WEBVTT

00:00:01.330 --> 00:00:05.157
THE COMPANY SAYING EARLY TESTING SHOWS A
SUBSTANTIAL IMPROVEMENT IN PERFORMANCE PER

00:00:05.240 --> 00:00:08.797
WATT, WITH GIGAWATT SCALE
DEPLOYMENT TARGETED TO BEGIN

00:00:08.880 --> 00:00:12.697
LATER THIS YEAR. JOINING US NOW IN A CNBC
EXCLUSIVE, OUR OPENING EYES PRESIDENT AND

00:00:12.780 --> 00:00:16.337
CO-FOUNDER GREG BROCKMAN
AND BROADCOM CEO HOCK E. TAN.

00:00:16.420 --> 00:00:19.137
GENTLEMEN, THANKS TO YOU BOTH
FOR BEING HERE. GREG, LET ME

00:00:19.220 --> 00:00:22.137
JUST START WITH YOU.
WE SPOKE TO YOU LAST OCTOBER WHEN YOU

00:00:22.220 --> 00:00:26.577
AND BROADCOM UNVEILED THIS EFFORT. AND I'M
I'M SORT OF CURIOUS. YOU KNOW AT THE TIME

00:00:26.660 --> 00:00:30.517
YOU SAID THAT THE GOAL WAS TO
DECREASE THE COST OF EACH TOKEN

00:00:30.600 --> 00:00:32.517
AT A FIXED LEVEL OF

00:00:32.600 --> 00:00:34.687
INTELLIGENCE. DOES THIS DO THAT.

00:00:34.770 --> 00:00:38.257
AND IF IT DOES IT IS IT

00:00:38.340 --> 00:00:42.257
SUBSTANTIALLY BETTER
OR IS IT MORE INCREMENTAL.

00:00:42.340 --> 00:00:47.297
>> FIRST OF ALL THANK YOU FOR HAVING ME.
IT ABSOLUTELY DOES THAT. THIS IS A REAL

00:00:47.380 --> 00:00:50.367
PERFORMANCE IMPROVEMENT
ON PERFORMANCE PER WATT IN TERMS

00:00:50.450 --> 00:00:53.437
OF PERFORMANCE PER DOLLAR.
AND SO FOR A FIXED AMOUNT OF

00:00:53.520 --> 00:00:57.007
INTELLIGENCE FOR A FIXED MODEL,
IT IS SO MUCH MORE PERFORMANT.

00:00:57.090 --> 00:01:00.107
BUT ALSO IT MEANS WE'RE GOING
TO BE ABLE TO SERVE JUST MUCH

00:01:00.190 --> 00:01:03.517
MORE INTELLIGENT MODELS
MUCH MORE BROADLY AT SCALE.

00:01:03.600 --> 00:01:07.847
>> AND WHEN YOU SAY SUBSTANTIALLY BETTER,
WHEN YOU TALK ABOUT PERFORMANCE PER WATT,

00:01:07.930 --> 00:01:10.417
YOU AREN'T GIVING US NUMBERS YET.
AND I'M CURIOUS, YOU KNOW,

00:01:10.500 --> 00:01:15.757
ARE WE GOING TO GET THEM IN TERMS OF
TECHNICAL BENCHMARK REPORTS AND THE LIKE?

00:01:15.840 --> 00:01:17.957
>> ABSOLUTELY. YEAH.
WE'LL HAVE A DETAILED TECHNICAL REPORT

00:01:18.040 --> 00:01:19.057
OVER THE NEXT COUPLE OF MONTHS.

00:01:19.140 --> 00:01:21.897
I'VE SEEN THE INITIAL NUMBERS
FROM OUR TEAM. THEY'RE LOOKING

00:01:21.980 --> 00:01:24.967
VERY, VERY STRONG. SO WE'RE
GOING TO BE VERY EXCITED TO

00:01:25.050 --> 00:01:26.837
SHOW ALL THE DETAILS TO THE WORLD.

00:01:26.920 --> 00:01:29.207
>> ALL RIGHT. I ASSUME YOU
HEARD OUR LAST GUEST SAYING,

00:01:29.290 --> 00:01:32.177
WELL, THEY'RE NOT GOING TO BE
GOOD ENOUGH TO MEET WHAT WE'RE

00:01:32.260 --> 00:01:33.977
DOING AT CEREBRAS. DO YOU DISAGREE?

00:01:34.060 --> 00:01:36.977
>> LOOK, I THINK THAT THE
MARKET IS VERY BIG, THAT THIS

00:01:37.060 --> 00:01:40.047
IS A CHIP THAT IS DESIGNED FOR LM
INFERENCE FROM THE GROUND UP.

00:01:40.130 --> 00:01:44.187
WE REALLY WENT FROM SCRATCH
AND HAD AN AMAZING TEAM WORKING ON

00:01:44.270 --> 00:01:47.727
THIS, ACTUALLY DESIGNED WITH OUR MODEL.
SO WE WERE ABLE TO

00:01:47.810 --> 00:01:51.397
DESIGN THE WHOLE THING FROM END
TO END TO TAPE OUT IN NINE

00:01:51.480 --> 00:01:56.437
MONTHS, WHICH IS REAL, JUST LIKE BLAZING
FAST FOR THIS INDUSTRY. AND IT'S REALLY

00:01:56.520 --> 00:02:00.167
TARGETED AROUND OUR KNOWLEDGE
OF LM INFERENCE, WHERE THE

00:02:00.250 --> 00:02:02.367
WORKLOADS ARE, WHERE THEY'RE
GOING, NOT JUST OUR MODELS, BUT

00:02:02.450 --> 00:02:05.137
THE WHOLE INDUSTRY.
AND SO IT'S SOMETHING THAT'S UNIQUE AND

00:02:05.220 --> 00:02:08.677
DIFFERENT AND REALLY CARVES OUT ITS
OWN NICHE THAT DOESN'T EXIST RIGHT NOW.

00:02:08.760 --> 00:02:12.947
>> YEAH. I WANTED TO COME
TO YOU TO THAT ON THE NINE MONTH,

00:02:13.030 --> 00:02:16.917
YOU KNOW, DESIGN TO TAPE OUT.
IT WAS AN AGGRESSIVE SCHEDULE.

00:02:17.000 --> 00:02:21.127
HAS IT CHANGED THE WAY YOU DO
THINGS AT ALL AT BROADCOM? DOES

00:02:21.210 --> 00:02:22.387
IT BECOME THE NEW NORMAL FOR

00:02:22.470 --> 00:02:26.597
SORT OF CUSTOM AI SILICON.

00:02:26.680 --> 00:02:31.597
>> IT IS FIRST OF ALL THANKS
AGAIN FOR HAVING ME. APPRECIATE

00:02:31.680 --> 00:02:33.537
SEEING YOU AGAIN DAVID. BUT ALWAYS.

00:02:33.620 --> 00:02:36.777
>> LIKE TO SEE YOU AS WELL. THANK YOU.

00:02:36.860 --> 00:02:39.607
>> WELL IT'S A GREAT
OPPORTUNITY BECAUSE THIS IS A

00:02:39.690 --> 00:02:41.747
KEY MILESTONE FOR US. BUT

00:02:41.830 --> 00:02:44.677
ANSWERING YOUR QUESTION, IT HAS

00:02:44.760 --> 00:02:47.347
ALWAYS BEEN OUR ASPIRATION TO

00:02:47.430 --> 00:02:50.787
DO THAT. WHAT WE HAVE HERE AND

00:02:50.870 --> 00:02:53.657
IN A BROADER CONTEXT, IT'S IT

00:02:53.740 --> 00:02:56.227
VALIDATES VERY WELL THE

00:02:56.310 --> 00:02:58.297
BUSINESS MODEL WE PUT IN PLACE,

00:02:58.380 --> 00:03:01.627
WHICH IS SIMPLY THAT EVERY

00:03:01.710 --> 00:03:04.067
MODEL MAKER, MLM, FRONTIER

00:03:04.150 --> 00:03:07.437
MODEL, DEVELOPER, WORLD

00:03:07.520 --> 00:03:10.907
EVENTUALLY GO TOWARDS BUILDING,

00:03:10.990 --> 00:03:15.917
DESIGNING, BUILDING THEIR OWN
SILICON COMPUTE CAPACITY,

00:03:16.000 --> 00:03:20.687
COMPUTE DEVELOPMENT, SIMPLY
BECAUSE THEY CAN DO IT MUCH

00:03:20.770 --> 00:03:24.117
BETTER. AND HERE WHAT WE HAVE

00:03:24.200 --> 00:03:28.227
IS WORLD CLASS, LEADING EDGE
FRONTIER MODEL DEVELOPMENT TO

00:03:28.310 --> 00:03:30.457
COMBINE WITH A TEAM, ONE OF THE

00:03:30.540 --> 00:03:32.897
BEST TEAM IN THE WORLD, IN THE

00:03:32.980 --> 00:03:35.727
WORLD OUT THERE TO CREATE AI

00:03:35.810 --> 00:03:38.637
CHIP DESIGN, WORKING WITH US

00:03:38.720 --> 00:03:41.137
AND WITH ALL THIS IP SILICON

00:03:41.220 --> 00:03:45.007
TECHNOLOGY. AND WHAT YOU GET IS

00:03:45.090 --> 00:03:48.777
A IS AN AI ACCELERATOR, THE

00:03:48.860 --> 00:03:51.277
JALAPENO, WHICH IS BUILT

00:03:51.360 --> 00:03:54.487
GUIDANCE ON THE FUNDAMENTALS OF

00:03:54.570 --> 00:03:58.117
OPEN AI. LLM IT'S BUILT ON

00:03:58.200 --> 00:04:01.287
CHATGPT AND IT'S THE FULL STACK

00:04:01.370 --> 00:04:05.727
FROM THE MODEL ALL THE WAY DOWN
HARDWARE AND SOFTWARE TO THE

00:04:05.810 --> 00:04:08.210
BASIC TRANSISTOR. WHEN YOU HAVE THAT KIND