Naver Just Bought Into a Data Center That Floats

Naver announced on August 13, 2026 that it has invested in Panthalassa, a US startup building offshore AI data centers powered by ocean waves.

Naver is not a speculative investor. It is South Korea's dominant internet company, it operates its own large-scale data center infrastructure, and it has spent the last two years trying to secure AI compute in a country with expensive land, constrained power, and a national strategy that assumes it will need far more of both.

Which is the whole reason this is worth taking seriously rather than filing under "interesting concept."

What the thing actually is

Panthalassa was founded in Oregon in 2016 and builds floating platforms it calls Nodes. Each one integrates three functions that are normally three separate industrial problems:

Power generation. The platform generates electricity from wave motion continuously — no separate power plant, no fuel supply chain, no grid interconnect.

Compute. AI workloads run on the platform itself.

Cooling. Seawater does the cooling, which removes the single largest non-compute energy draw in a conventional data center.

Results go back to shore over low-Earth-orbit satellite networks such as Starlink, which means no subsea cable and no landfall permitting.

The company has tested prototypes at sea — Ocean-1 and Ocean-2 — and plans to deploy its Ocean-3 pilot later this year, with commercial deployment targeted for 2027.

The constraint it is designed around

Every serious AI data center project in 2026 collides with the same wall, and it is not chips. It is the interconnect queue.

A large cluster needs a grid connection sized in the hundreds of megawatts. In the markets where anyone wants to build, those connections are spoken for years in advance. You can buy GPUs in months. You cannot buy a utility interconnect that somebody else reserved in 2023.

Panthalassa's proposition is to route around the queue entirely by not needing a grid connection at all. If the platform makes its own power and cools itself with the medium it floats in, the two slowest permitting processes in data center construction — power and water — both disappear.

There is a second constraint it addresses that matters specifically to Korea. Land near population centers is expensive and contested, and public opposition to large data center siting has become a real political factor. Ocean is neither expensive nor contested in the same way.

Why the physics is plausible

Wave energy has a long history of failed commercialization, and skepticism is earned. What is different here is the load.

Most wave energy ventures failed on the same problem: converting intermittent mechanical motion into grid-quality electricity, then getting that electricity to shore through a cable and onto a grid that requires stable frequency and voltage. The conversion losses and the transmission infrastructure ate the economics.

A colocated data center removes both. The load sits directly on top of the generation, so there is no transmission. And compute is unusually tolerant of variable power in a way that a grid is not — training and batch inference workloads can throttle up and down with available power without anything breaking. You are not selling electrons to a utility that demands 60Hz. You are feeding them to a machine that will accept whatever you give it and do proportionally more or less work.

That reframing is the actual insight. Wave energy was a bad electricity business. It may be a workable compute business.

Seawater cooling is the less speculative half. Cooling typically accounts for a substantial share of a data center's total energy draw, and an effectively infinite ambient heat sink at stable temperature is the reason submerged and coastal cooling concepts keep resurfacing.

The list of people already in

Panthalassa raised $140 million in May 2026, led by Palantir co-founder Peter Thiel, with Marc Benioff and Hanwha Asset Management participating. It has since been reported to be raising a further $225 million at a valuation near $2 billion.

Hanwha's presence in the May round is the detail that predicts Naver's arrival. Hanwha is a Korean conglomerate with deep shipbuilding and offshore engineering capability — precisely the industrial base required to actually build and maintain floating industrial platforms at scale. Korea does not have cheap land or spare grid capacity. It does have some of the best shipyards on earth.

That is a genuine national comparative advantage applied to a genuine national constraint, and it explains why this particular idea is finding Korean capital rather than, say, Texan capital.

The hard parts nobody has solved yet

The unsolved problems are real and none of them are about wave physics.

Maintenance. A failed GPU in Virginia gets swapped by a technician in an hour. A failed GPU on a platform in open ocean requires a vessel, weather, and a crew. Whatever the failure-rate math is, it has to be solved at the design level rather than the operations level.

Survivability. Marine environments destroy equipment. Salt, humidity, and constant motion are hostile to electronics in ways a climate-controlled hall is not. Ocean-3 is the test of whether the enclosure engineering holds.

Bandwidth. Satellite backhaul is fine for sending back the results of a training run. It is not fine for anything latency-sensitive or bandwidth-heavy. This constrains offshore compute to a specific workload class — batch training and large-scale asynchronous inference — and rules out serving interactive traffic.

Regulation. Industrial facilities in international or territorial waters raise jurisdictional questions that do not have settled answers.

The read

The reason to pay attention is not that floating data centers are obviously going to work. It is that a serious operator with real compute needs just put money behind them.

Naver knows what it costs to secure power and land in Korea, because it has done it. When a company with that knowledge invests in the alternative, it is making a statement about how bad the conventional path has become.

Ocean-3 deploys later this year. That pilot is the first honest data point on whether the ocean becomes a place we put compute, or stays a place we tried to.