Google paid up to triple the going rate for fiber that routes around the Red Sea

In March, Iranian drones struck Amazon's data centers in the United Arab Emirates and Bahrain, and bank and payment apps in the region stopped working ^[9]. Snowflake reported service disruptions ^[9]. That outage began on land. The Red Sea exposure is a strait less than 32 kilometers wide between Yemen, Djibouti and Eritrea, where the cables enter the sea ^[2].

On a network diagram, capacity on three cables looks like three routes. Paul Brodsky, senior research manager at TeleGeography, sees it differently. "You can have capacity on multiple cables, but if they're all routed through the same tight expanse of water, suddenly your route resilience isn't looking so resilient," he said. "The repair ship issue is very real." ^[5]

Contested water makes repairs slow. Three Red Sea cables were damaged in 2024, most likely by the anchor of a cargo ship disabled by a Houthi missile, and the repairs took months while crews waited for permits, Brodsky said ^[6]. "Depending on the availability of a repair ship, a cable repair might take one to two weeks," Doug Madory, head of internet analysis at Infoblox, told Rest of World. "But the last time there were cable cuts in this body of water, repairs took months because the repair ships needed assurances they would not be fired upon." ^[7]

What Google bought is a pair of fiber strands running along Turkish state pipelines ^[17]. At the price the person familiar with the deal described, the outlay comes to 12 to 18 million euros ^[19]. Long-distance fiber is usually sold as one upfront payment covering 15 years or more of use, plus a separate annual maintenance fee ^[3]. Over a 15-year term, the outlay works out to roughly 0.8 to 1.2 million euros a year before maintenance ^[20]. Google and Meta did not respond to Rest of World's requests for comment ^[16]. Rest of World's reporting does not include any figure on cloud prices or latency, so any claim that this premium will reach customers is a forecast.

For now, most of the new backup capacity stays idle. Madory expects cloud companies to test their land routes so they are ready, with most data still flowing under the sea until a cable breaks ^[11]. "Submarine cable routes are much cheaper, which is why they are so preferred," he said. "I would expect land route alternatives to remain dormant until pressed into service due to a cable cut." ^[12]

For a team running workloads between European and Asian cloud regions, two questions sort the exposure. The first is whether the workload depends on that cross-region path at all. The second is whether it can run for months on a rerouted path, which is how long the 2024 repairs took ^[6]. Needs the path and cannot wait months: this is the case where the provider's failover route matters, and it is fair to ask which land route carries it and whether it has been tested. Needs the path but tolerates a slow backup: the question moves into the contract, to service credits for a long degradation. Stays inside one region: the strait matters less, though the March outage showed that a regional data center carries risks of its own ^[9].