Intel seems like a somewhat tricky one for acquisition given how the people most interested in owning them(either for architecture and product portfolios or for manufacturing capabilities and R&D) are outfits who any vaguely sensible antitrust regulator would run screaming from(Nvidia, as you noted, got slapped away from ARM; them eating Intel would be the bad outcomes expected for mobile and embedded visited upon client and datacenter, possibly even worse given Nvidia's GPU edge; and Samsung or TSMC are already large scale foundries with foreign ownership which seems like a hard no); and how it seems like basically everyone who wouldn't be immediately forbidden doesn't obviously gain anything from a fairly massive acquisition that they couldn't get just by placing a large order with Intel.

Especially now that Intel has gotten serious about multiple chiplet/multiple tile parts in the relative mainstream(not just the white elephant Xeon Phi stuff that paper launched because they had a DoD contract they had to fill; or some of the low volume tech demo foveros stacking mobile parts) any of the hyperscale 'cloud' guys would presumably find it relatively painless(especially compared to taking over and running things themselves) to request their pet accelerator co-packaged with some Xeon cores or just an Intel DDR5 controller; or their particular semi-custom adjustment of one of Intel's list parts towards their specific needs.

It's hard to escape the unpleasant feeling that a PE chop shop operation will be involved; with the extra-distasteful prospect of one plundering everything that isn't nailed down and eating the seed corn, as usual, then hitting up Uncle Sam for cash to preserve this vital national security asset; playing at responsible management just long enough for regulators to turn away for a moment, then repeating the process.

I'd be a bit surprised if they'd bother in this case: they are already hemmed in by the existing supply of cameras whose hardware cannot be changed and by the demands of professional photographers for usability with things like external preview displays in the field; and, realistically, driver signing requirements in modern OSes mean that they can do something utterly trivial and, so long as it requires a driver, everyone but the nerds will need someone to sign it for them; and this would involve actual engineering effort; but it's hard to be entirely sanguine about driver re-implementation against an adversarial vendor.

The trouble is that the hardware vendor controls the hardware; and that a peripheral is physically plugged into your computer but may be logically plugged in somewhere else, or in more than one place. Sometimes fairly loosely(unverified firmware built for a relatively well known architecture that can be reflashed either by a deliberately provided update interface or some fairly accessible bug that doesn't mean cracking the case); sometimes inconveniently(weak or nonexistent firmware signing; but you need to crack the case for a debug header or to clip onto a flash chip); sometimes more or less unshakably(some contemporary smartphones and consoles where everything is signed to hell and back and no exploits short of ion beam reworking a sub 10nm IC are know). As for the physical/logical distinction; it's relatively simple for the driver to provide some mixture of 'local' features(like a UVC camera or a mass storage class device) and a logically remote one that just uses your computer to set up a tunnel between something running on the hardware and something running on a remote host.

Microcontrollers with some cryptographic capabilities continue to get cheaper; and even ones that will do full mutual TLS with their own unique factory key and all are downright cheap now; so the price of entry to a device that doesn't have its own NIC; but, if provided with an untrusted tunnel over USB, is fully capable of strongly authenticating a remote host(nope, DNS spoofing won't help) and strongly authenticating itself to a remote host is pretty low indeed. If (as with digital cameras, which tend to be built around a fairly punchy DSP/ISP ASIC that's essential to pulling data off the sensor, plus a weedy little general purpose core that does UI and housekeeping) you've got a vital, heavily integrated, component to put the cryptographic chip into you can also frustrate attempts to physically remove it or swap it out: you can pry the DIGIC X off the board of a Canon camera easily enough; but what remains is basically a nice sensor and a lens mount: you aren't just swapping out a commodity PMIC or the like.

Put these two aspects together and you have a recipe for hardware that will substantially resist driver re-implementation: you can still connect high bandwidth or latency critical things directly to fairly basic local drivers(the actual UVC video stream; mass storage access to the memory card, etc); but if you want to lock something down (like the actual image processing parameters being applied to the UVC stream) you can require that to be a cryptographically secure request by the onboard cryptographic element to the vendor's host and then a response, unique to the requesting device, whose vendor signature the device will verify before applying the settings. Adds some latency and an always-online requirement(something you can increasingly get away with); and if you lose interest in the product or go out of business the customer is screwed; but that's a them problem not a you problem. Plus, unlike attempts to keep things closed on the client side, you don't even need to hassle with any bullshit obfuscation: as long as the keys stay secret you can use totally normal, off-the-shelf, approaches to cryptographic authentication; and the format of your configuration requests and configuration files can be as sane and readable as you like; because it doesn't matter if other people can generate their own configuration files when you are the only one who can generate signed configuration files.

Please note, I'm absolutely not in favor of any of this. All super-mega dystopian anti-customer stuff for which I'd hope anyone bold enough to try it would be punished severely: just to note that it would work, so long as the aspects of behavior you wish to control aren't too latency sensitive or incompatible with online requirements; and your hardware includes some core component critical enough to keep it from being swapped out. Fancy digital camera? If you are going to rip out the image processor you might as well just buy the appropriate image sensor from digikey and start from there. Console? Basically nothing but breakout board and power delivery for a SoC that does everything. Laser printer? Rasterizing postscript is not a challenge; not sure exactly how exotic the laser driver reading out of raster memory is. Pen plotter or filament deposition 3d printer? Don't bother; you'll just get brain-swapped.

That seems like a claim at the unfortunate intersection of 'arguably true-ish in certain cases'(in the sense that you can, probably, with some looking around, find someone who is a total propellerhead at shaving the insides of your loops when provided the documentation for the target architecture but who should probably be kept well away from software architecture; and at least one example of a genuinely elegant spec written at roughly the same formality level as a really high level language); but tempting to believe for really stupid and generally destructive reasons(mostly the seemingly widespread belief that software engineering being hard is some kind of con, based on syntax obfuscation, by the nerds against the Big Handwave Idea People, rather than a reflection of the genuinely nasty delta between mostly knowing what you want and how to get it and precisely knowing what you want and how to get it.)

Stopping sale of the static cell modem flavor of 'fixed' broadband, in a market where you are still selling cellular service to cellular subscribers, is basically the lowest commitment show of displeasure you could reasonably imagine.

It makes me curious if they are actually serious; or if they'll quietly show back up within 6 months if they end up not getting what they want. A wireline doing a take-this-market-and-shove-it would show at least some level of actual willingness to sacrifice.

Clearly we just need to introduce a rigorously and formally specified 'spec description language'; in which don't-call-them-programmers will write an internally consistent and sufficiently comprehensive spec, which we will absolutely refrain from describing in any way that might make it sound like source code; and then send that through something that absolutely isn't either a language translation process, interpretation, or compilation because it's not source code....

Then, once that crashes and burns under its own weight; we can adopt the "Pre-formal specification process", for people who don't know what they want or are afraid of syntax; and then we can have the not-programmers convert the informal specification into spec description language; and pat ourselves on the back for having achieved the no-code dream!

Which Office version is that? I had no problems with 2007 SP3 and older. :)

There are conservatives on Bluesky, and starter packs to follow them. https://blueskydirectory.com/s..., with a variety of viewpoints.
Anyone who's not a bigot, who makes truthful arguments, who genuinely wants a discussion, them I will listen to.

Any virtualizations would be slow like from VMware Fusion, VirtualBox, etc. :(

I think the problem there is timing attacks: so long as resources get cached across domains itâ(TM)s possible(how practical and how precise would vary according to fiddly details of exactly how common or distinctive various cached assets are) for a site you visit to draw inferences on where you have been previously by referring to a potentially cached resource and seeing whether itâ(TM)s available essentially instantly or whether thereâ(TM)s a delay that suggests your browser needs to grab a copy.

Will they be burning down data centres and sending death threats and letter bombs to AI devs, and leading mobs of hundreds of people armed with axes, pikes and guns in the same way?

I find it pretty remarkable that you sympathize with the Luddites when they were unambiguously wrong. Their personal motives may have been good, but they were absolutely on the wrong side of history, and they used extensive violence to try to achieve it (to the point that some industrialists started building what were basically panic rooms). The later labour movements that followed focused their aim much better: don't be mad at the machines that make wealth, but rather, be mad at those who direct all the resulting wealth to themselves.

Why not just buy the GPU separately and stick it in a PC slot?

What PCI GPU do you think is out there that has 128GB of VRAM?

AI tasks don't need much in terms of CPU or fast local storage. They do however impose a lot of bandwidth / latency and cooling requirements. It's easiest to just design to the task.

$3k is the price of a good gaming rig.

If you're looking for something *really* cheap, something like the Orin Nano is more your price range. Normal GPU-level amounts of VRAM but only ~$250 total for a whole system.