WEBVTT - Tenstorrent Inc President and CTO Jim Keller Talks Chips

0:00:00.280 --> 0:00:02.480
<v Speaker 1>Let's just shift gears a little bit, talk chips more

0:00:02.480 --> 0:00:05.600
<v Speaker 1>broadly and how there's some new entrance into the AI

0:00:05.840 --> 0:00:08.240
<v Speaker 1>part of the equation. Jeff Bezos, for example, and Samsung

0:00:08.280 --> 0:00:11.880
<v Speaker 1>are betting big on AI chip startup ten Store, announcing

0:00:11.920 --> 0:00:14.200
<v Speaker 1>a seven hundred million dollar investment, which puts the company

0:00:14.200 --> 0:00:16.800
<v Speaker 1>now in a two point six billion dollar valuation. Joining

0:00:16.840 --> 0:00:19.520
<v Speaker 1>us to discuss is the CEO of the company, Jim Keller,

0:00:19.680 --> 0:00:22.720
<v Speaker 1>who is himself a legend and chip making world. Of course,

0:00:22.760 --> 0:00:26.279
<v Speaker 1>your resume includes know who's who of AMD, Apple, Tesla

0:00:26.320 --> 0:00:29.920
<v Speaker 1>and indeed Intel. Jim, the money that you're raising, what

0:00:30.040 --> 0:00:31.920
<v Speaker 1>is it that you're hoping to do. You're wanting to

0:00:31.920 --> 0:00:35.280
<v Speaker 1>make a more cheaper operation and option out there for

0:00:35.680 --> 0:00:36.880
<v Speaker 1>AI accelerator chips.

0:00:37.880 --> 0:00:40.760
<v Speaker 2>Yeah, that's right. We're building out the team, we're building

0:00:40.800 --> 0:00:43.879
<v Speaker 2>out our products, We're starting to ship our products, and

0:00:43.920 --> 0:00:46.280
<v Speaker 2>our real mission right now is to bring on developers

0:00:46.400 --> 0:00:47.840
<v Speaker 2>and early customers.

0:00:48.479 --> 0:00:50.800
<v Speaker 1>Let's talk about the customers and what you offer that

0:00:50.880 --> 0:00:52.640
<v Speaker 1>isn't in the market at the moment. Jim, what's your

0:00:52.680 --> 0:00:53.200
<v Speaker 1>selling point?

0:00:54.240 --> 0:00:56.800
<v Speaker 2>Well, we have a couple of things. We license our

0:00:56.800 --> 0:00:59.480
<v Speaker 2>IP and we're building a business on that for both

0:00:59.520 --> 0:01:03.680
<v Speaker 2>risk vice CPU and AI. We're selling low cost development

0:01:03.720 --> 0:01:06.600
<v Speaker 2>systems like I'm really interested in hitting a much lower

0:01:06.680 --> 0:01:09.959
<v Speaker 2>cost point. And then we open source to our whole

0:01:10.000 --> 0:01:13.720
<v Speaker 2>AI compiler stack and people like the transparency. They can

0:01:13.720 --> 0:01:17.080
<v Speaker 2>see exactly what the technology is doing. They actually it's

0:01:17.080 --> 0:01:20.440
<v Speaker 2>turned into a great recruiting vehicle for our software engineers.

0:01:21.680 --> 0:01:25.000
<v Speaker 1>Interesting. So let's just go first to the price points,

0:01:25.280 --> 0:01:27.680
<v Speaker 1>a cheap offering. A lot of that's done by having

0:01:27.720 --> 0:01:31.679
<v Speaker 1>less pricing components. For example, in videos, chips are incredibly expensive.

0:01:31.720 --> 0:01:34.600
<v Speaker 1>We know people pay it because of what they offer

0:01:34.760 --> 0:01:37.800
<v Speaker 1>the full stack in many ways, but they struggle from

0:01:37.840 --> 0:01:41.640
<v Speaker 1>a choke perspective on how foundwidth memory chips in particular,

0:01:41.760 --> 0:01:43.440
<v Speaker 1>you don't need them. Why don't you need them? Why

0:01:43.440 --> 0:01:44.400
<v Speaker 1>don't you want them?

0:01:44.840 --> 0:01:47.840
<v Speaker 2>Well, it's partly about choices. So there's two choices. One is,

0:01:48.200 --> 0:01:51.640
<v Speaker 2>if you pick a really expensive memory technology, you'll never

0:01:51.720 --> 0:01:54.920
<v Speaker 2>get the price down right, and so you need to

0:01:55.200 --> 0:01:58.880
<v Speaker 2>then make technology decisions to lower your costs. So we

0:01:59.120 --> 0:02:03.000
<v Speaker 2>have a tensor process or architecture, we keep the data

0:02:03.040 --> 0:02:06.640
<v Speaker 2>on chip. More, we have a software stack that's very

0:02:06.640 --> 0:02:10.960
<v Speaker 2>good at partitioning AI workloads across multiple chips, and then

0:02:11.000 --> 0:02:14.799
<v Speaker 2>they can send information and compute packets to each other

0:02:15.120 --> 0:02:18.240
<v Speaker 2>without having to go through memory, and that lowers the

0:02:18.240 --> 0:02:21.120
<v Speaker 2>memory VANDTH required and also lets it scale more naturally

0:02:21.160 --> 0:02:22.120
<v Speaker 2>across many chips.

0:02:22.760 --> 0:02:25.880
<v Speaker 1>So I'm a new investor. You brought me on board.

0:02:26.040 --> 0:02:29.400
<v Speaker 1>I'm interested Bezos Samsung. But you're saying, look, I'm going

0:02:29.440 --> 0:02:33.000
<v Speaker 1>to do this with more open architecture, with cheaper offerings.

0:02:33.040 --> 0:02:34.639
<v Speaker 1>How much of the market shed do you have to get?

0:02:34.680 --> 0:02:36.360
<v Speaker 1>What sort of price point are we looking at for

0:02:36.360 --> 0:02:37.160
<v Speaker 1>your chip offering?

0:02:38.760 --> 0:02:42.040
<v Speaker 2>Yeah, so this is really wild. The market is so big,

0:02:42.680 --> 0:02:44.840
<v Speaker 2>Like the percent we have to get is actually pretty

0:02:44.880 --> 0:02:47.839
<v Speaker 2>small to be a real value proposition. You know, next

0:02:47.880 --> 0:02:49.480
<v Speaker 2>year we'll die and go to heaven if we make

0:02:49.480 --> 0:02:52.280
<v Speaker 2>two hundred million dollars. We have already booked one hundred

0:02:52.280 --> 0:02:55.600
<v Speaker 2>and fifty million in revenue. I'm really interested in finding

0:02:55.760 --> 0:02:59.280
<v Speaker 2>small players, early players, people who want to own their technology.

0:03:00.040 --> 0:03:03.920
<v Speaker 2>People are excited about our software and hardware architecture, and

0:03:04.360 --> 0:03:06.960
<v Speaker 2>you know, I want to grow organically from there. I

0:03:07.000 --> 0:03:09.880
<v Speaker 2>think some people will focus on the big guys first.

0:03:10.400 --> 0:03:13.639
<v Speaker 2>That's not how new technology introductions go. You focus on

0:03:14.040 --> 0:03:17.680
<v Speaker 2>smaller players, people who need a differentiation, and that's what

0:03:17.720 --> 0:03:18.840
<v Speaker 2>we're really that's our mission.

0:03:19.240 --> 0:03:22.760
<v Speaker 1>So what therefore is the mission building in in terms

0:03:22.800 --> 0:03:26.400
<v Speaker 1>of long term value of generative AI of AI applications.

0:03:27.040 --> 0:03:28.480
<v Speaker 1>Do you think that we are in some sort of

0:03:28.560 --> 0:03:31.440
<v Speaker 1>hype cycle? Do you think that actually this market is

0:03:31.440 --> 0:03:33.040
<v Speaker 1>as big as manufacturing in.

0:03:34.000 --> 0:03:36.320
<v Speaker 2>Well, first of all, we are definitely in a hype cycle.

0:03:36.480 --> 0:03:39.280
<v Speaker 2>There's going to be multiple and people say, well, Jim,

0:03:39.360 --> 0:03:40.920
<v Speaker 2>you don't understand it's going to go down next year,

0:03:40.920 --> 0:03:42.000
<v Speaker 2>and then it's going to go up, and it's going

0:03:41.960 --> 0:03:45.440
<v Speaker 2>to down low pass filter all that stuff. Right, what's

0:03:45.480 --> 0:03:48.920
<v Speaker 2>really going to happen is AI is going to dominate

0:03:48.960 --> 0:03:52.280
<v Speaker 2>computing over the next ten years, and we're just starting

0:03:52.960 --> 0:03:55.120
<v Speaker 2>and the hardware is going to change a lot. The

0:03:55.160 --> 0:03:57.800
<v Speaker 2>software is going to change a lot. Like one day

0:03:57.840 --> 0:04:00.240
<v Speaker 2>you read in the papers lms can do anything, and

0:04:00.280 --> 0:04:03.360
<v Speaker 2>the next day you read they've hit a limit. Ignore

0:04:03.400 --> 0:04:06.160
<v Speaker 2>all that stuff. We're just starting this. There's going to

0:04:06.200 --> 0:04:09.600
<v Speaker 2>be massive transformations. So what I really wanted tense Torrent

0:04:09.880 --> 0:04:14.960
<v Speaker 2>to have is great computer designers, general purpose CPUs, AI processors,

0:04:15.320 --> 0:04:19.720
<v Speaker 2>chip design, system design, software design. And then the open

0:04:19.720 --> 0:04:22.440
<v Speaker 2>source stuff is really interesting because that engages a community.

0:04:22.480 --> 0:04:25.320
<v Speaker 2>People are really passionate about that to contribute back.

0:04:26.600 --> 0:04:29.960
<v Speaker 1>This is where you're thinking about your own talent team.

0:04:30.000 --> 0:04:32.559
<v Speaker 1>You've managed talent, as we said, sort of across the board.

0:04:33.040 --> 0:04:35.559
<v Speaker 1>You began like you were integral to the chip design

0:04:35.560 --> 0:04:37.440
<v Speaker 1>process in the US, and you think back to DC

0:04:37.680 --> 0:04:39.840
<v Speaker 1>and then the work that you then did an AMD,

0:04:40.279 --> 0:04:42.680
<v Speaker 1>moving to Apple, you were at Intel, and I just

0:04:42.720 --> 0:04:44.600
<v Speaker 1>have to go there on this day that Pat Gelsinger

0:04:44.640 --> 0:04:48.560
<v Speaker 1>hands in his well, hands his hat on off, he's

0:04:48.600 --> 0:04:51.000
<v Speaker 1>going to re resigning. Many would say it's because they

0:04:51.080 --> 0:04:53.960
<v Speaker 1>didn't get into AI quickly enough. That was prior to him.

0:04:54.200 --> 0:04:55.880
<v Speaker 1>What do you make of the transition.

0:04:56.839 --> 0:04:59.240
<v Speaker 2>Well, it's hard to say. So. I like Pat personally.

0:04:59.240 --> 0:05:01.760
<v Speaker 2>I've met him a number of times. He's a smart guy.

0:05:01.839 --> 0:05:06.719
<v Speaker 2>He really cares. I think the market focuses too much

0:05:06.720 --> 0:05:10.880
<v Speaker 2>on stuff like execution and AI transition. The way you

0:05:11.200 --> 0:05:14.600
<v Speaker 2>make money, get customers to build great products, like Steve

0:05:14.680 --> 0:05:17.600
<v Speaker 2>Jobs is the best at this, focus on the product,

0:05:17.760 --> 0:05:20.760
<v Speaker 2>make the best product you possibly can. And I think

0:05:21.000 --> 0:05:24.080
<v Speaker 2>they need somebody way more hands on, really focus on product,

0:05:24.600 --> 0:05:27.560
<v Speaker 2>and they need to stop talking about AI execution all

0:05:27.640 --> 0:05:30.440
<v Speaker 2>the nonsense. Right when they make the best product, the

0:05:30.440 --> 0:05:33.680
<v Speaker 2>best fab, the best chip, the best GP accelerator. They'll

0:05:33.720 --> 0:05:34.120
<v Speaker 2>do fine.

0:05:35.320 --> 0:05:38.200
<v Speaker 1>Jim. When you're thinking about all of the errors, you

0:05:38.279 --> 0:05:41.159
<v Speaker 1>want to own. How much does US and China factor

0:05:41.200 --> 0:05:42.039
<v Speaker 1>into your growth story?

0:05:44.320 --> 0:05:49.400
<v Speaker 2>So, first of all, we are building open source software,

0:05:49.760 --> 0:05:54.880
<v Speaker 2>risk five based technology. We think that's an open platform.

0:05:54.960 --> 0:05:57.320
<v Speaker 2>There's a lot of people interested in that. At the

0:05:57.360 --> 0:06:00.800
<v Speaker 2>same time, you know, there are you know, restrictions by

0:06:00.839 --> 0:06:03.320
<v Speaker 2>the US government and trade rules. We have really good

0:06:03.400 --> 0:06:06.440
<v Speaker 2>lawyers and trade compliance people. We work very closely with

0:06:06.480 --> 0:06:09.960
<v Speaker 2>those those guys. We want to do the right thing

0:06:10.080 --> 0:06:12.600
<v Speaker 2>both for the business and for you know, being compliant

0:06:12.600 --> 0:06:15.920
<v Speaker 2>to what's going on. And it's a changing world and

0:06:16.279 --> 0:06:20.800
<v Speaker 2>we're very involved in that. Short run, our business possibilities

0:06:20.839 --> 0:06:24.120
<v Speaker 2>are really big, and you know, if China plays in that,

0:06:24.120 --> 0:06:26.880
<v Speaker 2>that'll be interesting. If it doesn't for you know, various reasons,

0:06:26.880 --> 0:06:27.560
<v Speaker 2>that's fine.

0:06:28.480 --> 0:06:31.560
<v Speaker 1>Got a big evaluation come back when you're executing on

0:06:31.600 --> 0:06:34.080
<v Speaker 1>the latest round. Thank you, Jim Keller of ted Stone.

0:06:34.120 --> 0:06:35.320
<v Speaker 1>All right, great to have you.