Subsea Cables & Internet Infrastructure

Respect your cat. He or she is smarter than the best AI models according to the best known AI researcher, Yann LeCun, Chief AI Scientist at Facebook.  Yann LeCun is one of the world's leading AI developers and computer scientists: "The most intelligent AI systems don't have the common sense of a house cat. We are not even close to reproducing what cats do"."We are not yet at the level of reproducing the learning processes in a domestic cat's brain". Click here for the interview .  This is more evidence for the upcoming AI industry meltdown. It will lead to a massive oversupply of data centre space. 

The EU recently gave 6 million Euros to the Northern European Polar Connect Initiative which aims to build a subsea cable connecting Japan to Europe via the Far North. See the map below for routing. NorduNet is a network linking universities and research organizations in the five Nordic countries of Denmark, Finland, Iceland, Sweden, and Norway. It is taking the lead on this project which is looking at two alternatives routes to reach Asia. One route would be via the North Pole and the other via the Northwest Passage. This would give Europe access to Asia without traversing the US or Canada and hence offer better privacy and more secure communication as well much lower latency. Because the proposed paths are highly diverse to the usual suspects of Atlantic cables landing in Canada or the US, they might be attractive for resiliency purposes. One can imagine carriers splitting their traffic between the Polar routes and the more traditional cross-US routes for Pacific/European traffic. C...

 Amazon announced yesterday it is launching its Kuiper constellation service in 2025. The UK will be the first country to go live. Up to now Starlink, which has 4.6 million customers, has faced no competition. But the huge buzz around Starlink is not really warranted. Yes, it is a great technical achievement particularly given that a customer is being handed off from one service satellite to another approximately every 30 minutes. However, what ultimately matters are financial results. Undoubtedly, Starlink is bleeding lots of cash. There is no way one can build a massive network prior to significant sales and avoid it. Satellites cannot be upgraded. So they must be fully loaded from day one which sharply increases the capex. Furthermore, the key metrics determining profitability and net cash flow are unknown. These metrics include customer acquisition costs. The American CLECs mostly went under during the dotcom era because it cost too much to acquire customers. Starlink has also...

Facebook is building a six fibre pair cable directly connecting Taiwan to the the United States with landings at Toucheng, Taiwan, Hermosa Beach, California, and Manchester, California. RFS is 2Q2027. The Hermosa Beach CLS is the well known facility built by RTI Holdings before its bankruptcy. In the submarine cable landing license application, Facebook noted that its motivation was the fact that US-Taiwanese traffic is growing rapidly each year. Due to the 12,000 kilometer length of the cable and the absence of any island landings for power, the design throughput per pair is a relatively low 12.8 Tbps or 76.8 Tbps aggregate initial capacity.  ORCA is an open cable so each fibre pair owner will operate and control its own submarine line terminating equipment with only power being under collective control. Since Facebook is the cable's sole initial owner, the open architecture suggests it will sell capacity on the system to third parties to recoup capital expenditures and share comm...

 The Coming AI Crash: Faking Intelligence Versus Real Thinking Yesterday's mini-crash of AI stocks is not surprising. AI related stocks have been priced to perfection reflecting the view that AI is a revolution on par with the industrial revolution or the transition from horse to combustion engine for transportation. But the crash is just the first symptom of a greater problem, namely that the technology does not live up to the hype. The large language models are impenetrable black boxes due to their nature; they are nonlinear statistical models with huge number of parameters. It is the huge number of parameters that make them power hungry beasts of requiring GPUs to estimate them. For details on the crash itself, read https://thetechcapital.com/tech-stocks-tumble-as-deepseek-triggers-1-trillion-market-crash/.  The term 'training an AI model' means exactly the same thing as estimation of a large nonlinear statistical model like the one below. I am a time series statistica...

***5x 100Gs; CT1/LS1; Equiano The Stable Cable; $21K MRC Per Wave ***1x 100G; Johannesberg Teraco/London; WACS; $29.5K;Protected Back Haul; Photo credit: submarinenetworks.com

1. A point: London Telehouse West. Z point: Johannesburge Teraco Campus. 2. Service: 100G Wavelength. Layer 1 optical circuit. 3. MRC: $29.5K. 4. NRC: $20K. 5. South African Back Haul Protected. 6. Term: 3 Years. 7. Delivery: 4 to 8 weeks. 8. Customer responsible for cross connects. 9. Standard SLA. Map Credit: Submarinenetworks.com.