Subsea Cables & Internet Infrastructure

Stocks of publicly traded companies including Coreweave, META, Microsoft, and Oracle have lost 4% to 39% of their stock market value over the last year. Coreweave's core business is building data centers stocked with GPUs to rent AI service providers for estimation and inference. Obviously, this makes the standard data center look downright capital light. Racks must be populated with servers and GPUs and capable of handling 120Kw power loads. So power infrastructure and backup alone cost many multiples more on a per square meter basis than the standard telecom hotel. Moreover, these data centers must be bigger because a large language model might have trillions of parameters to be estimated. Coreweave's massive debt loads have led to a 37% decline in its market value over the last twelve months.  Secondly, customer switching costs are very low in the AI data center market. AI service providers are software companies. They download their data into the bare metal ...

NTT and Mitsui Leasing have teamed up together with fibre optic cable manufacturer Sumitomo Corporation to fund a 16 fibre pair subsea cable costing $500 million. Design capacity is 320 Tbps. This is the second project on which NTT Data and Mitsui Leasing have cooperated. The high capacity 20 fibre pair Juno cable was the first project. It is unusual to see a large incumbent player like NTT undertaking a subsea cable project with a non-telecom company as co-owner. But the advantages are clear. Mitsui will contribute cash, but not play a major role in design, vendor selection or wholesale commercials. So it not only reduces NTT's risk by sharing funding requirements, but it gives NTT a free hand in decision making. One lesson that has become perfectly clear is that large carrier consortiums increase the likelihood of deployment delays and lengthen the entire planning cycle because decisions require consensus. Moreover, the consensus requirement makes it difficult for new approaches...

MRC: $32.5K. A point: Most Mumbai DCs. Z point: SG3.  Cable: IAX.

The well respected wide area network carrier has lit a new, shorter path connecting Milano and Paris data centers. While I believe you should never limit yourself to one vendor, EUNetworks is a very safe choice because it does the fundamentals well and is commercially flexible. The new route spans 1057 kilometers. EU did not provide a latency estimate, but it takes roughly 5.1 microseconds to traverse a kilometer. So multiply by 1057*2 and the RTD is approximately 10 ms. One diversity edge is that the route avoids Lyon. I recommend network planners consider to further diversity their network portfolio and reduce risk as a financail analysts might say. 

A consortium consisting of TATA Communications, Lightstorm, Microsoft, and Singtel just announced a new 16 fibre pair subsea network tailored to serve the AI data center markets in Hyderabad, Singapore, and Kuala Lumpur. It includes two Indian landings, one in Southern Chennai and another at Machilipatnum, the latter being the shortest path to Hyderabad. Both Hyderabad and Lumpur host large numbers of data centers equipped with GPUs for rent to estimate AI large language models. Conversely, Singapore is a distribution point for estimated AI models. What is sometimes called AI inference. Both space and power in the city state are too limited and expensive for AI model estimation. Besides the Singapore and Indian landings, there will likely be a cable branch landing in the Malaysian province of Selangor, chosen because it is the shortest way to reach Kuala Lumpur. Public information on the new system is sparse, but it is also likely that cable will include be extended to Hyderabad and Ch...

Bypass the Red Sea and Persian Gulf quagmires as well as China and Chinese equipment. No Russian paper. Save at least 40 to 60 milliseconds over subsea routes via the Indian ocean.

SpaceX is the most over-valued stock in the Solar System with a stock valuation 100x current sales The smart money that invested in SpaceX as a private company knows it is wildly over-valued and will sell their stock when the IPO lockup periods expire this autumn. The float, the amount of shares registered for public trading, will increase 20% to 30%. That is a huge increase in supply. Moreover, the bond market, the best measure of risk, is pricing SpaceX bonds as junk. Unlike equity, bond investors are level headed and don't give a damn about technology hype. Only financials matter in the end. For a recent forecast of the impending implosion: https://finance.biggo.com/news/0e4c9f77-64fb-4bba-8193-43f26b36b74f. 

Another 29 Amazon satellites lifted into orbit yesterday. Total Amazon transmission capacity is now 396 Tbps. This is enough to begin initial service beginning of third quarter. The Musk Tribe has consistently told me that Amazon needs thousands of LEOs to begin service. Given Amazon's inked contracts with Delta Airlines, Apple, Vodafone, AT&T, and Hitachi construction company, it appears the Tribe was wrong. 😄It should be noted that Amazon's satellites are much higher capacity than Starlink's first generation, and Starlink began operations with 700 lower capacity LEOs. https://www.euronews.com/next/2026/07/03/amazon-set-to-launch-leo-its-rival-to-spacexs-starlink-internet-later-this-year

Fastnet is a 16 fibre pair submarine network landing in Maryland and in Southern Ireland near Castlefreke. Design capacity is 320 Tbps. This is the first cable in perhaps ten years to land in Ireland and the first landing ever in Maryland. The Irish landing makes perfect sense; Amazon has roughly 350,000 m2 of data center space in the country. It hosts an important cloud region (eu-west-1) there in large part due to a very low corporate tax rate. In addition, it just received clearance to build three data centers near Dublin totaling usable space of 42,585 m2. The choice of Maryland again shows that the hyperscalers value resiliency versus latency more than do the wholesale carriers and telecom incumbents. The latter focus on low latency routes for both Internet backbone and financial trading firms. The main US East Coast telecom hubs are Secaucus Equinix and Ashburn Equinix. If latency was Amazon's top priority, Maryland would not have been the landing point. The sa...

A turbidity current is the likely culprit for the WACS trunk outage. The same holds true for the 2Africa Ivory Coast branch failure. The Taiwan earthquake of December 2006 is instructive in this regard. The 22 knocked out cables failed in sequence over the course of several hours. So the sheer force of the earthquake was not responsible. Instead, the seismic event caused sediment to begin moving down the undersea slope of Taiwan's continental shelf. This was not a gentle slope, but rather the steep sides of the Kaoping subsea canyon, which is 4 kilometers deep. As the chart shows, cables went dark in sequence radiating from the epicenter outward as this undersea tidal wave traveled down the sides of the subsea canyon. The turbidity current traveled at speeds ranging from 3.7 meters per second to 5.7 meters (roughly 20 kilometers per hour). The sequence of events suggests there were at least 2 and probably turbidity currents involved.  It is probably not a coincidence ...

Heavy rains in Côte d'Ivoire likely caused rivers to dump large amounts of sediment at high speed into the Atlantic. These undersea mud Tsunamis accelerate down the deep gradient of the continental shelf off Abidjan and destroy anything in their path. In particular, they displace the sea floor up to several meters. It is likely that a turbidity current effectively disinterred the WACS trunk and severed it. The main value of WACS is moving traffic between Africa and Europe. So this outage imposes severe hardship on African ISPs. Outages off Côte d'Ivoire's shore are common and have disrupted Internet service in the country many times. For example, in 2024 a debris slide in the subterranean canyon off Abidjan took out four cables, including WACS. Côte d'Ivoire really needs high capacity fibre optic links into neighboring countries to better weather these network crises.

ISPs in Ghana are reporting their 2Africa waves are up and running. So 2Africa's main trunk appears untouched. Hence the Cote d'Ivoire outage must be due to a branching segment fault. It is thought a turbidity wave caused the damage. This is a powerful surge of water, debris, and mud that can wash away the sediment covering buried cables and snap them like toothpicks. I call it an underwater Tsunami. It is triggered by an undersea avalanche due to an earthquake or a flooding river like the Congo pouring into the Atlantic. For example, the 2006 Taiwanese earthquake caused a turbidity current that tore apart 22 cables off the country's Southeast coast. This current traveled several hundred kilometers and reached speeds as high as 72 kilometers an hour. It is not clear what caused the turbidity surge off Abidjan. What we do know is that there is a large subterranean cavern, Le Trou Sans Fond, at Abidjan's doorstep. It was likely involved.  The last few days Cote d'Ivoi...

This was initially diagnosed as a shunt fault, but the cable appears to be effectively severed with the approximate location of 9 kilometers from repeater 4 towards repeater 3. To the best of my knowledge this is 2Africa's first wet segment outage on the West African coast. No repair date is available at this time.

Abidjan clients on WACS report an outage began 4 AM local time today. Outage began with a severe drop of 385 volts. Investigation shows the fault lies in segment S3C between the first and second repeater on the Ivory Coast branching unit. So the outage should only affect Cote d'Ivoire. A ship has been dispatched to fix it. Two years ago four cables were severed off Abidjan due to a debris slide in the subterranean cavern known as Le Trou Sans Fond. Also a 2Africa outage started 1:30 PM Abidjan time. More information will be provided as it becomes available. First map below is 2Africa. Second is WACS. 

Reliable sources indicate that Google's 16 fiber pair Nuvem (Portuguese for 'cloud') cable has landed. There is typically a lag of a year between landings and RFS. So expect the system to be fully activated next Spring. I assume, without any insider knowledge, that either EXA or Telxius will acquire spectrum or a fiber pair on Nuvem. The cable's design capacity is 384 Tbps. Nuvem lands at Myrtle Beach, South Carolina, also home to Anjana and Firmina. For Google, which owns the South Atlantic Firmina cable, this solves a big problem. It can move traffic between South America and Europe via the Myrtle Beach CLS. It also diversifies the routing of Google's Atlantic traffic and reduces latency for both the Southern US and Southern Europe. Interestingly enough, Google used a subsidiary as the landing partner at Myrtle Beach, and at Sines, Portugal. The hyperscalers are becoming increasingly vertically integrated. In the past it was customary to outsource landings and cab...

Mombasa Icolo 1/Djibouti CLS; $23,500 MRC; 3 Years. Protected backhaul to Djibiouti data center only $1,050 extra. 🙂  Mombasa Icolo 2/JB Terraco; $25,400 MRC; 3 Years. Backhaul protected.

The National Institute of IT and Communication Technology (NICT) in Japan has been diligently working for many years on dramatically increasing fibre optic bandwidth by using a wider spectrum range than the traditional C and L bands. The long haul fibre optic networks have relied exclusively on the C and L bands because they offer the lowest optical attenuation for silicon-based fibre optics. Light in these frequency bands fades and loses strength relatively slowly as it passes through fibre optic glass. But as the chart below shows, the C and L bands comprise just a fraction of the available spectrum that lasers can use.  In this trial the NICT achieved 450 Tbps on a repeatered London metro fibre pair connecting Telehouse London to the University of London. Because existing optical amplifiers are designed to work only in the C and L bands, NICT has spent years developing amplifiers optimized for the O, E, S, and U bands. These bands required that the fibre optic strands in the am...

RTD: 135 milliseconds. Term: 1 Year. Only one wave available.

The chart shows US business paid AI subscription growth. Musk's AI company is in last place. He shoved xAI into SpaceX in order to siphon off SpaceX cash flow and IPO proceeds to finance his struggling venture. On its own, no one would lend to xAI given its third tier position in the AI service and its $1 billion per month cash burn. Now Anthropic is leasing a huge chunk of xAI data center capacity to estimate its models. Supposedly worth about $1.25 billion per month. I say supposedly because contracts can vary from rock solid to as soft as jello. Indeed, some industry contracts are really just letters of intent, non-binding expressions of customer interest. In this case both sides can cancel the agreement with only 90 days notice. Again, the Musk cheerleaders are prematurely celebrating. What this deal implies is that there's very little demand for xAI's models. So just as X pivoted from social media platform to AI, it is now pivoting to the AI data center b...

The standard description is that 2Africa is a 16 fibre pair cable with 180 Tbps throughput. It comes right off the 2Africa website. That public figure is a bit misleading. Only the Mediterranean and Red Sea segments have 16 pairs. The all-important West Coast segment has 8. Now these fibre pairs should individually be capable of 20 Tbps or slightly more. So West Coast transmission capacity probably does total 180 Tbps. Indeed, the Mediterranean segment could easily exceed 320 Tbps. And the same for the yet unfinished Red Sea segment. Design Capacity By Segment A. Mediterranean Sea: 16. B. Red Sea: 16. Not finished due to hostilities. C. Red/East Africa: 7. D. Red Sea/India: 9. E. West Coast: 8.