn't require a permanent fix, just a return to service.
What does the customer actually get if the SLA is breached?
A credit on their bill. And this is where it gets almost insulting. The credit is typically a percentage of the monthly recurring charge for the affected circuit — ten percent is common. Sometimes it scales — ten percent for the first hour beyond the SLA, another ten for each additional hour, capped at fifty percent of the monthly fee.
So if a business is losing thousands of dollars a minute because their circuit is down, the compensation is a fraction of the monthly bill.
And it's not automatic. You have to notice the breach, document it, submit the claim, and argue with the provider if they dispute the timing. Most businesses don't bother for a ten percent credit on a twelve-hundred-dollar circuit. A hundred and twenty dollars isn't worth the time it takes to file the paperwork.
The SLA is designed to be mostly unclaimed.
It's a risk transfer mechanism dressed as a guarantee. The provider caps their liability at a fraction of the service cost, and the customer carries the actual business risk. The SLA doesn't insure you against losses — it just gives you a small refund if the service was bad enough to cross a threshold that's hard to cross.
What about the things the SLA doesn't cover at all?
Latency and jitter are the big ones. Most standard DIA SLAs don't guarantee them, or they guarantee them so loosely the numbers are meaningless. They might say "latency under fifty milliseconds within the provider's network" — but that excludes the internet beyond their border, which is where most latency problems live. Jitter — the variation in packet arrival time — is critical for voice and video, and it's almost never in the SLA.
And the last mile?
Often excluded. The SLA might cover the provider's core network and their point of presence, but the local loop is frequently carved out. They'll call it "access tail" or "local distribution" and note that it's subject to separate maintenance terms. Which is absurd, because the local loop is where most physical failures happen — a backhoe in the street, a flooded vault, a damaged splice. That's the part most likely to break, and the SLA often doesn't cover it.
So you're buying a guarantee on the part that's least likely to fail, and the part that actually breaks is excluded.
That's the shape of it. There's also force majeure — the clause that voids the SLA during events beyond the provider's control. Storms, floods, power grid failures, fibre cuts by third parties. And maintenance windows — scheduled downtime doesn't count against the SLA, and the provider gets to define what's "scheduled." Some contracts require forty-eight hours notice. Others just say "reasonable notice."
It always does. And then there's the "reasonable efforts" language around repair. The provider commits to using reasonable efforts to restore service within the MTTR window. If they can't, because a part is out of stock or a technician is unavailable, that's not a breach — they made reasonable efforts. The SLA is full of escape hatches.
Let's talk about what else comes with a DIA circuit beyond the SLA. Daniel mentioned static IP allocations and BGP.
Right. A consumer connection gives you a dynamic IP address that changes periodically, and you're stuck behind carrier-grade NAT in a lot of cases. A DIA circuit comes with a block of static public IP addresses — typically a slash twenty-nine or larger, so eight or more usable addresses. You can run your own servers, your own mail infrastructure, your own VPN endpoints. And if you're a larger organisation, you can bring your own autonomous system number and run BGP — Border Gateway Protocol — with the provider.
Which means you're advertising your own IP space and managing your own routing.
You're effectively a small ISP at that point. The provider gives you a BGP session and you control how your traffic enters and leaves their network. That's a completely different product from anything in the consumer space. It comes with technical requirements — you need staff who understand BGP, you need your own IP allocations from a regional internet registry. But for a business that needs multi-homed connectivity or wants to switch providers without renumbering their entire network, it's essential.
How do you tell the difference between a real DIA circuit and a consumer product with a business label?
The quote tells you. A real DIA quote will specify the CIR, the port speed, the access tail cost, the SLA terms including MTTR and the credit schedule, the IP allocation, and whether BGP is available. A business broadband product — which is what a lot of small business "fibre" actually is — will quote a speed tier with an "up to" qualifier, no CIR, no MTTR commitment, and an SLA that's basically "we'll try to fix it during business hours."
"Up to" is the tell.
"Up to" means you're on a shared GPON or similar architecture. The provider has sold that same capacity to dozens of other businesses in your area. At peak times, you get whatever is left. The price is lower — maybe a hundred and fifty dollars a month instead of twelve hundred — but the product is fundamentally different. It's consumer infrastructure with a business support phone number. The way to spot it is to ask two questions. One, is the bandwidth symmetric and guaranteed? Two, what's the MTTR commitment? If they can't answer both with specific numbers, it's not DIA.
Let's talk about how this market has shifted. You mentioned prices dropping.
The overbuild has been significant. In a lot of metro areas, there are now three or four providers with fibre in the ground — the incumbent telco, a cable operator, and one or two competitive providers who built out during the low-interest-rate years. That competition has pushed DIA prices down. A one-gig DIA circuit that might have cost three thousand dollars a month ten years ago can be had for under a thousand in some markets now.
But the SLAs have standardised.
That's the tradeoff. As prices dropped, the SLAs converged. Most providers now offer essentially the same terms — ninety-nine point nine nine percent uptime, four-hour MTTR, ten percent credit per incident. The differentiation isn't in the SLA anymore. It's in the extras. Proactive monitoring, where the provider detects and responds to issues before you report them. Twenty-four seven support with a named account manager. Online portals where you can see circuit performance in real time.
And the rise of SD-WAN has changed what the SLA is worth.
This is the knock-on effect that's really reshaping the market. SD-WAN — software-defined wide area networking — lets you bond multiple circuits together and route traffic dynamically based on performance. A typical setup might be a DIA circuit as the primary, a cheap broadband line as secondary, and an LTE or five-g failover as tertiary. The SD-WAN box monitors latency, jitter, and packet loss on each link in real time, and if the DIA circuit degrades, it shifts traffic to the backup automatically.
So the SLA on the DIA circuit becomes less critical because the circuit isn't a single point of failure anymore.
If your SD-WAN setup can fail over in under a second, you can tolerate a DIA circuit that goes down occasionally. You might even choose a cheaper circuit with a weaker SLA because the redundancy handles it. The SLA shifts from being insurance against downtime to being a negotiating point on price.
Which means the provider's SLA is competing with the customer's own redundancy architecture.
And the customer's redundancy is often cheaper than the premium for a higher-tier SLA. Why pay double for five nines when you can get two three-nines circuits from different providers for less money, and have actual diversity? The SD-WAN box doesn't care about the SLA — it cares about whether packets are getting through.
That seems like it would push providers toward application-level guarantees eventually.
That's where it's heading. The "as-a-service" model. Instead of selling you a circuit with an SLA, they sell you a connectivity outcome — "your voice traffic will have less than twenty milliseconds of latency and zero point one percent packet loss." The underlying infrastructure could be anything — fibre, fixed wireless, satellite. The provider manages the mix and you just buy the result.
Which is a whole new set of fine print to worry about.
It is. Now the measurement becomes the product. How do they measure latency? From where to where? Over what time period? What's the sampling interval? Every one of those parameters can be tuned to make the numbers look better than they are. It's the same game in a different wrapper.
So the core dynamic doesn't change. The contract is still the product.
It always is. The technology changes, the packaging changes, but the fundamental tension is the same. The buyer wants a guarantee that the service will work. The seller wants to limit their liability when it doesn't. The SLA is the document that negotiates that tension, and it's always going to be written by the seller's lawyers.
Let's put some concrete numbers around this. If I'm a business buying a one-gig DIA circuit with a ninety-nine point nine nine percent SLA and a four-hour MTTR, what am I actually holding?
You're holding a circuit that delivers a symmetric gigabit with a committed information rate of probably eight hundred meg, maybe the full gig if you paid for it. You've got a block of static IPs, probably eight or sixteen. You can run BGP if you want. Your SLA allows for about fifty-two minutes of downtime per year, measured monthly. If the circuit goes down, the provider commits to having someone working on it within four hours of your call. If they breach the uptime target, you can claim a credit — probably ten percent of the monthly fee, maybe scaling to fifty percent if it's really bad.
And the monthly fee is what?
In a competitive metro market in twenty twenty-six, probably eight hundred to fifteen hundred dollars a month on a one-year term, depending on the access tail. Cheaper if you commit to three years. More if you're in a building that needs new fibre pulled.
And the business impact of an outage?
Depends entirely on the business. For an e-commerce operation doing fifty thousand dollars an hour in revenue, an eight-hour outage is a four-hundred-thousand-dollar problem. The SLA credit on a twelve-hundred-dollar circuit is a hundred and twenty dollars. The asymmetry is stark.
The SLA isn't insurance. It's a rebate coupon.
That's exactly what it is. And it's a rebate coupon you have to remember to clip and mail in.
What about the things that aren't in the SLA at all? Packet loss, for instance.
Packet loss is a funny one. Some SLAs mention it, but the threshold is usually set high enough to be meaningless — one percent, sometimes higher. For comparison, a VoIP call starts to degrade at about one percent packet loss. Video conferencing gets choppy at half a percent. So an SLA that guarantees "less than one percent packet loss" is guaranteeing that your voice calls might work. That's not a guarantee.
And jitter?
Rarely specified. When it is, it's often something like "less than ten milliseconds of jitter within the provider's network." But jitter compounds — every router hop adds variability. The provider's network might be clean, but the path to your cloud provider or your remote office crosses multiple networks, and none of those are covered.
So the SLA covers the part of the path the provider controls, which is the part least likely to be the problem.
And that's the structural limitation of any SLA. The provider can't guarantee the internet. They can only guarantee their piece of it. The SLA is a warranty on a component, not on the system.
Which brings us back to SD-WAN and multi-circuit setups. If the SLA only covers a component, the smart move is to stop treating that component as critical.
Right. Design your network so that no single circuit failure matters. Then the SLA becomes a nice-to-have rather than a must-have. You still want it — it keeps the provider honest and gives you some recourse — but you're not betting your business on it.
What should someone actually look for when they're reading one of these contracts?
Four things. One, the measurement period. Monthly or quarterly? Monthly is better — it limits how much downtime can accumulate before the SLA kicks in. Two, the credit calculation. Is it automatic or do you have to claim it? What's the percentage? Is there a cap? Three, the exclusions. What's carved out? Maintenance windows, force majeure, the local loop — know what's not covered. Four, the MTTR clock. When does it start? Who decides when the circuit is "repaired"? Is temporary restoration good enough?
And if the answers to those questions are vague, the SLA is vague.
If the provider won't put specific numbers in writing, the numbers don't exist. "Industry-standard uptime" is not a commitment. "Best-effort repair" is not a commitment. The SLA is only as strong as the specific, measurable commitments it contains.
I think the thing that gets overlooked in all of this is that the SLA is fundamentally a relationship document. It defines what happens when things go wrong, and that definition shapes the entire vendor relationship. A provider with a weak SLA is telling you, upfront, that they don't expect to be held accountable. A provider with a strong SLA — one that has real teeth, automatic credits, tight MTTR — is telling you they've built a network and a support organisation that can back it up.
The SLA is a signal.
It's a signal about how seriously they take their own infrastructure. If they're willing to guarantee four-hour MTTR, they have technicians on call and spare parts in stock. If they won't commit to anything specific, they probably don't.
Hilbert: I was the guy with the spare parts. Or not, depending on the month.
Go on.
Hilbert: Late nineties. Regional ISP in Ohio. My title was "network reliability engineer" which meant I drove around in a van with a rack of switches and a crimping tool. The SLA for our business customers promised four-hour MTTR. We had exactly two technicians for the entire metro area. If two circuits went down at the same time, one of them wasn't getting fixed in four hours.
What happened when you missed the window?
Hilbert: We wrote a lot of credits. Ten percent here, ten percent there. The sales team didn't care. They'd already booked the contract. The credits came out of a different budget.
So the SLA was a sales tool, not an operations commitment.
Hilbert: The operations team didn't even know what the SLAs said until someone filed a claim. I found out we had a ninety-nine point nine nine percent uptime guarantee when a customer read it to me over the phone while I was trying to figure out why their circuit was down. I had to ask him what the number meant.
What did you tell him?
Hilbert: I told him I'd be there in an hour. I was there in three. He got his credit.
You mentioned a customer with a Friday problem.
Hilbert: Printing company. Their DIA circuit dropped every Friday at about three in the afternoon. Like clockwork. It would come back after twenty minutes, sometimes thirty. Never more than forty. This went on for eleven months.
It never breached the SLA.
Hilbert: Forty minutes a month. The SLA allowed forty-three. We never paid a credit on that circuit, not once. The customer called every Friday. I drove out there six times. We replaced the CPE, we replaced the fibre patch, we checked the splice point. Nothing fixed it. Turned out to be a failing power supply in a street cabinet three blocks away. It would overheat on warm afternoons, the voltage would sag, and the switch would reboot. Took us eleven months to find it.
The SLA said you were meeting your commitments the whole time.
Hilbert: Perfect record. Ninety-nine point nine percent uptime, every month, for a circuit that failed every single week.
That's the averaging problem in one story.
Hilbert: The customer's business was losing about two thousand dollars every Friday. They couldn't process orders, couldn't send proofs to clients. Over eleven months, that's maybe ninety thousand dollars in lost revenue. The SLA would have paid them a maximum of fifty percent of their monthly fee, which was about six hundred dollars. They never even got that, because the numbers stayed inside the line.
Did they stay with the provider?
Hilbert: They left the day the contract ended. Signed with a competitor who had a worse SLA on paper but a better network in practice. The SLA didn't protect them. It protected us.
That's the thing about SLAs. They're a legal fiction that works in one direction. The provider defines the terms, measures the performance, and decides whether a breach occurred. The customer gets a document that looks like a guarantee but functions as a liability cap.
Hilbert: My brother-in-law sells enterprise fibre now. Different company, same game. He told me once that the SLA is like a warranty on a car that only covers the tires. The engine is the thing that breaks, and the warranty doesn't cover the engine. But the customer sees "warranty" and feels safe.
Does he think it's dishonest?
Hilbert: He thinks it's business. The customer wants to feel protected. The provider wants to limit exposure. The SLA lets both sides pretend they're getting what they want. He says the customers who actually understand SLAs don't read them — they just buy two circuits from two different providers and call it a day.
That's the SD-WAN logic. Redundancy beats guarantees.
Hilbert: My brother-in-law says the smartest customers are the ones who treat the SLA as a tiebreaker between otherwise identical quotes. They don't expect it to save them. They expect it to tell them which provider has the better operations team. A tighter SLA means the provider is confident they can deliver. A loose one means they're not.
The SLA is still useful. Just not for the reason it appears to be.
Hilbert: It's useful as a signal. As a contract, it's a coupon. My brother-in-law is not a trustworthy man, but he's right about that.
The signal function is real. If a provider offers a four-hour MTTR with automatic credits and no cap, they've invested in the infrastructure and staffing to back that up. If they offer "best effort" and vague uptime targets, they haven't. The SLA tells you what kind of organisation you're dealing with before you sign.
Even if the specific numbers in it are gamed.
Even then. The gaming is predictable. The investment level isn't.
The question Daniel started with was how this stuff is packaged and priced. I think the answer is that it's packaged as a contract and priced as insurance, but the insurance doesn't cover what you think it covers.
The price is dropping while the contracts are standardising. The market is mature enough now that the basic terms are table stakes. The interesting differentiation is happening in the monitoring, the support, the extras — and in the customer's own architecture, where SD-WAN and multi-circuit setups are making the SLA less relevant than it used to be.
Which raises the question of where this goes next. If SD-WAN becomes the default, and businesses routinely run two or three circuits from different providers, does the SLA on any single circuit matter at all?
It matters less. But it doesn't go to zero. There's still value in knowing that a provider will fix a broken circuit within a known window, even if you've failed over to a backup. Redundancy reduces the stakes of the SLA but doesn't eliminate them.
The shift toward application-level guarantees — connectivity as a service rather than circuits as a product — that could bring the SLA back to centre stage in a different form.
It could. If the provider is on the hook for the actual application experience — the voice quality, the video clarity, the transaction completion rate — then the SLA becomes meaningful again. But that's a much harder product to deliver, and the measurement problems get even thornier. We'll probably be talking about that in five years.
For now, the lesson is that ninety-nine point nine percent uptime is not a promise. It's a price tag. It tells you what tier of service you're buying, and it tells you what the provider is willing to put in writing. But if your business actually depends on the connection, you need something else — redundancy, architecture, a plan for when the circuit goes down. The SLA won't save you.
The SLA is the document you read after the outage, while you're waiting for the credit that won't cover the damage. The real protection is what you build yourself.
Thanks to our producer Hilbert Flumingtop.
This has been My Weird Prompts. 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.
