I just did a bit of digging. Ground-based solar panels cost about 20 to 50 cents per W. Space-based solar panels cost about 100 to 500 *dollars* per W.

Bonkers to think it would be cheaper to do this in space.

This is a completely different scale from Starlink -- orders of magnitude more compute and complexity. And the powergen is competing against ground-based solar.

What this really boils down to is that a space-based array only saves you money for a single component, the solar panels -- you need fewer of them for the same amount of power output. Everything else is more expensive. And guess what? Solar panels are cheap! They're so cheap that African families are buying their own and sticking them on their roofs so they can get reliable power. You're saving money on one of the cheapest components in a complex system. The most optimistic projection is that you can get away with one panel in space for every 40 on the ground. If a panel cost a million bucks, that would be very worthwhile. But a panel for a US solar farm is about 600W, and you can buy one for about 100 bucks! And none of this accounts for the fact that a ground-based solar panel will be much cheaper to engineer than a space-based panel -- about one or two *orders of magnitude* cheaper, because the operating environment is so much more extreme -- vacuum, radiation, crazy temperature swings, no engineer visits, etc. That immediately eats away the entire cost-based advantage of space-based AI compute.

This is a complete boondoggle of an idea.

Also, you ignored the cost and additional launches for sending the complete AI compute up into space, including the very massy chips, the radiators for cooling, and the structural trusses for linking it to the solar arrays. Because all of that power is going to become heat which must be radiated away. Plus there’s the chip hardening costs and indeed system hardening (micrometeroids). At least 100 further Starship launches, I reckon.

Meantime, a solar farm in the southern US that can deliver 1GW would cost about $1bn, and the battery bank another $1bn

Buttons are better, but dials or levers are better yet. Hitting a stupid toggle button that’s indistinguishable from the rest of the row of toggle buttons multiple times to change the temperature is just a stupid idea. I’m looking at you, Mercedes.

Of course. I'm just saying that trenching will still be the biggest component of costs, followed by site prep (eg upgrading electricity rooms) and hardware.

Smart for them, shit for consumers. Are you saying they shouldn't rely on Visa, Mastercard and Amex too? And why does this apply only to payment processing? Surely they shouldn't be dependent on third parties for any of their critical business functions, then? No use of third party logistics, heat and light and electricity and water and property assets and refrigeration and of course all the actual stuff they sell.

Wait it out, I guess. Supply will ramp up in response to heightened demand, some of the AI players will fall away, but we’re a long way off consumers deciding they don’t want a new phone every so often. It may end up with overcapacity in the supply chain meaning Apple gets to squeeze suppliers harder a few years down the line. I’m sure all the participants are busy running a ton of sensitivity analyses on different scenarios as they try to play this out.

I meant the biggest expense that will fall to the charger operators, rather than to the system as a whole.

I just don’t buy the argument that the cables supplying store power etc will be conveniently located to enable charger deployment just by drilling a few holes. You have to run the charger cables from the site’s electrical room or transformer, and that’s going to be a distance of dozens to hundreds of metres. I wish it weren’t the case, but it really is.

It’s completely mad, the US has vast tracts of land that are high insolation, you could absolutely have done this. It still could happen, and I guess to some extent it still will because the economics will drive it despite Trump’s admin doing their best to hold back the inevitable, but the pace is so far off the mark.

When you consider what still has to be done to get to industrial fusion, it’s clear it’s going to be decades more, as you say. We will need a device that produces sustained net electricity, not just net plasma energy; then a demonstration power plant that actually feeds power to a grid; then proof of materials surviving neutron bombardment at scale; then reliable tritium breeding at power-plant levels; and all of that has to be repeatable in engineering terms.

Fusion will have absolutely no part to play in decarbonisation because fusion is not going to be ready at scale in time, ie the next three through thirty years. Best guess is a first industrial plant selling power to the grid is going to be 2050 or so. We can’t wait till then.

Maybe at some point after that, fusion can take over from solar and wind, but fusion has major disadvantages that you’re skipping over: it’s centralised with all the political, regulatory, financing and single-point-of-failure risks that brings; and above all fusion WACC is going to be massively higher than renewables + storage + distribution. Especially when you consider how build costs for renewables and storage are going to continue to fall in the years up to 2050.

And you’re being altogether far too blithe on opex for fusion. 14MeV neutron fluxes are quite challenging, what with the embrittlement, swelling, activation, first wall replacements, superconducting magnets, tritium handling systems, cryogenics, robot replacement systems, etc.

Between them, they have tons of excellent EVs. The Inster is great and very cleverly done, and the Genesis GV60 is lovely inside

I think the modelling assumption is that the long term sales trend is inexorably up, no matter what happens in the next couple of years. Given how hard-nosed Walmart is, there's zero chance that they've modelled based on vibes or ideology, and so I actually take this as a quite bullish indicator on the future of EV sales in the US.

Plus increased dwell time. Shopping while you charge is the thing.