That seems like a good distribution for digital art and entertainment.

That claim isn’t just a lie regarding the consumer preferences, but also the technological developments. It’s not impossible to predict at all; when talking about long term limits the picture is pretty clear.

We already know that a fiber can carry 20 terabits per second, because it’s already doing it in backbone ISP connections today. If necessary and economical this can be done towards individual houses in 40 years. The only hard thing to predict is the speed of price decay. But as a sort of anchor value I can tell you a pair of transceivers for 100 Gigabit per second over 10 km can be bought for 500 USD right now.

We also know that radio links are fundamentally limited by the Shannon limit, which relates the possible bit rate to the spectral bandwidth given a noisy channel (like our atmosphere). Maybe you’ll get a few dozen GHz of spectrum usable from space, in the mmWave region. And with that you probably reach dozens of gigabits per second. There is still a lot more to be had than today, sure, but with a definitive limit not too far off.

To be clear the Shannon limit applies to the optical transport in fiber too, but there we are working with around 4500 GHz of usable spectrum just between 191.5 THz and 196.0 THz (the conventional band for DWDM), way way way more than radio will ever have by the nature of things. And people are expanding that range downwards to 184.5 THz already, and that’s in the market, not just in research.

Old news, this was clear in June: https://www.phoronix.com/news/Linux-Kernel-TSC-Unconditional

Honestly, I’d still pick hard. Breezing through games just isn’t as fun as having the occasional struggle on the way.

I think it was time on October 26, 2018.

If anyone wants some more pixels:

https://cloud.google.com/blog/products/infrastructure/introducing-americas-connect

Direct link to the current release of the Top500 list:

https://top500.org/lists/top500/list/2026/06/

Whew Alps is barely still in the top ten.

It does not.

Is the bottom one made to look better with AI, is that a feature of the used car market these days?

Same thing. The group of designers who founded the art style in the 80s called themselves Memphis, then the Corporate Art inspired by it in the last ten years or so was called Corporate Memphis. Alegria is the specific name for the art style that Facebook commissioned and used in 2017.

Don’t give them any ideas, someone will combine the two. Machine generated Memphis slop-slop.

Liking Lasagne and hating Mondays?

It’s funny in the first place to argue from what “they allow”. The customer doesn’t have to care about your internal policies, fucko. Fulfill your end of the bargain.

It’s the NVIDIA Jetson Orin NX 16 GB. From the Ampere family, 3.5 years old. Small Tegra SoC.

I guess something like this is a typical dual use good, and doesn’t fall under the AI Chip export restriction either because it’s too weak. Seems like it would be trivial for Russians to import them from China or India or whomever really.

Not just Nvidia.

For a start of who the shovel sellers might be, you can take a look at the list in his post that Torsten Slok used to calculate his margin numbers:

Note: Data as of 2Q 2026 and for OpenAI (1Q 2026 estimate from PitchBook) and Anthropic (2Q 2026 estimate from Financial Times). Averages are equal-weighted bucket averages of Energy & Grid (Constellation Energy, Vistra, NextEra Energy, Vertiv, Eaton, Arista Networks), Silicon & Equipment (Nvidia, AMD, Broadcom, Marvell, TSMC, SK Hynix, Samsung Electronics, Micron), Compute & Cloud (Super Micro, Dell Technologies, Foxconn, Equinix, Digital Realty, Amazon/AWS, Microsoft/Azure, Alphabet/Google Cloud, CoreWeave, Nebius) and Models & Applications (OpenAI, Anthropic). Sources: Bloomberg, PitchBook, Apollo Chief Economist

In short according to his research the people currently making a profit or at least positive margins, are the people who build chips and memory, who build energy infrastructure, and who build servers and data canters.

I would like to see the distinction between Servers (Super Micro, Dell Technologies, Foxconn) and Datacenters (Equinix, Digital Realty, Amazon/AWS, Microsoft/Azure, Alphabet/Google Cloud, CoreWeave, Nebius) but he seems to have put them together.

Also I heard from our local news that ABB (which is Swiss, Edit: apparently also Swedish) is profiting too, they belong in the power category.

And I wonder why Arista goes in the Energy category, I’d see them and Cisco, and Juniper, and Extreme, and Nokia in a separate network category.

‘Profits are currently being funded by investors rather than earned from customers’

Which profits?

Ah, now I see the actual quote is: “[…] the AI boom’s profits are currently being funded by investors rather than earned from customers.” he means the profits of the shovel sellers, not AI profits.

Sure, I’m not saying Arm or Qualcomm are irrelevant in the gaming space, in fact I think Arm is going to replace X86 more and more, like we see it happening on servers too.

But FEX only runs on ARM64. And you wouldn’t put a platform with a 15 year old efficiency-focused CPU into the Steam Frame. So this Kuno platform is probably unrelated to gaming and comes from some other project instead.

I don’t think this is likely to be gaming related.
That’s a 15 year old efficiency trimmed 32bit ARMv7-A CPU.

Man those quotes from Chris Hays in the article hit hard.

I think I need to watch the video they are from tonight.

By the way, regarding integration, I just learned from a colleague last week that he uses emacs for editing remote files too. You can just open a file as:

/method:user@host:filename, with method being ssh or ftp for example

https://www.gnu.org/software/emacs/manual/html_node/emacs/Remote-Files.html