blastdoor

I wonder if apple ever considered Global Foundries. Back then, Global Foundries was closer to the leading edge than they are today. 

From all that I’ve read about AVP it is possibly the most “pro” of all Apple’s “pro” products, or tied with the Mac Pro in pro-iness.

It seems unrealistic to expect this to become a consumer or prosumer product any time soon. Apple should lean into the pro market until the economics make it more feasible for consumers. 

Trump just hates the Biden policy because it was Biden's policy. 

But it's Biden's policy that resulted in a the TSMC fab in AZ and it's that fab that will enable Apple to get around this tariff. 

Earlier this year Taiwan changed the policy that TSMC can only fab on the latest node in Taiwan. This means that TSMC can bring the latest node to AZ much faster than before. 

So all Apple has to do is move a lot of inventory into the US before the tariff and pay TSMC to expedite the rollout of N3P in AZ. 

tundraboy said:

I think the market overreacted severely.  If the new programming techniques can match existing LLMs using less advanced chips.  Imagine what can be achieved using cutting edge chips.  It's not as if LLMs have already been perfected.  The last mile of development is usually the most difficult and most expensive.  Maybe cutting edge chips are needed to cross that last mile.

For example, I still don't trust AI generated summaries and shortcuts because they are still too prone to making up stuff.  AI hasn't solved the problem of how does an AI learn to distinguish what's true and not true, right and wrong?  Maybe going from 98% accurate to 99.999% accurate requires the most advanced AI chips, and lots of them.  I'm no expert, I don't know.  In fact nobody knows.

I think the market got it right with respect to Nvidia. Nvidia was vastly over-valued anyway, but the methodological innovations from deepseek seem to make Nvidia's advantages less relevant. In particular, they reduced the need for high speed interconnects across GPUs. They also greatly reduced the cost of inference, which means more inference can be done on devices -- like Apple's -- that have no Nvidia chips. 

The market got it dead wrong with respect to TSMC. This development is good news for TSMC because it means a wider diversity of firms can develop their own silicon for AI and be less dependent on Nvidia. Those firms then become customers of TSMC, cutting Nvidia out. It might even be good news for Intel, though Intel's board seems determined to sink the company.

AAPL actually went up yesterday and the market got that right, too. This is great news for Apple. It validates their approach to on-device computing and their unified memory architecture. It might even mean that they can use their own silicon for all of their AI needs, including training. 

Stabitha_Christie said:.

This is such an insanely bad take I have to question if it was meant to be satire  

Seriously, tell us you don’t know what open sourced software is without saying “I don’t know what open sourced software is.”  This is covered under the MIT license so it’s  not “controlled by hostile nation”.  Users can download it, change it and even include it proprietary software

The license itself makes it a fairly big deal. The other thing is getting it to run on lower end hardware. Your claim that whipping this up is somehow easy because other LLMs exist is just nonsense. If OpenAI, Microsoft, Google and Meta could get their models to run on cheaper gear, they would. All of them are spending a ton on data centers and trying to figure out how to power said data centers. This level of efficiency would be a boon for them. 

Totally this.