Subsea Cables & Internet Infrastructure

Goldman Sachs, the lead for the public offering, justified the $150 IPO price based on the assumption that SpaceX 2025 revenues of $18.67 billion would grow to $474 billion in 2030, including the AI division, which generated $3 billion in 2025, growing to $322 billion in 2030. Given that most of of the AI division's data center space has been leased to third parties in the last several months, this may be a bit generous. 🙃  The entire Goldman Sachs forecast was grounded in the premise that SpaceX will dominate the AI space. Yet its own employees don't use its trademark Grok product and the AI division's subscription growth is 24% per annum with total 2025 revenues of $3 billion. In contrast, OpenAI's revenues for the same year were $25 billion with a 233% annual growth rate. 

Keppel is a member of the Bifrost Consortium. It is essentially arbitraging the difference between per pair consortium cost and what third parties are willing to pay for them. The arbitrage business model comes back to the beginning of telecommunications competition when voice carriers engaged in various kinds of arbitrage to lure customers away from the incumbents. Note Keppel sells no lit services. You must go to Telin today for lit Bifrost services, at least until some of these fibre pair purchases turn into lit wholesale capacity.

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.