Huge DarknessSam T. Oates

Making the Pacific think

Why Panthalassa might finally get ocean power working

Sunlit Waves (1903), William Trost Richards

Alternatives to space compute that get around the terrestrial permitting problem look very attractive right now. Definitionally, pretty much anywhere on Earth is going to be easier to build in than orbit; the problem is logistics. A major part of the argument for space compute is that it’s easier (modulo Starship being cheap) to send things straight to orbit than, say, ship and construct a data centre in some random isolated location. Doing this without local amenities or port facilities is expensive.

For a viable alternative, you need someone actually doing it today. It’s not enough for a concept to be viable on paper; if you need to do five years of R&D before you start scaling, you’re not relevant to this discussion.

There is one under-discussed option here: Ocean compute. Panthalassa is the latest in a long line of failed wave-energy startups. But this time is different!

No, seriously. They have a pool of credible engineering talent from places like SpaceX. They’ve raised hundreds of millions in funding. They’ve proven prototypes at scale in deployment environments. They avoid the major failure modes of previous wave energy attempts like near shore permitting, cable or mooring connections, and grid integration. They also avoid the equivalents for past ocean compute failures - shore tethering, submersed maintenance, and power supply.

The product is basically a giant buoy; waves drive water into an upper chamber, which then drains out through a turbine, producing power. The whole thing floats free in the open ocean. They are aiming for 1MW units with one nine of uptime, supplying power for 2 cents a kWh; they’ve demonstrated 50kW peak power from a prototype.

The challenges facing them are merely extremely difficult rather than virtually impossible. They have to scale production of huge offshore structures and marine ops. They need to maintain cost discipline to avoid being eaten alive by alternatives; one downside is that there is absolutely no differentiation for a token generated in the ocean versus one generated in, for instance, orbit.

You can bet Panthalassa will be interested in compute because it’s their only business model. Unmoored open-ocean ops makes the engineering viable, but for the economics to work you need to be consuming the power at the source. They have two proposals here: compute and… uh… green hydrogen. The latter has been quietly dropped in recent comms.

Data connection is easy, assuming you stick to inference - Starlink. Cooling is free. Ocean swell is consistent and predictable, though a capacity factor of around 40% for waves seems rough considering GPU capex; it would be worth looking at the battery technoeconomics here. It does, however, completely sidestep grid interconnects and normal permitting.

There is also a highly developed offshore industry used to producing the simple structures demanded by Panthalassa. Yes, the offshore environment is horrendous, but it’s not impossible to deal with on a static structure like the Panthalassa units. Space is worse! They don’t need specialist installation vessels, port upgrades, HV cable, or seabed works. And if you can solve the manufacturing then there aren’t many caps on scaling: global wave power is theoretically in the terawatt range. Compute-wise, they can absorb demand from other scalers rather than being restricted to the SpaceX ecosystem.

In permitting terms, they’d likely be treated under maritime law and have to deal with US environmental laws, but this is a comparatively light burden.

The default position for this sort of thing should always be skepticism, but hopefully it’s clear why I think it should be much easier to scale this than space compute - even after correcting for the Elon factor.

My default expectation, however, remains the same. Normal data centres will find a way - and continue to outcompete everything else.