NetworkOcean Launches: Underwater Data Centers

NetworkOcean recently launched!

‍

^{"NetworkOcean builds underwater data centers to cut power usage by up to 30%"}

‍

^{TLDR: NetworkOcean's underwater data centers eliminate water consumption and reduce power usage by up to 30% through efficient cooling, creating more capital-efficient and sustainable AI infrastructure.}‍
_‍^‍‍

_‍✴️ NetworkOcean has 2,048 low priced H100s you can reserve now.

‍

Their 0.5 MW capsule in progress, to be tested underwater in the SF Bay in 1 month.

‍

‍Founded by Sam Mendel & Eric Kim

Sam and Eric met through robotics in 9th grade and worked together building underwater MHD generators throughout high school. Eric has patented a renewable energy device and developed various CV and LLM applications at Cornell, Merge, and Palantir. Sam, to fund his ocean tech ambitions, built various startups and hardware projects, most recently exiting a hiring marketplace for creators. A year ago, he built and deployed a floating web server that hosted our site in a port in the bay.

‍

Problem:

By 2030, we’re projected to see:

• $1T+ spent building data centers
• 1T+ gal/yr consumed by data centers
• 800+ TWh/yr by US data centers

💵 Cost:

Building a data center costs $10-20 million per MW of power capacity. 2/3rd of this cost is land, building, and cooling infrastructure. A GW facility requires a staggering $10-20 billion investment before purchasing any servers or switches.

🔌 Power:

Power capacity is the primary constraint in building new data centers. Real estate with high-power infrastructure and power availability is extremely high in demand.

💧 Water:

Since water usage doesn’t affect net-zero carbon goals, most hyperscalers prioritize energy efficiency by using water-intensive evaporative cooling systems. This approach is facing scrutiny, as evidenced by Google's recent permit issues in Chile.

‍
Solution: 🌊

Underwater data centers.

1. They’re cheaper to build than on-land data centers, significantly reducing costs in land, building, and cooling infrastructure.
2. They’re also cheaper to operate, with up to 30% lower power consumption and no water consumption.

Other benefits:

• Co-location with offshore power
• Speed to deployment
• 1/8th Hardware failure rate
• Low latency to coastal cities (<2ms to all of SF from the Bay)

‍

Have others tried this?

Microsoft, 2016: was the first to experiment with underwater data centers. They built a proof-of-concept capsule that showcased promising efficiency metrics. Despite this, the project remained a research testbed, with limited investment in underwater infrastructure, maintenance operations, and scaled deployments. Microsoft has officially retired the project and is focused on expanding its existing on-land data centers.

China moving ahead: since Microsoft’s Project Natick, Highlander was awarded an $880M contract to build 100 data center capsules around Chinese port islands.

‍

Learn More

‍

^{🌐 Visit networkocean.io to learn more.}

^‍_‍

^{🟢 As of this launch, they have 2,048 H100s available NOW. Reserve 1-2,048 H100s, minimum 1 week reserved, ≤ $2.10 / GPU / hr}

‍

^{🤝 NetworkOcean can operate GPUs cheaper than anyone else, but they will be constrained by capacity. If you know you will have future needs, please reach out to the founders via email soon and they'll work to get you your GPUs!}

^{‍
👣 Follow NetworkOcean on LinkedIn.}

‍