Hacker News Reader

Comment by peri-cl | Hacker News Reader

I'm very curious what happens now to the several gigawatts of fuel cells, that this project supposedly purchased to supply power.

> "Oracle signed a deal with solid oxide fuel cell (SOFC) developer Bloom last year for 1.8GW of power, expanding it to 2.8GW of capacity in April."

https://www.datacenterdynamics.com/en/news/new-mexico-regula...

https://en.wikipedia.org/wiki/Bloom_Energy

I was surprised to read we're already building fuel cell power plants on this scale (this is a first of its kind project—by far this startup's largest customer). I didn't realize the tech was that mature. Now I'm left wondering if they actually ever intended to, or whether this was some kind of circular financial-engineering vaporware.

View on HN

The tech has always been mature enough but simply too expensive. But since combustion engines are sold out for the next years, data centers are rushing to get anything they can. So the extra millions for power generation will be a rounding error

Electricity is about 3% of their total operating cost, so, yes.

Electricity is only 3% of opex for a data centre?

That seems surprising - source? What's the other 97%? Does this include ops staff?

Without having done any research at this point I would assume it's mostly the amortized cost of hardware - GPUs, memory etc

That's capex not opex!

The stuff inside the buildings (servers, switches, storage, cooling, etc) basically the stuff using the electricity and salaries/sub contractors (in addition to some techs handling the servers think security, cleaning, storage management, receiving shipments of new hardware, etc). Some taxes (depends a lot where)

I saw a breakdown (don't have on hand sorry) where server depreciation was by far the majority. Power was the largest cash opex, probably matching your intuition.

The cost of electricity is (almost) negligible compared to the cost of having GPUs sitting idle. Electricity from fuel cells can be twice as expensive, but even then, it represents only about an extra ~15% of the amortization cost associated with keeping the GPU hardware idle.

That's not an operating cost, it's capital cost.

Yes, but given the 3-5 year lifespan of the server infra capex, the separation becomes less meaningful. Better to count the entire TCO or discounted cashflow accounting for the rapid asset depreciation.

There is no way it's 3% if they're getting electricity from the grid, even if they get their cooling from somewhere else (eg lake/river water or outside air in winter). Median DCs are anywhere from 25-40% (depending on the jurisdiction's electricity costs and how much cooling they need). Hypercaler datacenters tend to be even more dense than the median traditional ones.

Energy is the largest OPEX by far (70-80%)... taxes second.

https://epoch.ai/data-insights/ai-datacenter-cost-breakdown

Anyone who puts a local model on a GPU understands how much heat they produce.

Its amusing watching cloud AI users struggle to grasp even direct impacts of their use.

That's if you don't count hardware amortization - which is 80% of the cost...

That's not OPEX, that's CAPEX and tangible assets are depreicated not amortized.

I would think someone would have run those numbers first.

No time to think, go go go.

You would think, right?

If the numbers dont add up it's fucked anyways who cares just send it

Also worst comes to worst government will have to bail them out to not cause massive unrest

someone did, determined it would make them look bad, and pawned the assignment off to a junior executive to take the blame.

if it works then they delegated successfully and the project is winning and if not then they blame the neophyte VP and walk away with another quarterly bonus anyway.

As far as I can tell Bloom has a working product, but I also couldn't find any examples of Bloom installations that are within two orders of magnitude of the New Mexico data center.

I think this was motivated by a mixture of greenwashing and financial shenanigans. That much gas turbine capacity has a leadtime of several years now. Nobody knows if Bloom can scale up to match the requirements. I'd bet against it just because the consumables and operations of the Bloom fuel cells is pretty complex and pushing the limits in terms of heat and other parameters.

I thought fuel cells were an energy storage system, not an energy generation system. Surely it’s cheaper to just connect the data center to the generation directly (either by upgrading the grid to handle it or building generators nearby).

Fuel cells are an electricity generation method. They oxidize fuels (methane, hydrogen) at low temperature, electrochemically, without combustion—going directly from fuel to electrons.

This would have been simply a natural gas power plant, of a different type.

I'm really wondering about the air filtering needed for this to work. Gas turbines are already quite sensitive to bigger particles in the air but for the fuel cell membranes, this could be a much harder problem.

I bet you could spin the natural gas in a centrifuge to eliminate larger particles.

I would guess the natural gas is quite purified by that point, but for the fuel cell to work you need to combine it with air (or to be precise, the oxygen in the air) and a lot of it. So for the fuel cell membranes not to clog with dust and gunk they will have to clean in in some way to prevent issues, ideally without to much extra-energy use or expense per unit of air filtered.

How about a HEPA or ULPA filter? These are very mature technology, work extremely well, and are not particularly expensive.

Sure, that would work - but with the massive ammounts of air needed I would assume those would be quickly clogged and unusable.

That was not my understanding. Unless this installation was supposed to be producing fuel on site somehow, fuel cells are a mechanism for turning fuel into electricity.

https://www.bloomenergy.com/technology/

In Motorsport a gas tank is called a fuel cell. That may be where the confusion lies

The literally dirty secret is that a Bloom fuel cell is a device for turning methane into hydrogen on demand to fuel a hydrogen fuel cell. It's still emits CO2 and oxides of nitrogen like a gas turbine. They should be a lot quieter though.

Yes, there is a completely different thing with the same name. Race cars can use fuel cells to prevent sloshing. Those are basically fuel tanks with foam in them.

Most fuel cells can't be recharged. They take combustible gases, re-assemble the molecules in the gas into H2O or something chemically stable, and release the energy delta as voltage potentials. The cells work purely as catalysts. Only the fuels are expended.

It's just a type of battery, that takes fuel, as its refillable energy source. Weird but it works.

Oh that's a fun use - I'd heard some data centres were using fuel cells/mains-gas as a back up power source but using it as main power is interesting.

They can be resold, they have great value for anyone wanting to provide their own power to a DC. Demand is so high they might be able to make a profit.

Reply on news.ycombinator.com