Fair, but my point is that a data center is not pets.com. They are a much more flexible asset, and are even more flexible than AI itself. Pets.com failed because the physical delivery infrastructure and consumer buying habits that support Chewy today did not yet exist and took years to create. Data centers can be repurposed from supporting the current generation of LLM-based AI models to an infinite variety of use cases. Pets.com could only ever hope to deliver pet food.
And Amazon.com could only ever hope to deliver physical books?
Anyway, it's clear an AI data center has utility for crunching AI inference, and that there is and will continue to be demand for AI inference. The trillion dollar question is whether you can make money doing this, and so far the answer is no.
You are not disproving but rather reinforcing my point. Amazon survived the dotcom bubble precisely because they could hope to do more than deliver physical books. It was a more flexible, generally applicable business model than pets.com, and they started out targeting a market more susceptible to ecommerce conversion than pet food delivery proved to be. Similarly, a company selling a product that can only be used for the limited use cases we have found for chatbot-style text generating LLMs will be more likely to fail than a company selling something that is more generally useful. Data centers are more generally useful than LLMs, and the companies building them are more likely to survive than those whose fortunes are pinned entirely to unrealistically high expectations for replacing white collar workers with LLMs.
Data centers aren't just physical buildings though - the vast bulk of data centre costs is the racks of servers in them, and those are being bought and installed on credit at very high costs.
Its not about "is a computer thoeretically useful" its about if it's going to deliver value to cover the cost of installing it in the first place over it's expected lifetime.
You aren't substituting generic compute: you're substituting highly optimized GPU dense server space. The application is more limited then it looks.
Coal isn't cheap. Specifically coal powerplants are not and it's got nothing to do with emissions. One of the pinch points in the Australian power grid has actually been that a bunch of coal plants are aging out and no private enterprise wants to even bother financing their replacement because it's a money loser compared to renewables build out.
Fortunately this is well on it's way to being solved by the massive battery build out which has happened which means evening peak electricity usage has declined enormously[1] - i.e the primary "sun not shining" point of the day is rapidly turning into an energy use low point for the grid (which makes sense: business powers down, people go home and their batteries which have been charging cover cooking - I have an induction cook top and a 10kw inverter covers everything).
There could be other regulatory and tax hurdles to an Australian coal build-out. We know that coal is at least as cheap as nuclear, and that the electricity prices went up sharply when Germany replaced much of its nuclear energy with solar energy. Therefore coal is cheaper than solar. The mistake that most people make is that they don't include the highly expensive battery or gas backups during the night in the cost of solar.
Yes there could be but there aren't, because conservative state governments have had to resort to trying to subsidize coal power plant investment because no one wants to touch it.
Two things can be true: solar can be unable to provide reliable base load power, and it can be far more profitable investment return then building a coal fired power station.
Which is what has happened: and now, with the falling price of batteries, batteries are a far more profitable return then building a coal fired power station.
Coal is expensive: solar, other then some light cleaning costs, is one and done. There's few mechanical parts and no fuel costs and you can build it anywhere there's an electrical grid (which is everywhere).
Batteries are similar: sure they might degrade, but that's over 15 years. At the right time of day, charging them can be between free, and profit generating due to all the solar.
A coal power plant is gigantic. It can't be built anywhere - it needs to be physically proximal to either a port, or an actual coal mine (which the one's near Hunter Valley in NSW are - I've driven next to the train lines for them, for example), and usually a water source too for cooling. It's a large, maintenance intensive structure (the reason the ones we have are shutting down is the maintenance has gotten very expensive because the whole structure is aging out).
And after all that, it also needs fuel - throughout it's lifetime it will need to constantly buy fuel. But it won't constantly be profitable: most renewables have a massive advantage in that they can switch off from the grid nearly instantly. So renewables are far, far better at reacting to negative power pricing conditions then thermal power plants which cannot spin down that quickly: so a coal power plant (really any thermal power plant) will usually have to pay fines when power prices go negative, whereas while solar might not like shutting down, it can do so immediately (and batteries can just hoover up all that power and claim the credits).
Grid stability is a different department, quite literally, to power generating investment which is a commercial enterprise chasing return. And solar was better, and batteries are catching up quick (partly because negative prices mean your solar investment won't be able to bank a solid 6 hours of profitable sunlight without them). The new thing is also batteries for wind - DC couple grid batteries at wind plants, turning them into much more dispatchable generating assets.
There is a world where scale and experience means it's about the same though, is the thing.
And memory is already expensive. It's downright hard to even get it though - you frequently would prefer not what's cheapest, but whatever is in largest scale production.
Scale and experience almost entirely share between HBM and normal memory. And they're both in large-enough scale production to not have a big difference on availability; if you're willing to pay HBM prices you should find even more sellers of DDR.
The only way I see HBM becoming competitive for consumer CPUs is if they solve the yield issues. Or if AI crashes so hard that people are putting those GPUs on fire sale and salvaging mass quantities of HBM off of them.
It seems unlikely as the entire machines with HBM on them are apt to be sold whole on the market and snapped up quickly. With how much demand is in the current market machines 3 years old may not be sold if they can't get newer faster machines fast enough.
For the better part of two decades I frequented a particular forum, and one of the most common reasons people tended to eventually get banned wasn't because of any particularly egregious political opinion, but because whatever opinion they held they tended to turn every thread they participated in anywhere into either a thread about that or a thread about them.
In Australia it's pretty great: my driver's license can be provided by an app on my phone and legally must be accepted if it is. So between that and NFC payment support, a phone is the only thing I need to remember to operate outside the house.
It's pretty great as long as it doesn't evolve into them no longer printing the cards and requiring use of their app, consequently on one of their approved OSes.
I mean yes but that is a case of the slippery slope fallacy: the existence of a future possible decision does not affect whether a current decision is a good one.
If they try to move in that direction, we can protest it then. Denying ourselves a useful capability now doesn't stop it from happening anyway, nor make it more likely.
I personally live in a country with a large religious population that rejects the smartphone, so thankfully I personally have not encountered a situation where needing a smartphone is vital. However we've seen many people right here on HN mention how some specific vital service - often government or banking - does require a smartphone.
In my HN Threads feed, the very next item is another HN user mentioning how they have difficulty paying for parking spaces without a smart phone, from this very page. Search "and sadly accept the cost of our principles" on this page to find it.
But I have no interest in winning internet arguments. You win, the cookie isn't worth it to me.
Keepass2Android without network permissions saving the file to a folder monitored by Syncthing is pretty good though - it keeps responsibilities nicely separated.
If passkeys were meant to be user friendly then there'd be a secure optical transfer mode to QR code them from device to device with the screen and camera.
Easy to implement (receiver flashes a public key, sender encrypts to that key and flashes the QR code back).
That it doesn't exist for a protocol meant to work with phones tells you exactly where the thinking was headed.
That they have purposely put effort into preventing you from doing that shows you everything you should need to know.
The entire point of passkeys is yanking the control over authentication out of the hands of people. These companies don't want you to be able to let someone else log in as you.
That's the entire point.
The problem they are solving has absolutely nothing to do with security.
The peak was my high school Gateway laptop.
reply