Most of Latin America’s internet still rides on cables laid when the region was on dial-up.
I’m not exaggerating. SAm-1, SAC, PAC, GlobeNet, Americas-II, Maya-1, ARCOS-1. Line up their ready-for-service dates and they all fall between late 1999 and 2001. Subsea cables are engineered for a minimum design life of 25 years. Do the math. It lands on now.
Some have already gone dark. Columbus-III in 2020. Atlantis-2, the link from Argentina and Brazil to West Africa and Portugal, in January 2022. Maya-1 was scheduled to retire last October, with a shorter ring taking over the Florida–Honduras–Cayman corridor. Américas-I, a 1994 system, is being phased out. TeleGeography has said it plainly for two years: the 2000–2001 generation is “approaching the end of its economic lifetime,” and most of it will be retired and replaced before the decade is out.
This is not a crisis. It’s a scheduled event. We’ve known the date since the day the cables went in the water. The only question is whether Latin America uses the moment or wastes it.
Figure 1. The 2000 generation, from ready-for-service to today. Hollow markers show the 25-year design-life point.
The old cables lasted this long for one reason. That reason is running out.
Here’s a number that should reset how you think about 25 years: for repeatered cables retired between 2010 and 2022, the average lifespan was 17 years. Cables rarely die. They get outcompeted. A newer system delivers so much more capacity per dollar that the old one stops earning its keep.
Latin America’s 2000 generation beat that average because of terminal equipment, not because of the glass. SAm-1 was designed for 1.92 Tbps across four fiber pairs. Upgrade after upgrade to the electronics at each end pushed it past 40 Tbps. Twenty-fold, on fiber nobody has touched since 2001. That’s remarkable engineering, and it bought the region a decade.
But the glass is the ceiling. Four fiber pairs from 2001 cannot be talked into behaving like sixteen or twenty-four from 2025. And demand does not care about our ceiling. Used international bandwidth in Latin America and the Caribbean passed 400 Tbps in 2025, more than triple 2021. TeleGeography’s forecast: over 1.3 petabits per second by 2030, over 3.5 by 2035. Content and cloud providers went from a fifth of that demand in 2021 to almost 40% in 2025. On the US–Latin America corridor they now carry more than the carriers do.
Now look at what the new generation delivers. Monet, 64 Tbps. Seabras-1 and Curie, 72 each. EllaLink, 100. Mistral, 132. Firmina, the newest US–Brazil–Uruguay–Argentina trunk, 240 Tbps on sixteen pairs. TAM-1, above 650. And this week someone announced a one-petabit-per-second transatlantic cable built on multi-core fiber. That’s where the next design cycle goes.
One modern cable now outweighs the entire 2000 generation. All of it. After every upgrade it ever got.
Figure 2. Design capacity by cable. SAm-1 is shown both as designed and after every terminal upgrade.
So the replacement will happen, and it will happen with fiber. That part isn’t interesting. What’s interesting is what else changes.
The question I get at every conference
“Can’t Starlink just do this with lasers?”
I get asked this in Miami, in São Paulo, in Amsterdam. It deserves a real answer, not a dismissive one, because the technology is impressive and I’ve spent time with the people building it. Starlink has more than 9,000 optical terminals in orbit. They form over a quarter of a million laser links a day, each up to 200 Gbps, over distances as long as 5,400 kilometers. In 2024 the mesh moved about 42 petabytes a day. 400 Gbps links are on the roadmap. The V3 satellites now flying on Starship are rated at over 1 Tbps of downlink each.
Now put those numbers next to a cable.
Forty-two petabytes a day sounds enormous. Spread it over 24 hours and it’s roughly 4 Tbps. One fiber pair on Firmina is designed for 15 Tbps. So the entire Starlink laser mesh, at its last officially published throughput, moved less traffic than a single fiber pair on a single cable. Firmina has sixteen of them. One 200 Gbps laser link is about 1% of one fiber pair.
Figure 3. Throughput on a logarithmic scale. Each step to the right is ten times the last.
Zoom out to the whole constellation and the story holds. SpaceX’s own network update put total Starlink capacity near 450 Tbps in July 2025, growing by more than 5 Tbps a week, and coverage of its 2025 progress report puts it above 600 Tbps by year-end. Real, and growing fast. But that’s capacity to end users, spread across 160 countries and 12 million subscribers. It is not trunk capacity between two points. Even if V3 delivers the promised hundred-fold jump, the whole planet shares it.
A cable is different in kind. It’s a private, dedicated pipe between two landing stations, and it gets a capacity upgrade roughly for free every few years when someone swaps the terminal cards. Satellites scale by launching mass. Fiber scales by replacing electronics on land. Over a 25-year horizon, that difference decides everything.
My estimate: lasers take 1 to 2% of Latin America’s international bits off the seabed over the next decade. Treat that as an inference, not a fact. But I’ll defend the direction against anyone.
Lasers don’t replace the capacity. They replace the failure.
That 1 to 2% badly undersells what LEO changes, because bits are not the only thing a subsea network carries. It also carries risk. And there are three things fiber does badly.
First, distance. Light travels about 47% faster in vacuum than in glass. Past roughly 3,000 kilometers, a LEO laser path beats any fiber path on the planet. Santiago to Miami is about 6,600 km. Buenos Aires to Miami, over 7,000. For the thin slice of traffic where milliseconds are the product, the sky is the faster road.
Figure 4. Simplified propagation model calibrated to Handley (2018). Bogotá–Miami stays on fiber; the Southern Cone routes cross over.
Second, the edge of the map. The Galápagos now has a 10 Gbps community gateway. When Tonga lost its only cable in January 2022, service took 38 days to come back. The stopgap was Starlink terminals flown in and a gateway hastily stood up in Fiji, and even Musk called the result “a little patchy.” For an island, a mining camp, or a town in Patagonia, the sky is not the backup. It’s the cable.
Third, concentration. Sixteen cables land on one beach in Fortaleza. That beach carries roughly 90% of Brazil’s international traffic. The ICPC counts 150 to 200 cable faults a year worldwide, most of them anchors and fishing gear. Give a network one small, independent, non-terrestrial path for control-plane traffic, out-of-band management, and emergency customer capacity, and you’ve built a materially more resilient network, even if that path carries a rounding error on a normal day.
关于 Starlink，正确的思路是这样的。它不能替代 SAm-1。它替代的是 SAm-1 及其同一海岸线上的三个邻居同时发生故障的时刻。
用海底替代海底。设计网络时，要假设存在天空级的连接能力。
这是 EdgeUno 的思考方式，也是我认为该地区在退役周期中应采取的措施。有五点。
将 2000 代系统视为桥梁，而非 Backbone。 如果您的产品路线图假设这些电缆在 2032 年仍然可用，您押注的不是物理学或经济学，而是许可证的续期。应购买新系统。在老系统仍能提供多样性时保留它们，并按照其实际价值进行定价。
解决许可审批问题。 Firmina 在巴西因许可审批而损失了大约两年时间。目前有提案将此整合到一个窗口，并将审批时间缩短至六个月。您应该支持这些提案。如果每根电缆在其使用寿命的四分之一都花在等待批文上，那么到 2030 年再将该地区的带宽增加三倍是不可能的。
不要将所有连接都登陆到同一两个海岸。 福塔莱萨和南佛罗里达是极好的枢纽。但它们也是单点故障。通往多米尼加共和国、巴拿马、哥伦比亚和太平洋海岸的新系统，是该地区构建网状网络而非两个漏斗的机遇。互联互通必须跟上。如果哥伦比亚波哥大到圣保罗的流量仍然需要绕道经由迈阿密，那么这种多样性仅存在于纸面上。
从第一天起就构建混合网络。 Fiber for the bits, LEO for the lifeline. Every EdgeUno market has terrestrial paths to multiple cables and, increasingly, a non-terrestrial path for the day the terrestrial ones fail together. That isn’t a slogan. It’s a design constraint we apply to every new point of presence we build.
Figure 5. Fiber for the bits, LEO for the lifeline: two terrestrial paths from different coasts, one non-terrestrial path for the day they fail together.
And don’t wait for the press release. Every cable in Figure 1 was economically obsolete before its owner said so in public. The operators who win the next decade in Latin America will be the ones who read the ready-for-service dates, not the announcements.
The class of 2000 did its job. It carried this region from dial-up to streaming on fiber that was never designed for either. What replaces it should be built for a world where the seabed is faster and the sky is safer, and where a network that uses both is the only kind worth running.
We’re building that network.
If you’re planning capacity in Latin America for the decade ahead, let’s talk about the routes, interconnection, and resilience your network will need, and how EdgeUno’s AS7195 network across 17 markets can support those plans.
