I am actually thinking about my next BMW (my current one is getting old). Knowing this, the iX3 is off the list.

BMW had a great control system, intuitive, possible to do many things without even looking. "Young people prefer" ? Really? The young people that actually buy your cars? Well, they'll learn and not prefer touchscreens anymore on the 2nd car.

Inference is neural network limited. On systems with entirely separated data and compute, this is a memory bus constraint. In quantized systems, ALUs and FPUs are illogical. A simple 16 cell LUT is suitable for noise free multiplication. A neural network circuit with a dedicated LUT and addressable buffers more or less eliminates bandwidth constraints. But this requires neural networks more similar to FPGA cells rather than classic compute.

That said, separating inference from data drops network depth considerably. And in transformers depth increases memory access exponentially which is why nearly identical MoE and dense models perform so differently. We of course need data centers for models with massive numbers of active parameters. But massive numbers of active parameters weaken models as that means depending on training rather than looking up facts. Huge active parameter sets will never be a good design. A real world analogy is that huge models are like Jeopardy champions who have crap loads of partial trivia facts in their heads. Smaller models are like librarians who don't have a trillions useless facts memorized but knows how to find the data for the researchers who will use the librarian repetitively to follow through to the answer. The model only needs enough training to use the data.

What really matters is, how fast it can find the data.

Inference suffers greatly when you treat weights as data. And bigger models have more trivia like facts memorized from their training. This means more active parameters and greater network depths with higher memory bandwidth needs. Even now, you see the chatbots are improving drastically because they rely substantially more on data and way less on inference.

I am not sure which part you consider gibberish, but it makes perfect sense to me. But it might sound better in my head.

P.S. I actually have considerable data to backup many of my claims. At work I'm sitting on about 50MW of compute. We build in shipping crates in a repurposed mine. There is a group of us where our goal is to avoid spinning up more compute. We need to deliver inference to 200,000 users eventually. (not customers, employees) and we could throw millions at NVidia and end up hosting some crap model, but we focus instead on cutting that memory bandwidth need. And yes, I'm far from being the smart guy on the team.

This seems smart.

The federal government needs
  a) more tax money to reduce their dependence on more bonds
  b) more tax money to redistribute to startups
  c) reshoring to avoid just being too far behind.

So, taxing imports is a great idea. It doesn't matter if the trickle down effect works or not. The government has bills to pay. The government is owned by the people. Therefore, the people need to pay their bills. These taxes are good... we need to reward Trump for finally bringing new socialism to America. And what is best is that this really will steal from the rich and give to the poor.

BTW... Little secret, if TSMC spun down there business right now over the next 12 months, we'd be fine. We might have to rewind our tech a year or two and it might take some time to grow capacity, but it really wouldn't matter. TSMC just gives us a 12-18 month boost over their competitors. If they scale down as Intel, GF, Samsung, SMIC and a few others scale up... it really wouldn't matter.

The absolute best data centers today should be written into history as a dark dark stain on the progression of humanity. It is a sign that no matter how stupid we have been up until now, we managed to find that hole has no bottom.

Consider this... The reason we need data centers that massive today is because the tech isn't ready.

NVidia B300 filled racks using 100kw each are barely capable of running modern frontier models. Even if we designed more powerful chips that could do the job, there isn't enough silicon wafers produced in the world to do the job. Even the most amazing semiconductor technologies on earth are not good enough. Consider that even if we could build 1TB HBM4 NPUs, the time it would take to generate a token on a transformer that large would be slow.

What we know

Transformers will leave the data center between 2030-2032 as it will never make sense to run transformer inference in racks. It's at least 1000x less green than local inference.

Intel, AMD, NVidia, Apple and every RISC-V/ARM chip supplier on earth is 100% focused on making local inference the thing.

We will soon stop training massive models that start becoming obsolete the moment the last parameter is fed. Even now, Qwen 3.8 27b is good enough for a year at least (first model to do this).

For LLMs, we'll just stop wasting cycles on huge networks and will focus more on stronger harnesses attached to better data sources.

Cloud AI will be more about acting as data providers. Models will become reasoning engines, the "smarts" will come from things like vector databases.

Vector databases will use kilowatts, not megawatts. Data storage will be much smaller.

Things like Intel's dead Optane product will become the future of datacenters. Google Search, Microsoft Bing, maybe Baidu are the best positioned companies for what comes next and they don't need 100KW per rack to deliver. If I were to enter this segment, I would pair a customized KLV object query engine onto a 1024 bit wide ecc memory bus and connect it with low cost networking (think Intel's Ethernet fabric tech). Then add a super capacitor to flush to flash on power loss... Or just hope Optane comes back. Superfast, massive, token based vector databases are the future of cloud LLM

What's next. In 3-5 years, every single rack in every single AI data center is trash

This is fact.

Even now H100 racks should be heading to scrap.

The chips are slow, they use old heat spreader tech, they get almost no performance per watt compared to B300. They are much harder to cool. They have so little VRAM it takes piles of them to run newer models. PCIe data center GPUs were always irresponsible purchases. SXM GPUs only run on special systems Noone but high end data centers can even run.

Consider that every single B300 or older equipped rack on earth is landfill in 3-5 years.

By then, the world will have overprovisioned RAM, we'll have moved 2-3 generations of cooling forward, we'll have designed denser racks, started employing all the awesome tech China built to work around lack of EUV... No really, Korean companies are already licensing CXMT tech for wafer stacking. We will have moved to lower fresh water depen....

You know what? It is going to just be cheaper to bankrupt the shell corporations who own the current data centers than to upgrade them to new tech. Besides, the environmental issues will be such a mess. Imagine trying to dispose of that much fr4 epoxy without causing environmental disasters? It's probably at least as much epoxy as the total mass of the twin towers.

I can keep ranting but I promise this, there isn't a single piece of tech in the most modern AI data center on earth that is worth using in 5 years... Not even the racks. It's all garbage.

Global AI tech spending is said to surpass 2.5 trillion this year.

Sorry, Gates is a small fish in that pond.

The two articles were probably written by two different ghost-writers.

AI isn't actually smart. It's an at best average intelligence thing with an insane amount of available knowledge. Like the village idiot with eidetic memory.

As soon as you do any work with AI that's not available as a ready-to-apply solution from its training data, you begin to notice how much you need to correct it, keep it on track, remind it of basic constraints you already mentioned five times, etc. - it's performing at a level I'd accept from an intern, but not from a co-worker.

What it does have is data. Tons of data. Any question that would take you and me half an hour on Google and Wikipedia to answer it can answer in 10 seconds. That's a great tool. It is not intelligence, however.

How is this "news that matter"?

Bill Gates is an old man who has zero AI credentials and has long ago become irrelevant. Not to mention his previous predictions of future developments are not exactly brilliant.

lol

Welcome to the 21st century, you might have to catch up on a few things. Start with Mozilla Media Engine and AVFoundation / WebKit Media Stack.

Your website sends me an H.264 stream (or whatever you decided you'll use). It doesn't need to know shit about my graphics hardware.

Complete bullshit. Even for fullscreen video, the website doesn't need to know my screen size. It should tell the browser "here are the resolutions I can offer you" and the browser picks one of those. All the website needs to know is if I prefer 1080p or 4K. And I might pick 1080p even if my screen is 4K because, say, the hotel wifi is slow.

Complete bullshit. None of these things break if I don't tell you my OS. I've got extensions in this browser that can fake my agent string and nothing breaks. Fonts? Are you kidding me? Be more specific, what exactly do you mean by that?

And there's no reason it needs to. What legit use case requires a website to know whether my GPU is an NVidia, AMD, Intel or whatever chip? A shitload of the capabilities are only needed for 3D scenes (like max texture size), which is 0.1% of the websites I visit - they can ask and I can whitelist just them. Why does it need my refresh rate?

Again: All of these things have a legit use case, sure. Often exactly one. That is rarely if ever used. There's absolutely no reason that EVERY WEBSITE can query information that only one website in a thousand actually needs.

You don't tell the baker, florist and car dealer your shoe size just because the shoe shop needs it to serve you, right?

Because I want to be irrelevant to the fucking ad industry. I don't want them to track me, profile me and try every trick in the book to manipulate me into buying shit I don't need.

Sure, the same way that putting a finger into another man's ass has a legit use case (prostate examination) - but that doesn't mean I want every man to do that to me all the time.

The point is that tech companies are exploiting the legit use cases by abusing them when you're not in that use case. Basically, they think they can finger up your ass whenever they want, because once a few years ago you consented to a doctor to give you an exam and maybe you will do that again in the future.

I don't mind it being capable of. I mind it doing all kinds of shit without telling me and without giving me a choice. For example, I absolutely do not want websites to access my microphone. Or even know that I have one. For the once-a-year case where I might actually use that, I can enable it.

The problem is the same problem that the cybersecurity field has been fighting for 40 years: Permit All being the default.

It doesn't NEED to know any of these things. My browser knows the window (not screen) size and if your CSS is well-formed, it'll manage or maybe your page looks shit if you build it shit. It needs to know my OS only if I want a download link, not on every page I ever visit. It doesn't need graphics capabilities unless it wants to bypass the browser rendering engine, in which case it should ask for permission or fuck off.

The thing is that we give up too easily and the major browser manufacturers (one of which is an ad company themselves) are complicit.

I want "fuck the ad industry" mode, where the browser generates random but plausible values for every non-essential information a website requests. Different ones every 5 minutes.

He has. For a long time. The problem is, apparently he doesn't like to pay them, and they're not aware of that before the fact.

This is essentially a high-stakes gamble. The payout is potentially millions, so they just try whatever. And I bet if it all turns against them and they're forced to pay legal costs etc. they'll just file for bancruptsy.

Perfect illustration of the thinking problem here. You don't even get WHAT the problem is.

Why is a driver able to fuck up a system? The OS should be the watchdog that says "WTF are you doing there?", not the obedient slave executing whatever.

A great illustration is W^X - first found not in Windows, but in OpenBSD. Because Windows happily goes "whatever you say, master", while the guys at OpenBSD said "no actual use case of non-broken, non-malicious software should ever behave like that, so we'll block it".

If you are responsible for a sensitive location, and a five-year old kid from the neighbourhood one day just wanders in and take a look around out of curiosity, pressing a few buttons to see what happens - you would definitely not keep your job by claiming that the little kid is actually the problem here.