The so called AI bubble will crash, stock prices will drop, but the demand for AI is obviously not going away; in addition it is only going to demand more and more RAM.
It looks like Micron and those who stand to lose profits from more RAM competition are the ones attempting to block Apple from using Chinese RAM.
Even if it does take 3 to 5 years to stand up a semiconductor fabrication plant, Apple alone can surely keep it busy.

Google says a single production line semiconductor fabrication plant that can produce DDR5X, LPDDR6, and HBM4 costs between $15 billion to $30 billion.
Google also claims Apple has access to maybe $40 billion in cash and up to $146 billion is liquid reserves.

So Apple can make a credible bluff that it will have a fab in 3 to 5 years to get its way today; and if they call it then Apple should just go all in and build the damn thing.

John will do it, Tim needs to step down already.

Yeah, out of the US frontier model providers Google seems to be giving us the best open weight solutions lately.
However, Gemma 4 is probably about a year behind the frontier models, and maybe nine months behind the Chinese open weight models.
It seems Google could do better than Gemma 4 today, given Qwen3.6 is very useful with similar model sizes.

Of course Google Gemini has been a distant third now in the frontier race as well.
I want Gemini 2.5 Pro open weights, last years frontier model, or Gemini 3.0 Pro.
Those are still far enough behind Gemini 3.1 Pro and Gemini 3.5 Flash that they would not compete with those services, plus they are probably so large that most could not run them on their own hardware anyway.
But they would compete with open weight Chinese models today and offer an alternative to running those on your own hardware.

XAi was supposed to release their previous best model once it was no longer competitive.
So Grok 3 and maybe even up to Grok 4.3 as of yesterday should have been opened.
But like Google, XAi with Grok is not a leader with US models, maybe taking third with Grok 4.5 until Gemini 3.5 Pro is released later this month.

I do not have the hardware to run Grok-2 at any reasonable speed, nor would it seemingly be worth it today.
I am sure the same would be true for Grok-3 and beyond.
I can run Gemma 4 at full 16-bit quantization, but Qwen3.6 seems to do better.
So I can appreciate the release of an open weight model that targets hardware a mere mortal pro consumer might have.
However, I will still rather give my $20 a month to a company that was releasing open weight models; and when my lender finally lets me take out a second mortgage for the rack H100s then I will be good to go.

It really seems like Google and the rest of them should be concerned about the Chinese models getting a foothold.
I would really like them to do something other than crappy legal tactics like blocking imports of Chinese models to avoid having to compete and give us open choices.
It is too bad that a US based company has not developed an open weight frontier model with a viable business plan.
Nvidia might be a hope there, since they will want to sell H100 level GPUs to every man, woman, and child in a few years; and we will want frontier level models to run on them.
But Nemotron is not quite competitive yet either.

This is the punchline.
Gold star for everyone!

If your exams are so easy students can cheat on them, that is a professor problem.
I had classes in college where you could have had the internet and phone a friend and still probably would not pass the exam without properly knowing the material; I respected those subjects.
The classes where you were not allowed to bring a back pack, nor wear a baseball hat, and only could use a #2 pencil to the exam were a joke; nothing taught, nothing learned.

If you want to pay tens to hundreds of thousands of dollars to "cheat" and not learn the material, what do I care?
A person using ChatGPT, the notes on the brim of their baseball cap, or the hired test taker gets it done; why are we concerned how?
Eventually it either catches up with them, or they are able to "fake" it and cheat through life; either way the work is completed.

Oh it is about academic integrity at Brown? Give me a break. Like there has been any of that anywhere for a century.

It really seems like OpenAi, Google, Meta, and SpaceXAi would do well to release more open weight models.
Their current models are still a few months ahead of the Chinese models, so their last versions would be right on par.
It still takes a lot of expensive hardware to run the truly competitive Chinese models, as it would the on release old American models as well.
As a result you are not really going to lose much if any of your customers to those running your models locally, because if they are capable of doing that they are already looking else where.

But sadly the American frontier Ai model companies have mostly regressed in releasing open weights.
Meta has gone closed and is no longer really even talked much about.
Grok open weight releases are are behind on the promised release schedule sadly; and generally even behind in terms of their latest frontier model.
Although it will be interesting to see how the new Grok 4.5, the first model it trained with Cursor, is going to score.

I would much rather hand my Ai subscription money to a company that released their model weights.
Better yet it would be nice if they shared their annotated data sets and training recipes as well.
However I am not sure how an organization that "gave everything away" like that would survive.
Right now it would have to be based on having massive compute and renting it out for profit that is put back into model development.

The Chinese companies either have a large side gig, like Alibaba, Bytedance, and Xiaomi similar to Meta, Google, and Microsoft here.
On top of that the Chinese government may be providing some support as well.

But if you at least released an open software stack that competed with or beat OpenAI and Anthropic models it seems like that should start to hurt their bottom line. The leading Chinese models appear to be no more than three months behind OpenAI and Anthropic. In addition their models seem to cost less to run.

I suspect their is a bit of distrust around anything coming out of China.
However a company capable of running these massive models would be much more secure using open weights they can run on their own hardware with much less worry that someone will take it away.
I would think this would be especially appealing for organizations outside China and the United States.
Mistral AI has said as much, working out of France.

Maximum? Most 4 day a week work plans I have seen still require 40 hours a week meaning four ten hour days.
I find that to be very appealing and perhaps more productive than five 8 hour work days.

However, anything less than 8 hours of work in a day is not what I what consider a full day of work by any measure.
Someone who works six four hour days is certainly not achieving a six day work week, but instead can perhaps at best claim 3 days of work spread out over 6 days.

But either way, if we set it at a maximum of 8 hour work days it only proves my point that workers in the EU are putting in fewer hours on average, less work, and it does not seem surprising that productivity has suffered.

I think you and I are in agreement.
In general I would assert that government regulation is absolutely the wrong solution.
This appears be no exception to that rule and driven by ulterior motives.
I am sure their is some incompetence on the law makers side, they do not understand how cloud computing works; but even if they did all they want is more power anyway.

According to https://4dayweek.io/countries it looks like the average work week is under 40 hours for most European countries.

Now sure you can stretch 32.5 hours out over 5 days working 6.5 hours a day, but I am assuming a minimum of 8 hour work days; so less than 40 hours is not filling 5 days.

My guess is that the US companies have data centers in Europe so an outage in the US would not affect Europe.

However I do not think that negates your point of being dependent on a foreign company or failing to have redundancy.
Your example of redundant Global Navigation Satellite Systems (GNSS) is a great example of redundancy benefiting everyone; I love having a GNSS chip that supports all three chips even if I am in the US.

Although cloud providers have always felt like they had pretty good interoperability and data portability to me; I just move my docker contain from AWS to Azure to Google Cloud to OVH; or have one on each cloud ready to go and one in my garage and my mom's basement just in case?

The issue is the services mentioned are centralized. My video streaming should be Bittorrent not Amazon or Netflix, my game server anyone can host not on Epic servers, and my communication pier-to-pier not all connected by a central host like Signal. Not sure why Docker made that list, just make a mirror of their repos and binaries? Really the list of services in the summary/article are kind of silly, most are not critical and the providers have a great incentive to make sure they are provided and redundant so not to lose customers.

Europe offers a progressive approach to work-life balance through widespread four-day work week initiatives and strict labor laws that make firing employees significantly harder than in the US. These rights prioritize job security and personal well-being over corporate flexibility. Who would have thought it had a side effect of being less productive?

Yeah the kid with no limit has their smartphone in their hand and is using it.

You should talk to someone about your violent fantasies before you hurt someone.

No, this will not be a Satanic Ozzy Dungeons and Dragon mistake, nor a big tobacoo like issue; this is like video games in the 90s!
Video gaming may be associated with better cognitive performance in children [https://www.nih.gov/news-events/news-releases/video-gaming-may-be-associated-better-cognitive-performance-children]
The children denied smart phones will be stupid like the ones that did not play video games.

Huh, made me think about a strait in California with an iconic bridge over it.
But you know maybe there is something wrong with me?

Eh, I am no Apple fan, nor a Steve Jobs fan, but the little I know does not support the assertion that Woz made Jobs; nor that Jobs was irrelevant without Woz.

Woz left Apple because Jobs' Macintosh was getting all the attention.
Woz did not go with Jobs to create NEXT, which lead Jobs back to his throne at Apple.

While Jobs created the next generation of computing that is the foundation to Apple today, Woz made the first universal television remote.

Now I would probably pick hanging out with Woz nine time out of ten over Jobs, and not just because Jobs is a rotting corpse and also dead.
Because yeah, Jobs was at his core it seems a "business man", but I am comfortable in asserting his vision was far more revolutionary than Woz's evolutionary engineering.

Or in short, Jobs would have found another Woz to use; Woz was lucky or cursed to have been found by Jobs.

How about a projector aimed right into them that plays a fake video feed. You would only have to corrupt a little of the data to make it not trust worthy.