Daniel's been thinking about fibre optic infrastructure again, and he's landed on a question that doesn't get asked nearly enough. He points out that we tend to think of fibre as cutting-edge because it's marketed as the consumer upgrade path, but it's actually been in the ground carrying Internet traffic for a long time already. The real question is how future-proof all that existing glass actually is. He brings up dark fibre and the enormous unused capacity sitting there, and then asks the thing that makes you sit up: if we shut down DSL and coax tomorrow and moved every subscriber to fibre to the house, what would happen? He also mentions the pandemic-era genomic data transfers, petabytes moving across continents on fibre and even on special flights, and notes that the case for fibre isn't just knowledge workers, it's scientific collaboration. And he's right that the companies laying the fibre are unreliable narrators about actual deployment reach.
The deployment claims are a whole separate irritation, we should come back to that. But the core question here is interesting because it inverts how most people think about fibre. The marketing frame is "this is the new thing, upgrade now," and the reality is the fibre in the ground is already, in some cases, forty years old. The first transatlantic fibre cable, TAT-8, went live in nineteen eighty-eight. That's nearly four decades of fibre carrying data under the ocean. The stuff isn't new. What's new is pulling it to the side of your house.
Forty years. So the cable's middle-aged and nobody's worried about it.
That's exactly the point. And the reason nobody's worried is that the capacity of a single fibre strand has increased by roughly a factor of a million since those early deployments. Same glass, or very similar glass, but the electronics at both ends keep getting better. The first transatlantic fibre carried about forty thousand phone calls. Today a single strand can carry something on the order of four hundred terabits per second in laboratory conditions. The glass didn't change that much. The lasers and the modulation schemes did.
So the physical cable is the part that ages well. It's everything plugged into it that gets swapped out.
Yes. And that's the answer to Daniel's hypothetical, or at least the beginning of it. If we shut down all the copper tomorrow and moved everyone to fibre to the home, the fibre in the ground could handle it. The bottleneck wouldn't be the glass. It would be the endpoints, the aggregation points, the backhaul from the neighbourhood nodes. But the strands themselves, the actual light-carrying infrastructure, have headroom that's almost absurd.
Almost absurd. Give me the numbers that make it absurd.
So a typical fibre to the home deployment today uses GPON, gigabit passive optical network. That's two point five gig down, one point two five gig up, shared across something like thirty-two or sixty-four homes. That's what most people are getting when they sign up for fibre. But that same fibre strand, if you change the optics at both ends, can carry forty gig, a hundred gig, four hundred gig. The limiting factor in consumer fibre isn't the fibre. It's the cost of the termination equipment. The glass itself is capable of vastly more than what any residential customer is being sold.
So the glass is bored.
The glass is profoundly bored. And that's before you get into the multiple wavelengths. A single fibre can carry dozens or even hundreds of different colours of light simultaneously, each carrying its own independent data stream. That's wavelength division multiplexing. The fibre that's already in the ground, the stuff that was pulled through conduits ten or fifteen years ago, can be upgraded just by changing the transceivers at each end. You don't need to dig anything up.
Which is the expensive part. Digging.
Digging is eighty percent of the cost of any fibre deployment. The cable itself is cheap. The labour and the civil engineering are what make it expensive. So once the glass is in the ground, you have an asset that gets more valuable over time because the electronics keep improving and you never have to touch the trench again.
So Daniel's hypothetical, everyone switches tomorrow. The fibre handles it. But you said that's the beginning of the answer. What's the rest?
The rest is that the places where fibre hasn't been deployed yet are the problem. And those places exist in large numbers, even in relatively developed countries. If you shut down DSL and coax tomorrow, you'd strand millions of households. Not because the fibre couldn't carry their traffic, but because there's no fibre connected to their homes. The last mile is still copper in a lot of places.
And the companies that own that copper are sunsetting it. Daniel mentioned DSL and coax retirement.
Right. AT&T has been aggressively retiring copper. They've been pushing to shut down their legacy DSL network for years, and they've largely succeeded in getting regulatory approval to do so. In the UK, Openreach is aiming to retire the PSTN, the public switched telephone network, entirely by the end of twenty twenty-five, which means copper phone lines go away. Parts of Europe are further along. Estonia shut down its copper network years ago. The trend is clear: copper is being turned off, and the replacement is fibre. But the transition isn't complete, and if you accelerated it to tomorrow, you'd have a coverage gap, not a capacity gap.
The coverage gap is real, but it's a deployment problem, not a physics problem. What about the physics? You mentioned Shannon limits and nonlinear effects when you were looking at this.
There is an actual physical limit to how much data you can push through a fibre. Claude Shannon's work on information theory gives us a channel capacity limit, and for optical fibre, we're starting to get close to it in the long-haul systems. The nonlinear Shannon limit for standard single-mode fibre is somewhere around one hundred terabits per second per fibre. In the lab, people have pushed past that using multi-core fibres and multi-mode fibres and spatial division multiplexing, which is essentially putting multiple parallel light paths inside a single cable. But for the standard fibre that's already in the ground, we're approaching the theoretical ceiling.
Approaching. How close?
Commercial systems are running at maybe twenty or thirty terabits per fibre. The lab record for standard single-mode fibre is somewhere around one hundred and fifty terabits. So we have maybe a factor of five or so left in the standard fibre before we hit the wall. After that, you need new fibre with more cores or more modes. But that factor of five is still enormous. And it'll take years, probably a decade or more, to exhaust it with real-world traffic growth.
And by then, the fibre that's being pulled today will have paid for itself many times over.
Many times over. The economic life of a fibre cable is typically depreciated over twenty or twenty-five years, but the actual physical life is more like forty or fifty years, and the useful life keeps getting extended by better electronics. So you have an asset that you book as a twenty-year investment and it keeps delivering value for twice that long. That's almost unheard of in technology infrastructure.
Contrast that with the copper it's replacing. DSL was a hack. A brilliant hack, but a hack.
It was a magnificent hack. DSL takes a twisted pair of copper wires designed for voice, which is a four-kilohertz signal, and it pushes megahertz-range frequencies over it. The distance limitations are brutal. After about a kilometre and a half, the signal degrades to the point where you're getting single-digit megabit speeds. And that's in ideal conditions. Old copper, bad splices, water in the conduit, all of that makes it worse. Coax is better, because it was designed for higher frequencies for television, but it's still a shared medium with noise issues and upstream bandwidth that's a fraction of the downstream.
Coax is the cable TV pipe being asked to be a two-way Internet connection.
DOCSIS, the standard that runs over coax, has been remarkably successful. DOCSIS three point one can do multi-gigabit speeds. DOCSIS four point zero promises symmetrical multi-gigabit. But every generation is squeezing more out of a medium that was never designed for this. It's a series of increasingly impressive tricks. Fibre isn't a trick. It's the thing the trick was trying to approximate.
That's the line. Fibre isn't a trick. It's the thing the trick was trying to approximate.
I'll stand by that. And it connects to Daniel's point about scientific collaboration. The pandemic genomic data transfers he mentioned, those were moving datasets that were too large for anything but fibre or physically shipping hard drives. The COVID-19 High Performance Computing Consortium was moving petabytes of sequencing data between labs. Some of that went over dedicated research networks like Internet2 and ESnet, which run on fibre at hundreds of gigabits per second. And some of it, as Daniel noted, went on planes. You load up a server rack with hard drives and fly it across the ocean, because for truly enormous datasets, the bandwidth of a 747 full of hard drives still beats a fibre link.
Sneakernet at scale. The latency is terrible but the throughput is unbeatable.
There's a famous quote, I think from Andrew Tanenbaum, never underestimate the bandwidth of a station wagon full of tapes hurtling down the highway. The pandemic updated that to a cargo plane full of sequencing data. But the point is that even with that brute-force approach, the endpoints are fibre-connected research institutions. The data gets generated on one continent, flown to another, and then pulled into a compute cluster over a fibre link. The fibre is the on-ramp and the off-ramp. The plane is the middle mile.
So fibre is essential infrastructure even when you're literally flying hard drives around.
It's the last hundred metres and the first hundred metres. And for smaller but still enormous datasets, the kind that don't justify a chartered flight, the fibre does the whole job. The Square Kilometre Array radio telescope, when it's fully operational, will generate something like one hundred and sixty terabytes of data per second. That's not a typo. Per second. They're building dedicated fibre links from the telescope in South Africa and Australia to regional processing centres, and from there to the world. Without fibre, that instrument doesn't work.
One hundred and sixty terabytes per second. So the fibre that's bored today carrying Netflix is eventually going to be carrying radio telescope data.
Or both, simultaneously, on different wavelengths. That's the beauty of the wavelength division multiplexing. You can lease a wavelength to a research network and another wavelength to a commercial ISP and they don't interfere. The fibre doesn't care what the photons represent.
Let's go back to the deployment claims for a moment. Daniel said the companies laying fibre are unreliable sources for actual reach. I've seen this in Jerusalem. You check the coverage map, it says your building is connected, and then the technician shows up and says there's no fibre within three hundred metres.
This is a universal experience. In Israel, the fibre market is competitive in theory. Bezeq has its own fibre network, IBC has one, Partner has one, there are a few others. But the coverage maps are aspirational. A street might be marked as covered because the conduit is there or because one building on the block has been connected, but your building might not have the drop cable pulled from the street to the basement. And that last fifty metres can take months.
And the ISPs are selling service based on those maps. You sign up, they take your money, and then they figure out whether they can actually deliver.
It's a known problem. The regulator in Israel has been trying to get the companies to report coverage more accurately, but the incentives are all wrong. If your competitor claims coverage on a street, you have to claim it too, or you look like you're losing. So everyone's coverage map is optimistic. The reality on the ground lags the maps by six to eighteen months.
Which makes fibre deployment feel like a political campaign promise rather than an infrastructure project.
And that's frustrating because the physical work is actually happening. Israel has gone from single-digit fibre penetration to something like seventy percent of households passed in about five years. That's fast by international standards. But the gap between "passed" and "connected" is where the frustration lives. "Passed" means the fibre is in the street. "Connected" means it's in your living room. Those are different things, and the industry is not always careful about the distinction.
So if we go back to Daniel's hypothetical, the issue isn't the fibre backbone. It's the last fifty metres times a few million households.
Right. And it's also the in-home equipment. If you forced everyone onto fibre tomorrow, you'd need tens of millions of optical network terminals, the little boxes that convert the light signal back to Ethernet. The supply chain for those is not infinite. There'd be a shortage. But these are solvable problems. They're logistics problems, not physics problems. The physics says the fibre in the ground can do it.
What about the coax networks? DOCSIS four point zero is promising symmetrical multi-gigabit. If coax can do that, why bother replacing it with fibre?
Because DOCSIS four point zero is squeezing the last drops out of a medium that's at its limit. The upgrades required to make coax do symmetrical multi-gigabit are substantial. You need to push the fibre deeper into the neighbourhood, to nodes that serve maybe a few dozen homes instead of a few hundred. You need to upgrade all the amplifiers along the coax plant. At some point, the cost of upgrading the coax approaches the cost of just pulling fibre the rest of the way. And once you've pulled the fibre, you're done. No more upgrades to the physical plant, ever, or at least for decades. With coax, you'll be back in ten years trying to squeeze out DOCSIS five point zero.
So coax is on the same trajectory DSL was on. Incremental upgrades getting more expensive until you hit the point where replacement is cheaper.
That's the arc. And the cable companies know this. Comcast and Charter in the US are deploying fibre for new builds and for areas where the coax plant is end-of-life. They're not ripping out working coax, but they're not installing new coax either. The future is fibre. The coax is being managed for decline.
Managed for decline. That's a phrase that should make anyone on coax a little nervous.
It should, but the decline will take a decade or more. Nobody's turning off coax tomorrow. The cable companies have too many customers on it, and DOCSIS four point zero will keep it competitive for a while. But the investment direction is clear.
Let's talk about the dark fibre Daniel mentioned. He's right that there's enormous unused capacity. How much are we talking about?
It's hard to get exact numbers because the fibre owners don't always disclose it, but estimates suggest that in long-haul networks, something like fifty to seventy percent of the fibre strands are dark. They were pulled during the dot-com boom, and then the traffic growth didn't match the expectations, so they sat unused. Now traffic is growing again, and those dark fibres are being lit up. But there's still a lot of headroom.
And dark fibre is different from unused capacity on a lit fibre.
Yes. A lit fibre might be carrying a few wavelengths out of the dozens or hundreds it could carry. That's capacity that can be turned up by adding transceivers. Dark fibre is a strand with no light in it at all. You need to put the electronics on both ends. But the glass is there. It's already in the conduit. The hard part is done.
So the overbuilding during the dot-com bubble turned out to be a gift.
An accidental gift. The companies that pulled all that fibre went bankrupt, but the fibre stayed in the ground. The creditors sold it for pennies on the dollar, and the companies that bought it got a century's worth of capacity for almost nothing. It's one of the great infrastructure windfalls in history.
There's something almost comforting about that. The bad business decision became a public good.
Infrastructure often works that way. The private sector overbuilds in a frenzy, goes bust, and the physical assets remain, providing value for decades that the original investors never captured. The railways were the same. The canals before that. Fibre is just the latest version of the pattern.
So if we're thinking about fibre as essential infrastructure, the way Daniel frames it, the market alone doesn't necessarily get you the right outcome. You get boom and bust, you get coverage gaps, you get optimistic maps.
The market is good at building fibre where the returns are high. Dense urban areas, business districts, those get fibre quickly because there are enough customers per kilometre to justify the trenching. The problem is rural areas, low-density suburbs, places where the cost per home passed is high. Those places need some form of subsidy or universal service obligation, or they get left behind. And if copper is being turned off, being left behind means no service at all.
Which is where the essential infrastructure argument gets teeth. If you accept that Internet access is essential, then you can't leave it entirely to the market.
And most developed countries have accepted that, at least in principle. The US has the Rural Digital Opportunity Fund. The UK has the Universal Service Obligation for broadband. Israel has a fibre deployment mandate with coverage targets. The debate is about the details, the speed thresholds, the funding mechanisms. But the principle that everyone should have access to fibre-grade connectivity is broadly settled. The execution is what's messy.
What about the developing world? Daniel mentioned knowledge economy nodes. If a city wants to attract tech investment, is fibre a prerequisite now?
Increasingly, yes. You can't run a data centre without fibre. You can't host a research institution without fibre. You can't attract remote workers without fibre. It's not sufficient, you also need reliable power and a skilled workforce and all the rest, but it's necessary. And the countries that are building it out aggressively, places like Rwanda and Kenya, are seeing returns in the form of tech hub development. Kigali has better fibre than some European cities.
Rwanda leapfrogged.
They skipped the copper generation entirely. They went straight from almost nothing to fibre and mobile. And mobile is the other piece of this. 5G needs fibre backhaul. Every cell tower needs to be connected to a fibre network to handle the traffic. So even if the last hop to the user is wireless, the network behind it is fibre. The wireless part is just the last few hundred metres.
So the sunsetting of 2G and 3G that Daniel mentioned, that's not just about spectrum refarming. It's also about the backhaul requirement.
Right. A 2G tower could be backhauled with a few E1 lines, which are basically copper. A 5G tower needs multi-gigabit fibre. As you turn off the old mobile networks and turn on the new ones, you're implicitly requiring fibre everywhere there's a cell site. The mobile operators don't always own that fibre, they lease it, but the fibre has to be there.
Let's go back to the future-proofing question. Daniel asked specifically how future-proof the fibre already in the ground is. You said we have maybe a factor of five left in standard single-mode fibre. What comes after that?
Multi-core fibre. Instead of one light path through the centre of the glass, you have seven cores, or nineteen, or more, each carrying its own independent data streams. The fibres are already being manufactured and tested. NEC and NTT in Japan have demonstrated multi-core fibres carrying over a petabit per second over long distances. That's a thousand terabits. The cable looks identical from the outside. Same diameter, same connectors, same installation process. But inside, instead of one core, you have multiple.
So the next generation of fibre is a drop-in replacement for the current generation.
In theory. The connectors are more complex because you need to align multiple cores precisely, and the splicing is harder. But the conduit is the same size. The trenching is the same. If you pull multi-core fibre today, you're future-proofing for decades beyond what standard fibre gives you. But almost nobody is pulling multi-core fibre for access networks yet. It's still too expensive, and standard fibre has enough headroom that it doesn't make economic sense.
So the fibre being pulled to homes today is not the most advanced fibre that exists. It's the fibre that's cost-effective given current demand.
And that's the right call. You don't pull multi-core fibre to a residential street when standard single-mode fibre can handle everything those homes will need for twenty years. By the time you need multi-core, the electronics will be cheaper and the installation techniques will be mature. The infrastructure approach is to pull what you need now plus a healthy margin, and leave conduit space for future pulls.
Conduit space. That's the other thing that matters.
Conduit is more important than the fibre itself. If you have a good conduit network with spare capacity, you can pull new fibre through it without digging. If you direct-buried the fibre without conduit, you're digging every time you want to upgrade. The smart deployments use microducts, small flexible conduits that can be blown into larger conduits, and then fibre can be blown through the microducts. It's a system designed for incremental upgrades without ever opening a trench.
So the truly future-proof infrastructure decision isn't the fibre, it's the conduit.
The conduit and the rights of way. If you've secured the right to use the land and you've put in good conduit, you can upgrade the fibre inside it forever. The physical path is the scarce resource, not the glass.
That's the thing that should be in the essential infrastructure conversation. Not just who's pulling fibre today, but who's putting in conduit that'll be good for a century.
That's where government policy can actually help. Requiring conduit to be laid whenever a road is opened. Coordinating between utilities so that water, gas, electricity, and telecom conduits go in the same trench. That coordination saves enormous amounts of money and disruption. Some countries are good at this. Singapore is famously good at it. Most countries are not.
Israel is not.
Israel is emphatically not. We have a long and storied tradition of digging up the same street three times in six months for three different utilities. It's practically a national sport.
The Jerusalem municipal orchestra. The percussion section is jackhammers.
The melody is the sound of a fibre being cut by a backhoe that didn't check the utility maps.
Which brings us to the actual most future-proof thing. Accurate as-built records of where the fibre is buried.
You'd be amazed how bad those records are in some places. Fibre was pulled in a hurry, the documentation was an afterthought, and twenty years later nobody knows exactly where it is. There are companies that specialise in finding and mapping unknown fibre. It's a whole industry.
Of course it is. So the answer to Daniel's question, how future-proof is the fibre in the ground, is: very, if it was installed well, with conduit, and someone wrote down where they put it. Less so if it was direct-buried in a panic during a bubble and the guy who knew the route retired.
That's about right. The technology of the glass itself is remarkably durable. The weak points are the installation practices and the business models. The glass will outlast the companies that laid it, just like it outlasted the dot-com companies. The question is whether the physical plant was built to be upgraded or whether it was built to be used once and abandoned.
The DSL and coax that Daniel mentioned, those were built to be used once. They've been stretched far beyond their design life, and they've served well, but they're at the end of that stretch.
The copper network was designed for voice. It was a triumph of engineering that it carried the Internet for thirty years. But it's done. The future is glass, and the glass we have is mostly good enough for the future we can see. Beyond that, we'll pull new glass through the same conduits. The conduits are the real infrastructure. The glass is consumable.
The glass is consumable. That's the thing to take from this. The fibre optic cable that everyone treats as permanent infrastructure is actually a replaceable component inside the permanent thing, which is the path through the ground.
The permanent thing is mostly a legal and civil engineering asset, not a technology asset. The right of way, the conduit, the access points. Those last a century. The fibre lasts forty years. The electronics last five to ten. The whole system works because each layer can be upgraded independently.
That layering is what makes it future-proof. Not the material, but the architecture.
Hilbert: Nineteen ninety-nine. I pulled fibre in four cities.
Four.
Hilbert: Chicago, Denver, Phoenix, and a stretch between two towns in Nebraska I won't name because the mayor still sends me a Christmas card and I don't want him to know where I am. The Nebraska job was seventy-two miles of single-mode fibre through cornfields. The farmer whose land we crossed got paid by the foot and he counted every foot himself, walking behind the trencher with a measuring wheel.
That's a man who understood the value of his right of way.
Hilbert: He understood the value of being a nuisance. He'd stand in the trench if he thought we were an inch off the easement. The fibre we pulled was Corning SMF-28. Standard single-mode. The same stuff that's in the ground in most of the world. I looked it up a few years ago. That same fibre, the exact same model, is still being manufactured and sold. Twenty-seven years later. The spec sheet has been revised, the attenuation numbers are a little better, but it's the same product. You can splice fibre pulled in nineteen ninety-nine to fibre pulled in twenty twenty-six and the light doesn't know the difference.
The standard hasn't changed in three decades.
Hilbert: The standard is the standard because it works. The innovation is in what you push through it, not in the glass. I was at a trade show in two thousand four, and a vendor was demonstrating a new modulation scheme that could push ten gig over SMF-28. People were losing their minds. Ten gig, over standard fibre. Now they're pushing four hundred gig over the same glass. In twenty years it'll be a terabit. The glass doesn't care.
The glass really doesn't care. It's just a waveguide.
Hilbert: The thing I remember from Nebraska, the farmer asked me how long the fibre would last. I told him fifty years, which was the standard answer. He said, "Fifty years. I put in drainage tile in nineteen sixty-two and it's still working. Your fifty years is not impressive." And he was right. The drainage tile was just clay pipe in a trench. Same principle. The hole in the ground is the asset. The pipe is replaceable.
The farmer understood infrastructure better than the telecom industry.
Hilbert: The farmer understood that digging is expensive and everything else is negotiable. We finished that Nebraska job in six weeks. I drove back through there about ten years ago. The fibre was still in the ground. The corn was still growing. The farmer had died, but his son was still farming the same land. I didn't stop.
The easement outlasts the farmer.
Hilbert: The easement outlasts everything. That's the point. The fibre in that trench has been upgraded three or four times at the endpoint since we pulled it. New lasers, new multiplexers, new protocols. Nobody has touched the glass. Nobody will touch the glass until someone decides they need more cores, and even then they'll probably pull a new cable through the same conduit, if we left conduit. We did. I made sure.
The answer to Daniel's question is that the fibre in the ground is future-proof because the hard part, the trench, was already dug. The glass is just the first thing that went in it.
Hilbert: The glass is the first thing. It won't be the last. But it'll be there longer than any of us. I've got a spool of that Nebraska fibre in a box somewhere. Kept a few hundred feet. Still good.
If you take one thing from this, it's that the future-proofing isn't in the fibre. It's in the hole. The trench, the conduit, the right of way, the access points. The glass is consumable. The path through the ground is the permanent thing. Everything else gets swapped.
The glass we already have is bored out of its mind. It can handle everything we're asking of it and a lot more. The bottleneck was never the cable. It was always the last fifty metres and the business model.
The business model and the willingness to dig where the digging isn't profitable. That's the real future-proofing question. Not whether the fibre can handle the traffic, but whether we'll actually connect everyone to it.
Something to chew on. This has been My Weird Prompts. Thanks to our producer Hilbert Flumingtop for keeping us in the trench.
If you enjoyed this, leave us a review wherever you listen, or email the show at show at my weird prompts dot com. We read everything.
We'll be back soon.
