@bloombergbusiness: Cerebras CEO Andrew Feldman displays the company's WSE-3 chip, the largest AI chip ever built, measuring 46,225 mm² and containing 4 trillion transistor. #AI #chips #Tech #CEO #BloombergTech
4 trillion transistor count, 900,000 cores, 48GB SRAM, 21K TB/s memory bandwidth. that last stat is STAGGERING
2026-06-07 02:04:35
73
Max Bantz :
I could of done this but didnt want to.
2026-06-08 17:38:06
0
sycicnok :
not enough memory baby.
not enough memory. cerebras will fail biiiig time
2026-06-06 14:48:06
0
jh52780 :
Big chips are what the microprocessor solved. Nobody wants a big chip, they want smaller ones with more performance.
2026-06-06 16:50:53
88
tiktok0214365 :
That is probably one of the hardest line ever. “It only cost half a billion dollars and 10 years of my life” straight out of a movie.
2026-06-07 17:14:40
13
Csbls :
Took so long and he couldn’t make it smaller.
2026-06-06 03:18:09
93
nervous racoon :
nobody laughed at his joke
2026-06-05 19:29:37
32
Savage Meerkat :
I thought this was the Onion 🤣
2026-06-06 14:48:17
41
faceone :
He acts like he paid for it
2026-06-08 20:04:02
0
Vlad 🍆 :
We’ve got micro ATX chips before GTA 6
2026-06-07 12:59:01
6
fernanditoo :
AI bubble is about to burst guys
2026-06-05 18:32:45
19
slurpslurp :
@slurpslurp:you don't understand this technology and that's fine... so this big chip solves a lot of problems.. instead of putting a lot of energy and a lot of space and a lot of components that fail and cause time lag they have put it all on a chip... this chip replaces an entire server cabinet... that's what this chip does. they have basically figured out a way to create all the processes that would be done on an entire server cabinet and did it on a single chip... that's why their products are extremely expensive because they can do an entire data center in a few u-blade server towers!!
2026-06-06 19:30:22
7
magnumninja :
wasted his entire life to make a chip bigger only for it to fail and cost to much money to do AI tasks for any company
2026-06-07 03:26:10
0
nunyadamnbness :
I don’t understand, what was the hard part of this? Lithography should have been easy, wafers all run through at 300mm these days. It should be easier to make a large chip instead of smaller chips. I would imagine quality would be a huge factor as particles could kill en entire wafer at this size.
2026-06-06 01:04:42
6
creempie80085 :
who cares bro
2026-06-07 21:11:20
0
slurpslurp :
@slurpslurp:put another way this does the work of 20 Nvidia b200 chips... and takes up a lot less space creates a lot less heat is more reliable and does it faster!! this big chip can have the entire large language model operate on a single chip. because it doesn't have to share information with other chips to operate the large language model it is supremely more efficient. when you have to connect computers in massive parallelism to run a large language model you are at a disadvantage... by placing the entire large language model on a single chip you use less energy you make results in far less time you take up less space..... this big chip is 125 petaflops, it has 900,000 cores.. and it's composed of 4 trillion transistors
2026-06-06 19:46:50
8
jayden14311 :
Nvida just made the smallest chip 😂and it’s better
2026-06-07 22:12:16
3
I Love Ohio :
Wow ten year old technology. Sounds revolutionary.
2026-06-06 09:28:29
2
ivan7475577 :
Which motherboard support this?
2026-06-08 13:06:08
1
AI :
I thought the idea was to make stuff smaller, not bigger.
2026-06-06 21:14:36
0
kromm :
The phone
2026-06-06 22:52:03
4
Tembusa :
is this a joke?
2026-06-08 09:32:40
0
matty.slick :
“We solved a problem that didn’t exist”
2026-06-06 23:00:28
4
To see more videos from user @bloombergbusiness, please go to the Tikwm
homepage.