So Daniel sent us this one — he's been thinking about fiber rollout in Jerusalem, which he calls a true value multiplier for cities trying to cultivate knowledge work and human progress. But his real question is about what's happening under the hood right now. If you place a call from a DSL landline that's way past its sell-by date, does that call ever touch an IP network? Do two mobile subscribers talking today actually travel over IP end to end? And when exactly did that shift happen? He wants us to trace the actual routing path — the messy, incomplete migration from old-school circuit-switched telephony to the all-IP world we're supposedly living in.
And the honest answer to every one of those questions is: it depends. Which is the most frustrating and also the most interesting answer in telecom.
It depends on what, exactly?
On whether there's a media gateway sitting at the local exchange. On whether your carrier has decommissioned its Class 5 switches yet. On whether you're in a coverage gap that forces a fallback to circuit-switched. The infrastructure stack in twenty twenty-six is this bizarre layer cake where nineteenth-century copper pairs, mid-century circuit switching, and modern IP routing all coexist — sometimes on the same call.
A layer cake where some of the layers are made of actual dirt.
Pretty much. So let's trace this through three specific call scenarios, because the answer is genuinely different for each one. Scenario one: a landline call from a DSL-connected home to another landline. Scenario two: mobile to mobile. Scenario three: the hybrid — mobile calling landline. That third one is where things get especially weird.
Start with the DSL landline. This is the case Daniel's really asking about — the person who's stuck on DSL and picks up what looks like a regular phone.
Right. And here's the first thing most people get wrong. If you have DSL, your internet traffic is digital — it's modulated onto frequencies above the voice band on that copper pair. But the voice call itself? In a huge number of markets, including Israel, that voice call is still plain old analog circuit-switched from your handset all the way to the local exchange. DSL rides on top of POTS — Plain Old Telephone Service. They share the same copper pair, but they're logically separate services.
So the internet part is digital, but the voice part is still your great-grandfather's telephone call.
And what happens at the exchange is the critical variable. If that exchange has been upgraded to what's called a softswitch — an IP-based call control system — then your voice gets converted from analog to digital at a media gateway right there at the exchange edge. From that point forward, it's SIP and RTP packets traversing the carrier's IP backbone. But if the exchange still has a Class 5 switch — which is the traditional circuit-switched telephone exchange hardware — then your call stays circuit-switched until it reaches a tandem office further up the network hierarchy.
And how many Class 5 switches are still out there?
Far fewer than a decade ago, but not zero. AT&T decommissioned its last Class 5 switches in twenty twenty-two when it fully retired its PSTN network. That was a landmark moment — the largest carrier in the United States saying "we're done with circuit-switched." But that doesn't mean every copper line disappeared. AT&T stopped offering new POTS lines in twenty nineteen, but they're still maintaining existing ones, and they still operate media gateways to bridge those legacy copper connections to the IP core.
So AT&T pulled the plug on the switches but kept the gateways running. That's like demolishing the train station but keeping a shuttle bus to the platform that no longer exists.
That's exactly what it is. And the shuttle bus — the media gateway — is expensive to maintain. It's specialized hardware that converts between TDM, time-division multiplexing, which is how circuit-switched networks chop up and interleave calls, and SIP slash RTP, which is how VoIP works. Every conversion step adds a tiny bit of latency and potential jitter. The more gateways in the path, the worse the call quality.
What about the UK? Daniel mentioned he's originally from Ireland, and BT has been making noise about copper switch-off for years.
BT's timeline has been a moving target. They announced a "stop sell" on new copper-based services in September twenty twenty-three — meaning you couldn't order a new copper line in most of the UK from that point. The full switch-off was originally targeted for December twenty twenty-five, but it's been delayed multiple times. The regulatory hurdles are significant, especially around lifeline voice services and emergency calling.
Lifeline meaning elderly or vulnerable people who rely on a landline that works during a power cut.
And that's one of the hidden tradeoffs in this whole migration. Copper POTS is line-powered from the central office. The central office has backup generators. If your neighborhood loses power, your copper landline still works. Fiber requires an ONT — an Optical Network Terminal — at your premises, and that ONT needs local power. If you don't have a battery backup, your fiber-based voice service goes dead the moment the electricity cuts out.
Which we saw play out in Texas.
The twenty twenty-one winter storm was the case study that changed the regulatory conversation. Widespread power outages knocked out fiber-based voice services across huge swaths of the state because most ONTs didn't have battery backup — or the batteries had degraded and nobody had replaced them. Meanwhile, the copper POTS lines kept humming along, powered from central offices that had generator backup. That single event has probably slowed copper retirement in a dozen states.
So there's a genuine resilience argument for keeping copper alive, even if it's technologically ancient. It's the cockroach of infrastructure — unglamorous, but nearly impossible to kill.
And that resilience argument is strongest for voice. For data, everyone's fine with fiber. But voice has regulatory obligations — E nine one one reliability mandates, lifeline service requirements — that make regulators very nervous about pulling the copper plug. Japan is an interesting comparison here. NTT has pushed fiber to over ninety percent coverage, which is extraordinary, but they still maintain copper for emergency voice services. Their compromise has been deploying fiber to the home with mandatory battery backup.
Alright, so that's the DSL landline story. The call might go IP after the exchange, or it might stay circuit-switched for much of the path, depending on what hardware your local telco has and hasn't retired. Now scenario two — mobile to mobile.
This is the cleanest story. Since roughly twenty fourteen to twenty sixteen, all the major US carriers — Verizon, AT&T, T-Mobile — have mandated VoLTE as the default voice layer. VoLTE stands for Voice over LTE. The call is encoded as IP packets right at the handset, travels through the carrier's IMS core — that's the IP Multimedia Subsystem, which is a standardized architecture for delivering multimedia services over IP — and is routed entirely over IP from end to end.
So if I'm on my smartphone calling another smartphone, and we're both on modern networks with good coverage, my voice never touches a circuit-switched network.
Correct. It's IP all the way. The handset converts your voice to packets using a codec — typically AMR-WB, Adaptive Multi-Rate Wideband, which is why VoLTE calls often sound noticeably better than old-school cellular calls. Those packets traverse the carrier's IP backbone, hit the IMS core for call control and routing, and get delivered to the other handset as IP packets. The other handset decodes them back to audio.
And when did the old way die?
The old way was circuit-switched cellular. GSM and CDMA networks used circuit-switched fallback for voice — the phone would essentially drop down to a older network technology to make a call. AT&T shut down its 2G GSM network in twenty seventeen. Verizon shut down its 3G CDMA network on December thirty-first, twenty twenty-two. Those shutdowns forced the transition. Once there's no circuit-switched network to fall back to, VoLTE becomes the only option.
So Verizon's New Year's Eve twenty twenty-two was less about champagne and more about finally killing CDMA.
I suspect their engineers celebrated both. But here's the nuance Daniel's question deserves. VoLTE is the default, but circuit-switched fallback — CSFB — still exists in coverage gaps. If you're in a rural area with weak LTE signal, your phone might still drop to 3G for a voice call, assuming the carrier still operates a 3G network. T-Mobile, for example, still runs a 2G GSM network in some areas, primarily for machine-to-machine communications and legacy roaming. So a call that starts as VoLTE could, in theory, fall back to circuit-switched mid-call if you drive through a coverage hole.
Though that's increasingly rare.
Increasingly rare, yes. And with 5G standalone — true 5G with its own core network, not just 5G radio bolted onto an LTE core — there's VoNR, Voice over New Radio, which is essentially VoLTE for 5G. As that becomes universal, the last vestiges of circuit-switched fallback will disappear from mobile. But we're not quite there yet.
So mobile-to-mobile is the success story. IP native, end to end, with a few fading edge cases. Now the messy one.
Scenario three: mobile calling landline. This is where the media gateway becomes the star of the show, and not in a good way. Your mobile call starts as IP — VoLTE from the handset, through the IMS core, all IP. But at some point, it needs to reach a copper landline that may or may not be connected to an IP exchange. If the landline is on a modern fiber connection with VoIP, great — the call stays IP the whole way. But if the landline is on legacy copper behind a Class 5 switch, or even behind a media gateway at an older exchange, the call has to be converted from IP to TDM.
And that conversion happens at a media gateway somewhere in the carrier's network.
Yes. And here's the thing — carriers have been consolidating these gateways. A decade ago, you might have had a media gateway at every local exchange. Now, a carrier might have a handful of them serving an entire region. So your mobile call might travel hundreds of miles as IP packets, get converted to TDM at a regional gateway, traverse another chunk of circuit-switched network, and then hit the copper local loop at the destination.
Every one of those conversion steps is an opportunity for things to get slightly worse.
Jitter, packet loss, added latency. The conversion from IP to TDM and back again is not lossless. TDM is synchronous — it expects a continuous stream of bits at a fixed rate. IP is asynchronous — packets arrive at irregular intervals and need to be buffered and re-clocked. The media gateway has to do that re-clocking, and if the buffer is too small, you get packet loss. If it's too large, you get added latency. It's a hard engineering problem, and it's one of the reasons why a mobile-to-landline call can sometimes sound worse than either a pure mobile call or a pure landline call.
So Daniel's intuition about fiber rationalizing the infrastructure stack is basically correct — if everyone were on fiber with end-to-end IP, the media gateways would disappear, and with them all those conversion artifacts.
That's the vision. And in places with high fiber penetration, it's largely the reality. But the bottleneck isn't the backbone — the backbone has been IP for years. The bottleneck is the last mile. That copper pair running from the exchange to the house. As long as millions of those are still in service, the media gateways have to stay.
Let's bring this back to Jerusalem, since that's where Daniel is and where he's thinking about this. He mentioned there are still patches of the city where DSL and coax are the only options. What does a call actually look like from one of those neighborhoods?
It depends on the provider. Bezeq is the incumbent — they own most of the copper plant. If you're on a Bezeq DSL line and you pick up a landline phone, that call is almost certainly analog circuit-switched over the copper pair to the local exchange. At the exchange, Bezeq has been upgrading to softswitches for years, so there's a good chance it hits a media gateway and goes IP from there. But I can't tell you with certainty that every Bezeq exchange in Jerusalem has been upgraded. Some of the older exchanges in the city's more historic neighborhoods may still have legacy switching equipment.
And if Daniel's calling from a VoIP service that rides on top of his internet connection?
That's a different path entirely. A VoIP call from a DSL-connected home goes: copper local loop to the DSL modem, through the ISP's IP backbone, out to the internet, to the VoIP provider's servers, and then — depending on the destination — either stays IP all the way if the recipient is also on VoIP, or hits a PSTN gateway somewhere to reach a traditional landline. That path can have five or more distinct conversion points.
Compare that to a fiber-to-the-premises call.
Fiber ONT at the home, direct IP routing through the carrier's network, and if the destination is also on fiber, it's IP end to end. Two endpoints, one protocol, no conversions. That's the rationalized stack Daniel's talking about. Fewer conversion points means lower latency, fewer failure modes, and better call quality.
So the practical question for someone in a DSL-only area who cares about call quality — is there anything they can actually do?
Yes. If you're stuck on DSL but want better voice, consider a VoIP provider that peers directly with major carriers, bypassing as many PSTN gateways as possible. Services like Vonage or Ooma still go over the public internet, but their routing is optimized — they have direct interconnects with carrier networks, which reduces the number of conversion hops. It's not as clean as native fiber voice, but it's better than the default POTS path.
And for someone who does have fiber, the thing to check is whether your ONT has a battery backup.
If you're on fiber-based voice service and your ONT doesn't have a working battery, you lose phone service the moment the power goes out. Some providers include a battery backup unit by default. Many don't. It's worth checking — and worth replacing the battery every couple of years, because they degrade.
The Texas lesson.
The Texas lesson exactly. And for cities planning fiber rollout — and this is relevant to Jerusalem's ongoing deployment — mandate battery backup for ONTs from day one. The "fiber is better" argument is weakened considerably if it fails during the one scenario where voice service matters most: an emergency during a power outage.
There's also a question listeners can ask their carrier that cuts through a lot of the ambiguity. "Is my voice call routed through a media gateway or is it end-to-end IP?"
That's the right question. For AT&T customers, the answer is: if you're on fiber, it's VoIP end to end. If you're on legacy copper, it hits a media gateway at the exchange. For Verizon, FiOS voice is VoIP, but their copper footprint still has POTS lines. For Bezeq and HOT customers in Israel, it's worth asking directly. The customer service rep might not know the answer immediately, but the question itself signals that you understand the infrastructure, and they'll often escalate to someone who does.
I want to zoom out for a second. Daniel framed this whole thing around fiber as a value multiplier for cities — the idea that a versatile, high-throughput data pipeline can attract knowledge work and scientific progress. And he's right. But the infrastructure migration story suggests something messier. The transition isn't a clean upgrade. It's a coexistence that might persist for decades.
That's the core tension. From an engineering perspective, retiring the PSTN makes perfect sense. Maintaining two parallel networks — one IP, one circuit-switched — is enormously expensive. The operational cost of keeping Class 5 switches and media gateways running, training technicians on legacy equipment, maintaining spare parts for hardware that hasn't been manufactured in years — it adds up. Carriers want out.
But from a regulatory and reliability perspective, copper has attributes that fiber can't easily replicate. Line power. Proven reliability over a century. No dependency on customer-maintained batteries.
And those attributes matter most to the people who are hardest to migrate. Elderly customers with lifeline services. Rural communities with unreliable power grids. Emergency services that need guaranteed uptime. The last ten percent of the migration is always the hardest, and in telecom, that last ten percent might take another decade.
So the open question is: will we ever fully retire the PSTN? Or will we shrink it to a minimal footprint and keep it alive indefinitely as a resilience layer?
I think the answer is the latter. We'll shrink it. The number of POTS lines will continue to decline. More exchanges will be upgraded. But a minimal circuit-switched core, with media gateways at the edge, will persist for emergency services and legacy customers. The operational cost of that minimal footprint is high per-line, but the regulatory cost of eliminating it entirely is higher.
And on the mobile side, the transition is essentially complete. VoLTE and soon VoNR are the defaults. The circuit-switched fallback will fade as 5G standalone coverage expands and carriers finally shut down their last 3G networks. The mobile world is all-IP in a way the fixed-line world may never fully achieve.
Which creates an interesting asymmetry. A call between two smartphones in twenty twenty-six is IP-native, end to end, with wideband audio codecs that make voices sound remarkably clear. A call between two landlines might still be analog circuit-switched for most of the path, with the narrowband audio of a nineteen-fifties telephone network. The infrastructure you're standing on determines the quality of the experience, even though both are "phone calls."
And most people have no idea which infrastructure they're actually using.
That's the invisible layer cake. You pick up a phone and dial. The fact that your voice might travel through a nineteenth-century copper pair, a twentieth-century circuit switch, and a twenty-first-century IP backbone — all on the same call — is completely opaque to the user. But it's there, under the street, shaping the quality and reliability of every conversation.
Daniel's instinct about fiber rationalizing the stack is correct in the long run. A single IP-based infrastructure carrying voice, data, and video is simpler, cheaper, and more capable than the layered mess we've inherited. But the transition is slow because the old stuff works — especially for voice — and the new stuff has failure pattern the old stuff didn't.
The deprecation arc for fiber is gentle, as Daniel put it, because it was designed for high throughput from the start. We're not going to outgrow fiber the way we outgrew copper. The physics of light in glass gives us headroom for decades. But the deprecation arc for copper voice isn't about throughput at all — it's about resilience and regulatory obligation. Those are harder to engineer away.
So if you're in Jerusalem, in one of those DSL-only patches, and you pick up your landline — your voice is probably going analog over copper to the local exchange, then IP from there. Unless the exchange hasn't been upgraded, in which case it's circuit-switched for much of the journey. And if you're on your smartphone calling another smartphone, you're IP all the way, VoLTE from handset to handset, unless you hit a coverage dead zone.
And if you're on your smartphone calling a landline, you're IP until a media gateway somewhere converts you to TDM for the final leg over copper. That gateway might be at the local exchange or it might be hundreds of miles away, depending on how aggressively your carrier has consolidated.
The invisible hand of infrastructure consolidation, reaching into every phone call.
And now: Hilbert's daily fun fact.
Hilbert: In the nineteen eighties, a marine biologist studying axolotl limb regeneration on a research trip to Tuvalu discovered that the handwritten field notes from the expedition had been accidentally bound into a government fisheries report on tuna migration — meaning the world's most detailed observations of axolotl blastema formation sat mislabeled in a Tuvaluan ministry archive for over thirty years.
That is the most specific mis-filing I've ever heard of.
I have so many questions about how that binding error happened, and I know none of them will ever be answered.
The bigger question, the one that'll be with us for years, is whether we ever truly say goodbye to the circuit-switched world. The operational cost of keeping it alive is enormous. The regulatory and reliability arguments for keeping it are strong. My bet is we land somewhere in the middle — a skeleton PSTN, stripped down to its most essential functions, persisting alongside an otherwise all-IP world.
And as 5G standalone with VoNR becomes universal, the mobile side will be fully IP. But those PSTN gateways will bridge the two worlds for at least another decade. The layer cake isn't going away — it's just getting thinner.
Thanks to our producer Hilbert Flumingtop. This has been My Weird Prompts. You can email the show at show at my weird prompts dot com.
We'll be back soon.
