You are currently viewing SemiWiki as a guest which gives you limited access to the site. To view blog comments and experience other SemiWiki features you must be a registered member. Registration is fast, simple, and absolutely free so please, join our community today!
Is this the golden age of the semis industry? Even if the AI bubble pops, it feels like big tech, especially with cloud computing offerings, isn't going to wait on Intel to provide compute chips anymore. It's like Apple opened the gates to custom silicon for their offerings, and now everyone else is going to do the same or am I wrong?
With its newest AI chip, Microsoft plans "wider customer availability" after its initial product was only used for internal projects.
www.cnbc.com
Is this the golden age of the semis industry? Even if the AI bubble pops, it feels like big tech, especially with cloud computing offerings, isn't going to wait on Intel to provide compute chips anymore. It's like Apple opened the gates to custom silicon for their offerings, and now everyone else is going to do the same or am I wrong?
Lego’s first-ever CES appearance brought a brand-new line called Smart Play, featuring a Smart Brick that lights up, interacts with motion and proximity, and has sound effects onboard. In one Star Wars demo, the brick glowed like a TIE fighter’s stare and played lightsaber whooshes — no phone or app needed. Even Luke Skywalker’s new minifigure can make his saber emit sound on its own, pointing to a broader shift in how Lego combines tech and tactile play.
Is this the golden age of the semis industry? Even if the AI bubble pops, it feels like big tech, especially with cloud computing offerings, isn't going to wait on Intel to provide compute chips anymore. It's like Apple opened the gates to custom silicon for their offerings, and now everyone else is going to do the same or am I wrong?
I agree completely. Apple set this all in motion, what I call fabless systems companies. Off the shelf computing has been dying a slow death for years now. With billions of dollars at stake, domain specific chips are well worth the effort. It also looks better on Wall Street if you control your own silicon.
I have mentioned this before, it is all about prototyping and emulation. Apple builds their software in concert with the chip so it is ready to go when the silicon is in production. That is a huge competitive advantage.
I think the door is beginning to close for new entrants for the data center barring any disruptive technologies (optical AI computing), though there remains a lot of open territory for custom silicon at the edge. I say that because the table stakes for data center have moved to co-optimized rack-level systems and beyond for the silicon suppliers. Some data center level new entrant chip suppliers still might survive by doing specialty chips for specific hyperscalers, but not on their own.
I agree completely. Apple set this all in motion, what I call fabless systems companies. Off the shelf computing has been dying a slow death for years now. With billions of dollars at stake, domain specific chips are well worth the effort. It also looks better on Wall Street if you control your own silicon.
I have mentioned this before, it is all about prototyping and emulation. Apple builds their software in concert with the chip so it is ready to go when the silicon is in production. That is a huge competitive advantage.
There are also several cost savings or design advantages that custom CPU chips can provide:
1. No hidden costs for marketing, sales, product management. In Intel and AMD these are large expensive functions.
2. No added cost or die area for features you don't need, or optimizations you won't use.
3. You can save expensive die area and power requirements if you don't need (for example) various hardware accelerator blocks or higher clock speeds.
4. You can focus on what makes a difference for your specific applications, like larger caches, more memory channels, more I/O channels, rather than worrying about industry benchmarks.
5. You can synchronize your CPU development pipeline with your deployment pipeline.
I think we are forgetting on thing with ASIC software you need software for these stuff which takes time to mature your Chip is just a Piece of Sand if you can't run software on it.
So where is the Intel 18A product of Microsoft?
I have read that Microsoft is one of the customers of 18A, and I originally think this product is ASIC.
But Maia 200 is N3, and I heard Maia 300 is N2, so where is the 18A product?
I highly doubt Microsoft or any other company will put a high volume chip through through Intel or any other new foundry on the first run. That would be a pretty high risk decision. I do believe Intel Foundry will get 4-6 big name customers but it will not be on a high volume mission critical chip like Maia or any other complex chip.
I'm a little shocked that Tesla will use Samsung 2nm for their next AI chips ($16.5B) considering they have much more experience on TSMC. Tesla does buy Samsung products for their cameras and other stuff but this will be the first chip using Samsung. I'm guessing the price was right. Maybe the deal is a most favored nation type deal like what Apple got from TSMC?
In parallel I would bet Tesla will use TSMC for a plan B.
I think we are forgetting on thing with ASIC software you need software for these stuff which takes time to mature your Chip is just a Piece of Sand if you can't run software on it.
S2C Inc. is a worldwide leader of FPGA prototyping solutions for today’s innovative designs. S2C was founded in 2003 by a group of Silicon Valley veterans..
Custom silicon only creates value when software, tools, and deployment are co-developed alongside it—otherwise, even the most advanced chip never leaves the lab.
Looking for leadership, state of the art, silicon proven, semiconductor IP & Software for your next SoC? T2M, your one stop technology supplier and provider
Custom silicon only creates value when software, tools, and deployment are co-developed alongside it—otherwise, even the most advanced chip never leaves the lab.
Looking for leadership, state of the art, silicon proven, semiconductor IP & Software for your next SoC? T2M, your one stop technology supplier and provider
That's true for specialized ASICs, but many (most perhaps) of these custom chips are Arm-based CPUs using mostly Arm IP for functional blocks (especially the cores). And every cloud computing company has their own VMM and OS distributions. For TPUs, as an example, I agree with you, and Google has many people developing custom software for every TPU generation. But for Graviton, Cobalt, and Axion, I think the software tasks are not so difficult.
www.cnbc.com
