Subsea Cables & Internet Infrastructure

AI data centers has two challenges: big capex requirements and high operating expenses due to the skyrocketing cost of electricity. Another challenge is low customer switching costs. A customer like OpenAI or Anthropic rarely puts equipment into an AI data center. It simply downloads its software into the data center's bare metal servers, estimates its model, and then uploads the estimated model back to its own servers. Oracle has only one major customer and so it faces huge customer risk. Contracts tend to be soft and often include no penalty cancellation clauses. This business model only works if the vendor has access to cheap electricity like hydropower in the Nordics and very low customers. Solar and wind sell rates do not generally make the grade because fully loaded they are in fact quite high. In the case of Oracle, it went from a typical debt free software company to a 2.4 debt to revenue ratio. This includes bonds, long term lease liabilities, and short term obligations. 

Not only do I offer Equiano 100Gs below $20K with you contracting directly with the network operator, but also available is Layer 2 50G protected MPLS circuits on Equiano for $13,500 MRC on two year terms. This allows you to migrate from granular Layer 2 to Layer 1 as your business scales. African ISPs need to seize the day before capacity tightens and the window of opportunity closes due to the lack of any new cables for 3 to 5 years.

Singapore currently has 26 operating fiber optic cables. The government is aiming to double that figure over the next ten years. It is also wants to increase residential broadband to 10G, which seems frankly far fetched and perhaps overkill. The biggest drawbacks to landing in Singapore is the limited number of cable landing stations and the congested sea harbor. I would tell the government that it needs more cable landing stations and that they must be physically diverse to the existing facilities. More Malaysian/Singapore terrestrial crossings for Trans-Continental cables, but that conflicts with the sovereignty angle dominating policy making today. https://www.edb.gov.sg/en/news-and-insights/singapore-to-double-capacity-for-subsea-cable-landings-to-enhance-digital-connectivity

1. Fully diverse and fully buried dark fibre pairs or individual strands available for lease or IRU between IColo1, IColo2, Telehouse, Nyali Exchange, Seacom CLS, and 2Africa Shanzu CLS. Unlike other providers, the fiber connects to the data centers via buried conduits. Not power poles. End to end burial. 2. Sixty Euros per kilometer on three year lease deals. 3. New ultra-loss fiber manufactured in 2024. 4. Repair SLA based on 4 hours mean time to repair.

 At the 2026 Submarine Networks World in Singapore Ciena pitched the MOFN business model, the managed optical fiber network. This is a network dedicated to a single customer like a hyperscaler, large enterprise, government, and the like. It involves dedicated fibre pairs and equipment. Very often the customer buys the equipment and ships it to the network operator. It is quite common in Africa where network operators provide such services to the American Tech Giants. But carriers have often found that hyperscaler forecasts of capacity needs are often wildly off. The result is poor MOFN economics and low rates of return.

Ciena seems correct when it asserts massive growth from here to 2035. Heck, that is a 35 year Internet trend. But claiming that traffic will be 85% AI sounds like a very bad forecast. Right now the world is spending $2 trillion each year on AI infrastructure versus $100 billion in AI service revenues. A pretty big gap. We have experienced cycles of excess technology enthusiasm over the last 150 years and bullish forecasts have almost always proved wrong.