BEFORE — A normal land data center

Microsoft in 2018 sealed 864 servers inside a steel tube the size of a shipping container, filled it with nitrogen and dropped it 35 meters down into the Orkney Islands. And then two years later, they pulled it up. And the servers had failed at 1/8th the rate of identical servers on land. And the reason is, servers don't usually die from being old. They die from environmental stress. Temperature swings, loose connections, humidity, corrosion, oxygen, ionized contacts, humans walking through the racks, static and cable damage. The underwater server pod removes every single one of those. Stable temperature because the ocean has huge thermal mass and zero humidity, no oxygen, no people. So the payoff after two years underwater: 1/8th the failure rate. Land control group, failure rate 8x. Land servers. 1/8th Natick underwater pod, failure rate 1x. Same servers, same workload. The reason is, servers don't usually die from being old. They die from environmental stress. Temperature swings, loose connections, humidity, corrosion, oxygen, ionized contacts, humans walking through the racks, static and cable damage. The practical takeaway is pretty simple. If your hardware fails because the environment it sits in is uncontrolled, controlling the environment is the highest leverage move you have. And I can see definitely someone taking up this idea again in the age of AI.