This would be a very interesting future. I can imagine Gemma 5 Mini running locally on hardware, or a hard-coded "AI core" like an ALU or media processor that supports particular encoding mechanisms like H.264, AV1, etc.
Other than the obvious costs (but Taalas seems to be bringing back the structured ASIC era so costs shouldn't be that low [1]), I'm curious why this isn't getting much attention from larger companies. Of course, this wouldn't be useful for training models but as the models further improve, I can totally see this inside fully local + ultrafast + ultra efficient processors.
> I'm curious why this isn't getting much attention from larger companies.
I can see two potential reasons:
1) Most of the big players seem convinced that AI is going to continue to improve at the rate it did in 2025, if their assumption is somehow correct by the time any chip entered mass production it would be obsolete.
2) The business model of the big players is to sell expensive subscriptions, and train on and sell the data you give it. Chips that allow for relatively inexpensive offline AI aren't conducive to that.
Well, there's a limit to how small we can make transistors with our current technology. As I understand it, Intel is already running into those limits with their new CPUs (they had to redesign the fins IIRC). I can imagine that without an actual breakthrough in chip manufacturing the size could stay large. That's not to say that a breakthrough won't happen, though.
Yes, in 2D, but NAND has been using layers for a while. We call HBM interposers 2.5D. 3D breakthrough would be pretty easy but for those pesky problems like power delivery and cooling. (/s)
But give that time (e.g. microfluidics) - something interesting is that it would be extra hard to use all layers at once, but NN might be a good fit, imagining that computation will be sparse (subsets activating simultaneously)...
That's the part that people are missing: it won't get smaller. It already required heroic optimization to get 8B on one megachip. Taalas is more expensive but faster. It is cheaper per token when running 24x7 but not cheap to buy. It will never be small and never be cheap.
The hardware isn't there yet. Apple's neural engine is neat and has some uses but it just isn't in the same league as Claude right now. We'll get there.
> I'm curious why this isn't getting much attention from larger companies.
Time is money and when you're competing with multiple companies with little margin for error you'll focus all your effort into releasing things quickly.
This chip is "only" a performance boost. It will unlock a lot of potential, but startups can't divide their attention like this. Big companies like google are surely already investigating this venue, but they might lack hardware expertise.
> I'm curious why this isn't getting much attention from larger companies
I would be shocked if Google isn’t working on this right now. They build their own TPUs, this is an extremely obvious direction from there.
(And there are plenty of interesting co-design questions that only the frontier labs can dabble with; Taalas is stuck working around architectural quirks like “top-8 MoE”, Google can just rework the architecture hyperparameters to whatever gets best results in silico.)
The price difference could be explained by LCSC purchasing in bulk directly from TI or similar and selling them at lower margins because their volumes are larger. I've seen "clone" chips sold at LCSC, but they're listed under a different brand (I can't recall one off memory unfortunately).
Hey HN, this past summer I went to NextPCB's factory and toured the facility. I asked questions from a variety of creators to factory workers at random.
Yup, all the software except the blobs for the SoC, display FW, and trackpad FW (since it's COTS) are open source! If I wanted to have those open-source, the entire machine would take a huge performance hit. I think of striking a balance.
The entire linux install is OSS, the keyboard is in ZMK, and the EC firmware is written in Arduino.
I'll just echo the amazement and congratulations of all the other comments. I do have a question though - your post stated "The hardest class I’ve taken so far was quantum mechanics in my junior spring term." Kudos to your educational system that allowed you to take quantum mechanics as a junior in high school - it looks like it clearly provided you a framework that allowed you to excel. Without giving away your privacy, is this some sort of special program where you live, or is it a standard opportunity?
He goes to Phillips Exeter Academy, where Zuckerberg once matriculated. Students' voices are heard through the Harkness method of teaching. There is plenty of opportunity for students to grow curricularly (e.g.,dynamic chaos theory in math, senior projects, though not required ...) and extra-curricularly (e.g. competition robotics, physics, bio,chem clubs, etc ...) which may not be a norm in most public or even private schools.
Hey Nirav, super super honored that you saw this! I've always looked up to you guys for inspiration and guidance. Thank you for the offer! Although I probably won't be mass-producing open-source laptops like you (i have a framework 16!), I would love to meet you. Would that be possible?
This is the best of the internet. Connection based on interest, appreciation, and mutual respect facilitated with a high degree of good faith. Hope you folks connect fruitfully, and also appreciate that you kept some of the "sent an email" and "thanks" public. Getting to see that this happened has given me a real boost.
Other than the obvious costs (but Taalas seems to be bringing back the structured ASIC era so costs shouldn't be that low [1]), I'm curious why this isn't getting much attention from larger companies. Of course, this wouldn't be useful for training models but as the models further improve, I can totally see this inside fully local + ultrafast + ultra efficient processors.
[1] https://en.wikipedia.org/wiki/Structured_ASIC_platform