WEBVTT - Palantir CTO Shyam Sankar Talks AI Demands & Company Transformation

0:00:00.080 --> 0:00:03.480
<v Speaker 1>Palenteer outward results, boosting its revenue guidance for the year

0:00:03.520 --> 0:00:05.920
<v Speaker 1>to arrange of two point seventy four billion to two

0:00:05.960 --> 0:00:09.680
<v Speaker 1>point seven five billion, ahead of estimates, with AI demand

0:00:09.760 --> 0:00:13.160
<v Speaker 1>also helping Palenteer boost its profit outlook. There's been growth

0:00:13.160 --> 0:00:17.239
<v Speaker 1>in business with government and in commercial customers. AIP Palenteers

0:00:17.600 --> 0:00:22.239
<v Speaker 1>Artificial intelligence platform and other products has quote transformed the

0:00:22.320 --> 0:00:26.040
<v Speaker 1>business in a level more than a year. CTO Sean

0:00:26.120 --> 0:00:31.040
<v Speaker 1>Sanka joins us to discuss. Sean on the call this

0:00:31.200 --> 0:00:34.280
<v Speaker 1>transformation that doctor cart was talking about. I've been to

0:00:34.360 --> 0:00:38.760
<v Speaker 1>a few AIP cons, but as CTO, you're working on it.

0:00:39.000 --> 0:00:41.640
<v Speaker 1>I think just explain the basics of that. What's changed

0:00:41.640 --> 0:00:45.600
<v Speaker 1>within Palenteer from a technology perspective and in the domain

0:00:45.640 --> 0:00:46.800
<v Speaker 1>of artificial intelligence.

0:00:48.320 --> 0:00:50.120
<v Speaker 2>Well, what you really see in the market is this

0:00:50.240 --> 0:00:54.640
<v Speaker 2>massive bottleneck between prototype being and production, and that happens

0:00:54.640 --> 0:00:57.560
<v Speaker 2>to be where AIP is most differentiated. And that differentiation

0:00:57.680 --> 0:01:01.160
<v Speaker 2>is built on a decade of deep technical investments, investments

0:01:01.160 --> 0:01:05.720
<v Speaker 2>like the ontology, the OSDK, the security and business primitives

0:01:05.760 --> 0:01:09.240
<v Speaker 2>that we built throughout the platforms like functions, actions, automations

0:01:09.480 --> 0:01:11.720
<v Speaker 2>in this pipeline that we have that's really focused on

0:01:11.760 --> 0:01:13.920
<v Speaker 2>addressing that. And I think what's changed.

0:01:13.840 --> 0:01:15.959
<v Speaker 3>As you've been to these AIP.

0:01:15.760 --> 0:01:19.160
<v Speaker 2>Cons is that the market now understands how severe that

0:01:19.200 --> 0:01:19.800
<v Speaker 2>bubblenek is.

0:01:19.840 --> 0:01:22.000
<v Speaker 3>I think your last guest was just talking about that.

0:01:22.480 --> 0:01:26.640
<v Speaker 2>Where is so easy to build a charismatic AI prototype.

0:01:26.680 --> 0:01:29.400
<v Speaker 2>That's about the amount of effort of building a PowerPoint slide,

0:01:29.480 --> 0:01:33.880
<v Speaker 2>But it's also that amount of utility. And unlike traditional

0:01:33.920 --> 0:01:38.280
<v Speaker 2>deterministic software, this kind of powerful stochastic genie that is LMS,

0:01:38.600 --> 0:01:40.959
<v Speaker 2>it requires a lot more work to get to production,

0:01:41.040 --> 0:01:43.399
<v Speaker 2>maybe ten to one hundred times as much work, and

0:01:43.440 --> 0:01:46.000
<v Speaker 2>that requires the tool chain that we've assembled with AIP.

0:01:47.880 --> 0:01:50.440
<v Speaker 1>I've had a few conversations with your CI his cop

0:01:50.440 --> 0:01:53.880
<v Speaker 1>about his frustration with the PowerPoint deck versus the reality

0:01:53.920 --> 0:01:54.920
<v Speaker 1>of shipping products.

0:01:55.040 --> 0:01:56.560
<v Speaker 3>We'll put that to one side for now.

0:01:56.600 --> 0:01:59.520
<v Speaker 1>Show. There was a heavy emphasis on working with the

0:01:59.560 --> 0:02:03.200
<v Speaker 1>military in the written materials and on the call. Could

0:02:03.240 --> 0:02:05.360
<v Speaker 1>you just get the basics of what you'll work with

0:02:05.400 --> 0:02:07.360
<v Speaker 1>the military looks like present day.

0:02:08.480 --> 0:02:10.400
<v Speaker 2>Look In the commercial world, you call it a value

0:02:10.440 --> 0:02:12.200
<v Speaker 2>chain from the hand of your supplier to the hand

0:02:12.240 --> 0:02:14.280
<v Speaker 2>of your customer. In the military, it's about a kill

0:02:14.360 --> 0:02:16.799
<v Speaker 2>chain from censor to shooter. But really at the most

0:02:16.840 --> 0:02:19.680
<v Speaker 2>abstract level, it's the same thing. We're trying to enable

0:02:19.680 --> 0:02:23.000
<v Speaker 2>our war fighters to have information, dominance and decision advantage.

0:02:23.160 --> 0:02:25.160
<v Speaker 2>How can I see you know, to quote Sun Sue,

0:02:25.840 --> 0:02:27.440
<v Speaker 2>if you know your enemy and you know yourself, you're

0:02:27.440 --> 0:02:29.480
<v Speaker 2>going to win, and so you know, how can I

0:02:29.480 --> 0:02:32.000
<v Speaker 2>see everything there is to know about the threat? How

0:02:32.000 --> 0:02:34.440
<v Speaker 2>can I understand everything I have to combat and deter

0:02:34.880 --> 0:02:36.760
<v Speaker 2>any aggression from those threats? And when you look at

0:02:36.800 --> 0:02:40.000
<v Speaker 2>the geopolitical landscape right now, it could not be more dangerous.

0:02:40.000 --> 0:02:43.040
<v Speaker 2>Everything that's going on in Eastern Europe, the massive tensions

0:02:43.040 --> 0:02:45.480
<v Speaker 2>that are that exist in the Middle East, and the

0:02:46.080 --> 0:02:48.960
<v Speaker 2>ongoing issues that we have and deturing aggression in the Pacific.

0:02:50.880 --> 0:02:55.320
<v Speaker 1>Beyond the importance of data, where does plant to sit

0:02:55.360 --> 0:02:58.200
<v Speaker 1>in the defense ecosystem or even the defense supply chain.

0:02:58.440 --> 0:03:01.960
<v Speaker 1>Are you sort of very closely aligned with hardware makers,

0:03:02.000 --> 0:03:05.240
<v Speaker 1>the aerospace community or it is the MOLTI more focused

0:03:05.280 --> 0:03:08.280
<v Speaker 1>on you just going direct to different arms of the

0:03:08.280 --> 0:03:09.560
<v Speaker 1>defense base with government.

0:03:10.760 --> 0:03:12.320
<v Speaker 2>Yeah, we're touching all of it. One of the things

0:03:12.320 --> 0:03:14.440
<v Speaker 2>I'm most excited about. I've been calling the first Breakfast

0:03:14.520 --> 0:03:17.160
<v Speaker 2>as an antidote to the nineteen ninety three last Supper

0:03:17.200 --> 0:03:19.840
<v Speaker 2>that led to the consolidation of our defense industrial base,

0:03:19.840 --> 0:03:22.200
<v Speaker 2>where it went from fifty one primes down to five,

0:03:22.320 --> 0:03:24.880
<v Speaker 2>because we forget that at the dawn of World War Two,

0:03:24.960 --> 0:03:26.799
<v Speaker 2>we didn't have a defense industrial base.

0:03:26.840 --> 0:03:28.640
<v Speaker 3>We had an American industrial base.

0:03:29.000 --> 0:03:32.600
<v Speaker 2>Chrysler made missiles, General mills, this Serial company made inertial

0:03:32.639 --> 0:03:35.440
<v Speaker 2>guidance systems. And so we have this moment right now

0:03:35.560 --> 0:03:38.640
<v Speaker 2>in defense tech where one hundred billion dollars or more

0:03:38.640 --> 0:03:41.280
<v Speaker 2>of capital has been deployed, a boluss of founders have

0:03:41.320 --> 0:03:43.080
<v Speaker 2>shown up. There's a lot of creativity, a lot of

0:03:43.200 --> 0:03:46.520
<v Speaker 2>energy as a company that's kind of been pathfinding over

0:03:46.560 --> 0:03:49.360
<v Speaker 2>twenty years. Not only have we developed our software that

0:03:49.400 --> 0:03:53.400
<v Speaker 2>gives these warfighters unique advantage like the Maven contract that

0:03:53.520 --> 0:03:56.240
<v Speaker 2>CDO just recently awarded for nearly half a billion dollars,

0:03:56.520 --> 0:04:00.800
<v Speaker 2>but really this software infrastructure that's required to deliver modern

0:04:00.800 --> 0:04:05.040
<v Speaker 2>American software to the battlefield in air gapped environments at speed,

0:04:05.080 --> 0:04:06.240
<v Speaker 2>at pace, well.

0:04:06.080 --> 0:04:08.000
<v Speaker 3>Ahead of the threats of twenty twenty seven.

0:04:08.160 --> 0:04:11.640
<v Speaker 2>So we've been working very closely with both traditional primes,

0:04:11.720 --> 0:04:14.960
<v Speaker 2>integrating our software to their hardware, helping them actually with production.

0:04:15.080 --> 0:04:16.760
<v Speaker 2>If you think about the fifty percent of our business

0:04:16.760 --> 0:04:19.240
<v Speaker 2>that's commercial oriented. How do we build jet engines and

0:04:19.240 --> 0:04:21.040
<v Speaker 2>satellites faster, better, cheaper?

0:04:22.240 --> 0:04:25.080
<v Speaker 3>In addition to the new entrants who need to achieve.

0:04:24.800 --> 0:04:29.400
<v Speaker 2>Scale and time, value of money is everything for them.

0:04:29.480 --> 0:04:32.880
<v Speaker 1>Whenever we have an executive on this program, I always

0:04:32.880 --> 0:04:34.960
<v Speaker 1>go to our audience and say, you know, what is

0:04:35.000 --> 0:04:37.680
<v Speaker 1>it that you want to know from? In this case, Palenteer,

0:04:38.120 --> 0:04:40.640
<v Speaker 1>I would say most of the questions were about warp speed,

0:04:42.080 --> 0:04:46.000
<v Speaker 1>very basic ones. Why now with warp speed and the

0:04:46.040 --> 0:04:49.440
<v Speaker 1>obstacles to rolling it out making it sort of more

0:04:50.440 --> 0:04:52.880
<v Speaker 1>readily available, pervasive out their shop.

0:04:53.839 --> 0:04:54.720
<v Speaker 3>Yeah, that's great.

0:04:54.800 --> 0:04:56.760
<v Speaker 2>I'm so excited about Warpspeed is what I'm spending all

0:04:56.760 --> 0:04:58.360
<v Speaker 2>of my time on and really shaping the R and

0:04:58.440 --> 0:05:02.599
<v Speaker 2>D roadmap around. Speed is our modern American operating system

0:05:02.600 --> 0:05:05.760
<v Speaker 2>for manufacturing. And the reason why now, you know, for

0:05:05.800 --> 0:05:08.400
<v Speaker 2>the better part of twenty years we have helped traditional

0:05:08.440 --> 0:05:11.160
<v Speaker 2>manufacturers build planes, trains, automobiles, and ships.

0:05:11.960 --> 0:05:13.360
<v Speaker 3>But most of those folks are.

0:05:13.240 --> 0:05:15.560
<v Speaker 2>Stuck in a legacy mode of how they're operating, and

0:05:15.600 --> 0:05:18.159
<v Speaker 2>you're able to help here or there. But what's unique

0:05:18.240 --> 0:05:21.840
<v Speaker 2>about the reindustrialization movement that's happening right now in America

0:05:21.960 --> 0:05:25.960
<v Speaker 2>is that these these founders, they're alumni of Palenteer, of Tesla,

0:05:26.000 --> 0:05:30.640
<v Speaker 2>of SpaceX, and they understand that the traditional erp plm

0:05:30.760 --> 0:05:34.480
<v Speaker 2>PLC software doesn't really work. That most of these successful

0:05:34.520 --> 0:05:36.760
<v Speaker 2>companies have had to build their own software and that

0:05:36.880 --> 0:05:39.360
<v Speaker 2>is really an unaffordable journey. So that there's this massive

0:05:39.400 --> 0:05:43.279
<v Speaker 2>opportunity to take the power of AIP and the historical

0:05:43.279 --> 0:05:46.800
<v Speaker 2>experiences we have throughout the value chain of production to

0:05:46.920 --> 0:05:50.400
<v Speaker 2>help our customers bend their atoms better with bits.

0:05:52.360 --> 0:05:53.919
<v Speaker 1>I have you know, Ceto, I kind of have some

0:05:54.000 --> 0:05:56.719
<v Speaker 1>quick fire questions that I've never been able to answer

0:05:56.720 --> 0:05:59.919
<v Speaker 1>about Palenteer in the first sUAS and since AIP and

0:06:00.080 --> 0:06:03.400
<v Speaker 1>warp speed over the year. What's the kind of key

0:06:03.480 --> 0:06:06.440
<v Speaker 1>foundational model LLLM that you've been building on top of.

0:06:06.520 --> 0:06:08.680
<v Speaker 1>I know you partner as well as sort of think

0:06:08.760 --> 0:06:13.039
<v Speaker 1>in house about the model or foundation level, but that's

0:06:13.040 --> 0:06:14.800
<v Speaker 1>a question that comes up quite a lot. Who are

0:06:14.839 --> 0:06:16.560
<v Speaker 1>you building on top of and working with?

0:06:17.920 --> 0:06:19.800
<v Speaker 2>Well, I think all the value is really going to

0:06:19.839 --> 0:06:21.880
<v Speaker 2>create at the application layer, and what we've seen in

0:06:21.920 --> 0:06:24.560
<v Speaker 2>production is that you actually need a menagerie of models

0:06:24.839 --> 0:06:26.360
<v Speaker 2>and you you know, if you think about the most

0:06:26.400 --> 0:06:29.799
<v Speaker 2>expensive frontier model out there right now, it's a thousand

0:06:29.839 --> 0:06:33.680
<v Speaker 2>times more expensive than the cheapest open source model for

0:06:33.839 --> 0:06:36.560
<v Speaker 2>ten percent more ELO ten percent more IQ. As a

0:06:36.640 --> 0:06:39.520
<v Speaker 2>rough proxy, that's not a compelling price performance trade off.

0:06:39.839 --> 0:06:42.279
<v Speaker 2>You already see that the first versions of GPT four

0:06:42.320 --> 0:06:44.680
<v Speaker 2>have been sunset. You know, we need to think about

0:06:44.680 --> 0:06:47.599
<v Speaker 2>this problem as Okay, what is the infrastructure I need

0:06:47.839 --> 0:06:50.320
<v Speaker 2>to make I'm turning my software into something that is

0:06:50.360 --> 0:06:51.200
<v Speaker 2>now stochastic.

0:06:51.360 --> 0:06:52.880
<v Speaker 3>It's not deterministic anymore.

0:06:53.240 --> 0:06:55.719
<v Speaker 2>Computer scientists that we have trained, they are all used

0:06:55.720 --> 0:06:57.360
<v Speaker 2>to writing deterministic code.

0:06:57.640 --> 0:06:57.880
<v Speaker 3>You know.

0:06:58.000 --> 0:07:00.560
<v Speaker 2>This is this harkens more to how we think about

0:07:00.680 --> 0:07:03.480
<v Speaker 2>the transition from analog circuits to digital circuits, the sort

0:07:03.480 --> 0:07:05.640
<v Speaker 2>of error correction you need, the sort of infrastructure you

0:07:05.640 --> 0:07:07.680
<v Speaker 2>need to think about having so that you can have

0:07:07.760 --> 0:07:08.640
<v Speaker 2>the abstraction to.

0:07:08.600 --> 0:07:11.160
<v Speaker 3>Think about these things as being digital. That's where I

0:07:11.160 --> 0:07:13.920
<v Speaker 3>think the colony is going to be go ahead. Sorry,

0:07:15.760 --> 0:07:16.720
<v Speaker 3>And so we're very.

0:07:16.640 --> 0:07:19.120
<v Speaker 2>Focused on helping our customers get the right models for

0:07:19.160 --> 0:07:23.120
<v Speaker 2>the right use cases, auto evaluate that at automatically generate

0:07:23.720 --> 0:07:26.240
<v Speaker 2>iterations on the prompts that get them there. And part

0:07:26.240 --> 0:07:28.520
<v Speaker 2>of our theory is that really prompts are for developers,

0:07:28.600 --> 0:07:31.400
<v Speaker 2>chat is a dead end. We're guiding our customers through

0:07:31.400 --> 0:07:33.560
<v Speaker 2>this journey here. They get there pretty quickly to realize

0:07:33.560 --> 0:07:35.480
<v Speaker 2>that we should be thinking of llm's as a new

0:07:35.520 --> 0:07:37.200
<v Speaker 2>type of run time the way I might write a

0:07:37.200 --> 0:07:37.960
<v Speaker 2>Python function.

0:07:38.280 --> 0:07:40.240
<v Speaker 3>Well, okay, I'm going to write an LEM function too.

0:07:41.640 --> 0:07:44.080
<v Speaker 1>Sean, for the commercial half of the business, is there

0:07:44.120 --> 0:07:47.760
<v Speaker 1>a specific Hyperscale cloud platform or partner that you work

0:07:47.800 --> 0:07:49.080
<v Speaker 1>with to support its growth.

0:07:50.640 --> 0:07:52.640
<v Speaker 2>We're working with all of them. We have deep and

0:07:52.720 --> 0:07:56.120
<v Speaker 2>very valuable relationships there. So we're very happy with that.

0:07:57.920 --> 0:08:01.120
<v Speaker 1>This last year has been about a IP and growth.

0:08:01.160 --> 0:08:03.840
<v Speaker 1>What's the next twelve months like for Palenter Schaump.

0:08:04.640 --> 0:08:07.040
<v Speaker 2>Well, I think it's really about deepening the investments that

0:08:07.080 --> 0:08:10.280
<v Speaker 2>we have with AIP that are dressing this bottleneck between

0:08:10.280 --> 0:08:12.920
<v Speaker 2>prototyping and production. You know, if you look at David

0:08:12.960 --> 0:08:15.680
<v Speaker 2>Kahn at Sequoia's article on the six hundred billion dollar

0:08:16.000 --> 0:08:17.320
<v Speaker 2>whole and revenue.

0:08:16.920 --> 0:08:18.000
<v Speaker 3>I think this is where the whole is.

0:08:18.080 --> 0:08:20.040
<v Speaker 2>This is the bottleneck in the market is the most

0:08:20.080 --> 0:08:22.680
<v Speaker 2>important problem to solve, and the folks who solve at first,

0:08:22.680 --> 0:08:24.480
<v Speaker 2>and I think we're in the pole position there, have

0:08:24.560 --> 0:08:26.240
<v Speaker 2>the opportunity to take the entire market.

0:08:27.560 --> 0:08:30.160
<v Speaker 1>Sham Sangkas, CTO of Palenteer, it's great to have you

0:08:30.200 --> 0:08:31.480
<v Speaker 1>back on Bloomberg Technology.

0:08:31.520 --> 0:08:32.200
<v Speaker 3>Thank you so much,