Apparently the real big ones are like $100k.

Spirit's assets are almost entirely their slots at ORD, EWR, LAX and LGA. They don't many of their planes.

I hear this a lot but their fleet is valued upwards of a billion, they own most of their fleet it was just financed with debt (it's not like a house where the bank has equity in the asset, the lender has debt with spirit and levers on that debt that control what it converts to on default but Spirit legally owns the planes). The slots are <100M total

Tough to decide who I trust less, denuvo or a ring-0 hypervisor I downloaded off BitTorrent.

Yes it did. You just couldn’t use it. I could send a text message while listening to music. Sometimes the music would crash due to OOM.

Decent? I'm not sure the new CAS models do anything that the TI-89 didn't.

Open loop cooling can work fine if they use greywater. The water isn’t potable anymore, but it goes into the sky and becomes clean again.

It’s all just a lack of imagination.

Only (mostly) water evaporate, salt and most contamination don't, so you get a brine that you must manage because otherwise it clog your heat exchangers and evaporation towers. Also, it must be returner to a river carefully to not kill all fish and life forms there.

I think a lot of patents have recently expired for 10GbaseT. I’ve noticed it getting a lot more accessible.

80 percent as good for 20 percent the cost.

But it is worse and more expensive…

Some basic math supports it. A GB300 NVL72 is about $6.5 million. Lets say that you need $6 million worth of cooling and another $6 million worth of electricity. At current rates, that's 720 billion tokens worth of Claude Opus 4.7. At 100,000 tokens per second, it pays for itself in about 3 months.

Obviously this is an extremely rough calculation. I can even be off by a factor of 10 and it's still a pretty good return.

Unless you're serving Chinese open-weight models - you have to consoder training costs. If you're off my 10x, then the amortization period is 30 months - far longer than the useful lifetimes of SoTA models. Frontier model development is a Red Queens race: you have to run as fast as you can, just to maintain your position.

The discussion was if Anthropic makes money on inference. They do. They lose billions on training.

No, because Anthropic can't serve their models unless they train them.

Training is akin to the cost of building the software/product. Inference is selling the product.

It's quite easy to sell something for a profit if you ignore the costs. Ultimate free money hack. I will start selling canned beans for the price of the beans plus a few cents. I will just ignore the cost of the cans, labor, power, machines, maintenance, distribution, storage and facility space. If I do that the few cents extra are pure profit.