Subsea Cables & Internet Infrastructure

Amazon's core network business is cloud services. These services generate less network traffic over long distances than Google's web search and advertising busiess. Customers download their data, compute, and then retrieve it or store it there. Amazon may back up its customer's data in another data center in the same cloud region, but traffic flows from cloud region to cloud region are probably low. As a result, Amazon has been a follower in subsea cables. Often a silent or minority investor like in the Bifrost where it is getting lit capacity in exchange for providing the US landing station. In some cases, Amazon is not a consortium member, but a fibre pair customer. Amazon bought a fibre pair IRU on the Marea, which connects Virginia Beach to Spain, from Telxius. It was probably their first physical capacity ownership on a subsea cable.  More recently, Amazon announced its first solar project, Fastnet, which connects Ireland, where it owns several large data centers, to M...

 I am heading to Nairobi for ITW Africa. I will also use Nairobi as my base for visiting different African nations. Available to meet with anyone in Nigeria, South Africa, Dubai, India, and Singapore who has mutual interests in subsea cables, procurement, and network capacity sourcing. 

MRC: $19,200. Term: 2 Years. Lisbon/London backhaul route protected. 

1. The submarine line termination equipment, consisting of active electronics and transponders (lasers and optical receivers), was traditionally kept at the cable landing station. This began changing when carriers wanted to own and manage their own dedicated capacity (fibre pair or spectrum slice). Telekom Malaysia was an early example. It took two fibre pairs on the ASE cable as opposed to a fraction of the consortium lit capacity. Later the hyperscalers crystallized this evolution in the open cable model which limits common ownership to the wet segments. Each owner has either fibre pairs or spectrum and makes its own decisions concerning where to place the SLTEs. Because cable landing stations are not designed as large carrier neutral interconnection points, the SLTEs are almost always housed in a cage at a carrier netural data center. It is this flexibility in choice of location and equipment vendor that makes the open cable model so appealing. It also facilitates hyperscalers build...

This is my first trip outside the Schengen Zone in 2.5 years. I intend to stay in Nairobi as my base of operations with trips to Singapore, India, Pakistan, Dubai, Oman, South Africa, and Nigeria this autumn. 

The FEC (forward error correction) encoder takes the bit stream and modifies it so the FEC decoder can detect and correct bit errors due to optical or electronic noise. To accomplish this, FEC adds extra redundant bits to the overhead that the decoder can use to identify and fix bit errors. The redundant bits are a mathematical function of the data payload values. This scheme enables transmission to keep bit errors very low without retransmission of the payload. It is similar in philosophy to the checksum field in IPV4. The difference is that the IPV4 field is strictly limited to detecting overhead errors and the routers respond to checksum failure by dropping the packet and forcing retransmission. In contrast, FEC is designed to detect and fix payload errors without retransmission.  The next interesting step is the bit to symbol conversion. Coherent optics represents bit patterns using combinations of optical properties like phase, amplitude, and polarization (there are two modes)...

And I have the solution to your Lagos metro network nightmare. Competitively priced and protected 100G waves using a unique right of way largely immune from civil and private construction harm.

Facebook is deploying the first deep sea 1 Petabit per second fibre optic cable with two optical cores per fibre strand. A core is the optical transmission path in a fiber strand. Standard fiber has one core. The cable will connect North America to Europe and will land in Virginia Beach, and in Bordeaux, France (where the Amitié cable also comes ashore). So it will effectively link Ashburn Equinix to Paris. It is likely Facebook will use the existing cable landing stations serving Marea and Dunant at Virginia Beach. Similarly, in Bordeaux the 1 petabit cable is likely to use the Amitié CLS. Today the highest capacity communication cable in service across the Atlantic or any ocean for that matter is the half petabit Anjana, another META project. It has 24 fibre pairs operating at 20 Tbps per pair. Shannon's law tells us that a fibre pair operating at 20 Tbps is pretty close to full spectral efficiency.  Hence boosting per pair output is a dead end if the goal is to rai...

Facebook's subsea cable subsidary, Edge Cable Holdings, has petitioned the FCC to land a new cable at the NJIX data center in Wall Township, New Jersey. The cable lands in Europe at an Arelion owned facility in Blaabjerg, Denmark. META uses both NJIX and the Arelion facility for the power feed equipment, but is installing the submarine line termination equipment, which includes the lasers and electronics at a Piscataway carrier neutral data center and its Danish counterpart, an unnamed Esbjerg site. There are two huge data centers in Esbjerg. Facebook currently owns two fiber pairs on the consortium AEC-2 subset network linking the US and Denmark. AEC-2 has exactly the same end points as Aurora. Hyperscaler traffic has grown so rapidly that consortium membership in many cases simply makes no economic sense. Most hyperscalers need all the capacity that a cable provides and hence sole ownership as emerged as the most economic business model. We know very little about A...

Communications Authority of Nigeria reports more than 5,000 fiber cuts in the first six months of 2026. If you are operating a Lagos network and need reliable 100G waves to connect your POPs, I have the solution. My provider offers route protected 100Gs as the default service and most them have few outages each year due to a unique backbone almost completely immune to construction digging cuts. The MRC for a route protected 100G connecting OADC, MDXI or Server House is $5,750 per month.

Google is a good example. The Umoja cable directly links South Africa to Australia. Humboldt links Chile to Australia. There is also a new planned cable directly linking the Indian East Coast to South Africa. No carrier consortium would ever entertain lighting such routes.  The answer is simple. As Google revenues become bigger and bigger, the opportunity cost per millisecond of any network disruption rises. More money is lost per unit of time and hence it makes sense to plough more resources into resiliency. As economists say, marginal benefit exceeds marginal cost. Google isn't building South Africa to India because there is a flood of traffic between the two countries. A key reason to do is greater routing options and hence a more stable network. The other part of the answer is Google's cloud aspirations. Its current global market share is 15% versus Microsoft at 20% and Amazon's 28%. In other words, it is playing catch up. Now Amazon has most of the Fortune...

An optical amplifier being lowered in the sea during a cable deployment. They are designed to work at depths as great as 10,000 meters where pressure equals a thousand and one earth atmospheres. Their design lifespan is 25 years and they rarely fail. Almost all subsea cable damage involves the fibre cable itself as opposed to the amplifiers. Besides immense pressure, the amplifiers must resist sea salt corrosion. To do so their hulls are generally comprised of titanium or high strength beryllium copper alloys. Titanium offers an exceptionally high strength to weight ratio as well as unmatched corrosion resistance at ocean depth pressure. Beryrillium copper dissipates heat extremely well. This is essential as the amplifiers contain many lasers and they generate significant heat.

On the left is a deep sea fibre optic cable. It consists of fibre pairs, ranging from two to 24, placed in a hermetically sealed stainless steel chamber surrounded by closely packed steel wires, then a copper power conductor layer, and finally a polyethylene sheath. The diameter is usually just 17 millimeters or 1.7 centimeters. A standard home garden hose in the US is about 16 mm. Deep sea cables must be thin because the cable's length can be as much as 12,000 kilometers. It would be very difficult to transport and lay a thick cable over long distances. Deep sea cables have always been thin. The engineering challenge is to make it thin and highly reliable.  Note that in shallow waters, usually defined as either 2,000 or 1,000 meters deep and less, the cables may have external armor in the form of steel wires wrapped around the polyethylene sheath. This protects against fishing and anchors. The cable sample on the right is armored. An armored cable typically has one to three sets o...

Routing: Singapore/Tokyo/LA/Secaucus Equinix/Slough Equinix.

 I recommend this book on subsea cable technology. I am ordering it today for myself. https://shop.elsevier.com/books/undersea-fiber-communication-systems/chesnoy/978-0-443-29886-8

As most of you know, Kenya is one of East Africa's two subsea hubs (Djibouti is the other). The Communication Authority of Kenya requires cable operators to report total lit capacity on a quarterly basis. The chart below shows the market is tight. Only Seacom was able via a large upgrade to increase lit capacity. It appears to me the local cables are maxed out. Seacom itself is getting old. It went live in July 2009. Hence it is approaching 20 years. So it is likely unless Ciena whips up more modulation magic that Seacom is at its lifetime max capacity. Note that 2Africa data has not been integrated into the official statistics, but it is important to note that international traffic is largely between Kenya and Europe, not Kenya and South Africa. Since 2Africa remains incomplete in the Red Sea, it may not affect total lit bandwidth that much. We know the local market is tight not just from the lack of growth in lit capacity, but also the high prices that 100G waves com...

I don't see any compelling reason for totally excluding Chinese Layer 1 capacity from your network. Data should be Layer 3 encrypted in any case and most DWDM box manufacturers such as Ciena offer Layer 1 encryption. The SEA-H2X fibre optic subsea cable went live in May. It offers 200 Tbps throughput on several key routes such as Thailand/Singapore, HK/Singapore, and Malaysia/Singapore. Moreover, it is an open cable system. This means each fibre pair owner selects their submarine line termination equipment. So you might even find a non-Chinese carrier on the system who purchased spectrum or fibre pairs. SEA-H2X is a 8 fibre pair cable whose main trunk connects Singapore to two Chinese landings. Hauwei was the system integrator with China Mobile, China Unicom, and Converge as consortium members. Converge is a large Philippine fibre-to-the-home provider with 2 million subscribers.

Asia United Group East (AUG East) cable's main trunk links Tokyo and Singapore with branches to South Korea, Taiwan, Philippines, Brunei, Malaysia, and possibly Indonesia (the consortium has stated the cable will land in Indonesia, but the map below shows Indonesia is off-net). Few project details are available. Fibre pair count is unknown, but it will be at least 16 and likely as much as 24 fibre pairs given the region's incredible Layer 3 traffic growth. Consortium members include Microsoft, Amazon, Singtel, the Japanese long haul carrier ARTERIA, Chunghwa Telecom (Taiwan PTT), DREAMLINE (Korean competitive carrier), GLOBE Telecom, and Unified National Networks (Brunei). Like Candle, it avoids the South China Sea by hugging its perimeter in order to avoid Chinese permitting (China claims the South China Sea as territorial waters and requires Chinese government permits for all subsea cables laid in it) and put distance between itself and the large number of cables packed toget...

Candle is one of the more interesting Intra-Asian cables to go live over the next several years. It is the first 24 fibre pair system to serve this region. Total transmission capacity is 570 Tbps. This makes it one of the highest capacity cables on the planet. Like Apricot, Candle avoids the South China Sea due to Beijing's territorial claims and permitting demands. The main trunk links the inseparable couple of Tokyo and Singapore with branches into the Philippines, Taiwan, Indonesia, and Malaysia. Consortium leaders include META, Softbank, and Telekom Malaysia. This project, like Apricot, exemplifies the geopolitics of American-Chinese rivalry. Hyperscalers avoid China and HK and exchange traffic at neutral Singapore. However, the routing is shorter than Apricot because Candle is deployed at the edge of the South China Sea as opposed to a long detour around Indonesia and up the East Coast of the Philippines. Like Apricot, Candle offers physical diversity to the usual suspects mak...

 The human brain runs on 20 to 30 watts. AI (large language models) model estimation consumes gigawatts. A teenager can learn to drive in a couple hours. In contrast, Tesla AI requires 10 billion miles of actual car driving data. And it still makes elementary driving errors like failing to stop for trains: https://www.bloomberg.com/news/articles/2026-08-22/nvidia-customers-notified-about-ai-related-price-hikes-above-15? So is betting the house on a resource-intensive, but second rate technology? Large language models appear too expensive and too mediocre in performance to be the future of AI.