Yep, most of the "purchases" made by AI companies were done to screw over the other AI companies.

When the bubble bursts, the contracts will evaporate and the big memory manufacturers will slow down their output. The existing stock of memory will be sold to the likes of Corsair and Kingston who will then harvest the memory chips for consumer products.

I am a bit more optimistic. I think once one of these companies goes IPO it will flop and the money will leave (see SpaceX IPO's plummeting and DeepSeek eating their lunch). Nvidia will pivot to pushing local AI and remain healthy (and probably buy back a lot of their stock on the cheap)

Except for HBM, memory chips are memory chips. Server memory (ECC) modules just have 9 of them instead of 8.

AI generates output based on whatever was loaded into its model. Just as AI images all look the same, so to will be the stories they produce (until AI starts consuming AI generated stories, then it will become bat shit insane). AI generates stories will be no better than the bargain bin knock off crap that existed before. AI can't innovate, it can only slap stuff together based on probability.

It isn't that easy, there are a lot of APIs and expected behavior that would break. They would need to make their own WINE.

Plus they would loose a lot of control over the eco-system that they have now. In the grand scheme of things Windows doesn't cost them a lot of money and it gives them the ability to push standards.

Innovation happens from failure. I like that Intel was trying new things. Even failures bear fruit (Sandybridge benefited from a lot of things introduced by Netburst). Just like how AMD learned from the disaster that was Sledgehammer.

No it couldn't have. The design was discovered (to be fair, thanks to Itanium's failure) to be fundamentally broken. The compiler can't know what will be hitting the CPU when the program is actually run.

Even without Itanium they would have died. Intel and AMD caught up to the performance and were far cheaper with a larger user base. With the exception of ARM the RISC cpus were on the retreat from workstation to server to specialty case.

He mentioned new deserts. Where are they?

Where are these new deserts he mentioned?

I live in Southern Europe, I have no idea what you are talking about.

Which "major" platform doesn't have a rust compiler?

The current CEO's focus is not going bankrupt. The last CEO spent too much and almost killed the company.

This 100%. If I had the points I'd vote you up.

Bitcoin will never beat the US dollar for criminal use.

Every time you use an AI platform the company doing the processing losses money. Raw GPU performance (which is also raw AI performance) is not increasing fast enough to make it profitable.

Chip fabrication gains has been slowing down massively over the last decade. Compare a 1990 computer to a 2000 computer, massive advancement. 2015 to 2025, not nearly as much. More and more cost is now required for less and less gains on state-of-the-art chip manufacturing. This happens with all technology, it matures to a point where it plateaus.

AI requires '90s like gains to become profitable, that isn't happening anymore. So AI will find itself used in specialized areas (movie SFX, search, modeling, etc...).