WEBVTT - Cerebras CEO Andrew Feldman Talks Earnings

0:00:02.520 --> 0:00:07.040
<v Speaker 1>Bloomberg Audio Studios, podcasts, radio news.

0:00:10.640 --> 0:00:13.800
<v Speaker 2>Cerebres reported quarterly earnings for the first time since going

0:00:13.840 --> 0:00:17.600
<v Speaker 2>public last month. Its sales outlook beat Wall Street estimates

0:00:17.600 --> 0:00:20.919
<v Speaker 2>but still disappointed investors hoping to see the company carve

0:00:20.960 --> 0:00:24.840
<v Speaker 2>out a bigger slice of the aident center market. Right now,

0:00:24.840 --> 0:00:28.400
<v Speaker 2>shares down seventeen point five percent, its biggest drop in

0:00:28.440 --> 0:00:31.680
<v Speaker 2>its quite short history as a public company. What's behind that?

0:00:31.760 --> 0:00:35.159
<v Speaker 2>CEO Andrew Feldman's with us and welcome back to Bloomberg Tech. Andrew,

0:00:35.320 --> 0:00:36.800
<v Speaker 2>you know, there was a time where we would talk

0:00:36.840 --> 0:00:40.959
<v Speaker 2>about the merits of top to tail server ownership, how

0:00:41.040 --> 0:00:42.600
<v Speaker 2>owning all of the content, and now we're going to

0:00:42.600 --> 0:00:45.160
<v Speaker 2>talk about margin contraction and we're going to talk about

0:00:45.200 --> 0:00:48.519
<v Speaker 2>the stock being down seventeen percent. That to me is

0:00:48.800 --> 0:00:51.400
<v Speaker 2>kind of the mismatch that the outlook on the sales

0:00:51.479 --> 0:00:54.040
<v Speaker 2>side beat Wall Street estimates. I think a lot of

0:00:54.040 --> 0:00:58.240
<v Speaker 2>people are trying to understand the sequential margin decline and

0:00:58.280 --> 0:01:01.520
<v Speaker 2>for me, this is about ramping output for two big customers.

0:01:01.920 --> 0:01:03.600
<v Speaker 3>Is that true? Yeah?

0:01:04.120 --> 0:01:06.600
<v Speaker 1>I think what we did is we put forward a

0:01:06.640 --> 0:01:09.760
<v Speaker 1>plan in the start of twenty six, we shared it

0:01:09.760 --> 0:01:11.840
<v Speaker 1>with investors as we went public.

0:01:12.240 --> 0:01:13.520
<v Speaker 3>And we're ahead of plan.

0:01:14.040 --> 0:01:16.360
<v Speaker 1>You know, we delivered record revenues of one hundred and

0:01:16.400 --> 0:01:19.399
<v Speaker 1>ninety one million of ninety two percent year over year,

0:01:20.160 --> 0:01:22.720
<v Speaker 1>and for our cloud business, you know, it was up

0:01:22.720 --> 0:01:25.840
<v Speaker 1>one hundred and sixty seven percent year over year. We

0:01:26.600 --> 0:01:32.920
<v Speaker 1>beat margin consensus substantially, and then we guided for full

0:01:33.000 --> 0:01:36.440
<v Speaker 1>year the growth margins would be ten percent better than plan.

0:01:37.400 --> 0:01:40.520
<v Speaker 1>We also shared the in Q two and Q three

0:01:41.360 --> 0:01:45.280
<v Speaker 1>we would go back to some of our customers and

0:01:45.319 --> 0:01:47.400
<v Speaker 1>we would rent back gear that we'd sold them to

0:01:47.400 --> 0:01:49.440
<v Speaker 1>try and keep up with demand, and then that would

0:01:49.480 --> 0:01:52.120
<v Speaker 1>have a margin impact on the order ten or fifteen points.

0:01:53.040 --> 0:01:57.360
<v Speaker 1>We did that to keep our customers close, to be

0:01:57.440 --> 0:02:00.120
<v Speaker 1>sure we could keep up with their extraordinary demand and

0:02:00.440 --> 0:02:06.280
<v Speaker 1>for our product for fast inference, and so that was

0:02:06.880 --> 0:02:07.880
<v Speaker 1>the story.

0:02:09.919 --> 0:02:13.000
<v Speaker 3>On every metric we put out, we're ahead of plan.

0:02:14.360 --> 0:02:16.520
<v Speaker 2>Have the proceeds from the IPO actually allowed you to

0:02:16.520 --> 0:02:19.720
<v Speaker 2>move more quickly in ramping up capacity?

0:02:20.320 --> 0:02:25.399
<v Speaker 1>Yeah, I think capacity is the largest constraint right now

0:02:25.440 --> 0:02:32.400
<v Speaker 1>for everyone. Data centers are and we've significantly increased our

0:02:32.800 --> 0:02:36.120
<v Speaker 1>our ability and our pipeline for data centers which is

0:02:36.200 --> 0:02:42.760
<v Speaker 1>now very large. You know, we announced a data center

0:02:42.960 --> 0:02:46.279
<v Speaker 1>partnership with Bell, Canada for one hundred and twenty megawatts

0:02:46.880 --> 0:02:50.640
<v Speaker 1>that will be delivered in twenty twenty seven. We are

0:02:50.680 --> 0:02:55.359
<v Speaker 1>pursuing data centers across the US, in Canada, in Europe,

0:02:55.800 --> 0:02:56.520
<v Speaker 1>in the Middle East.

0:02:57.560 --> 0:02:58.239
<v Speaker 3>There are.

0:02:59.840 --> 0:03:04.240
<v Speaker 1>The The vast resources that we have now at our

0:03:04.240 --> 0:03:07.680
<v Speaker 1>disposal give us tremendous advantage in the pursuit of this

0:03:07.800 --> 0:03:09.400
<v Speaker 1>The limiting factor data centers.

0:03:11.240 --> 0:03:14.600
<v Speaker 2>What you're talking about, like matter of factly, is buildings

0:03:14.919 --> 0:03:17.679
<v Speaker 2>not necessarily the compute, right, It's not what you guys

0:03:17.720 --> 0:03:20.760
<v Speaker 2>are offering. How difficult right now is it is to

0:03:20.880 --> 0:03:24.920
<v Speaker 2>get moving in America or other markets, to get planning approval,

0:03:25.120 --> 0:03:27.280
<v Speaker 2>get the concrete, get the labor, get the thing built.

0:03:28.440 --> 0:03:31.480
<v Speaker 1>Now that that's the irony of this market, that the

0:03:32.880 --> 0:03:37.360
<v Speaker 1>AI market is moving at blistering speed, and we are

0:03:37.400 --> 0:03:40.320
<v Speaker 1>being constrained by data centers which move with the speed

0:03:40.320 --> 0:03:44.000
<v Speaker 1>of real estate. And so that is a problem that

0:03:44.120 --> 0:03:47.000
<v Speaker 1>is being confronted by everybody in the category.

0:03:49.000 --> 0:03:50.920
<v Speaker 3>By the hyperscalers, by the neo.

0:03:50.760 --> 0:03:55.120
<v Speaker 1>Clouds, by the new generation clouds. Everybody is confronting this

0:03:55.200 --> 0:03:57.200
<v Speaker 1>similar problem.

0:03:57.400 --> 0:04:01.640
<v Speaker 2>Andrew Cerebris does not rely on traditional off chip HBM.

0:04:02.080 --> 0:04:05.080
<v Speaker 2>Would you just explain that the basics of the technology,

0:04:05.280 --> 0:04:08.440
<v Speaker 2>but how insulated are you from the memory bottleneck that

0:04:08.480 --> 0:04:09.600
<v Speaker 2>others are experiencing.

0:04:10.600 --> 0:04:12.680
<v Speaker 3>Yeah, that's a really good point.

0:04:12.720 --> 0:04:16.680
<v Speaker 1>Because of our innovative architecture, because of our wayfer scale approach,

0:04:17.400 --> 0:04:21.800
<v Speaker 1>we don't use HBM. HBM is a type of DRAM

0:04:21.839 --> 0:04:23.919
<v Speaker 1>and it's made by three companies, one of whom is

0:04:23.960 --> 0:04:28.919
<v Speaker 1>reporting shortly right, that's Micron, Heenex and Samsung. There's a

0:04:28.960 --> 0:04:32.800
<v Speaker 1>global shortage, it's extremely expensive, lead times are long, and

0:04:32.880 --> 0:04:36.000
<v Speaker 1>we don't use it, so we have a tremendous advantage there.

0:04:36.800 --> 0:04:42.840
<v Speaker 1>The other constraints in the supply chain for many are cooths,

0:04:42.880 --> 0:04:45.120
<v Speaker 1>which is a process inside of TSMC.

0:04:45.440 --> 0:04:46.600
<v Speaker 3>Again we don't use it.

0:04:47.520 --> 0:04:51.479
<v Speaker 1>And the third is a capacity at the three nanometer node.

0:04:51.720 --> 0:04:56.720
<v Speaker 1>That's space in TSMC's factory that makes three and nanometer chips.

0:04:56.760 --> 0:05:00.160
<v Speaker 1>Again we don't use it. We're at the five nanometer node.

0:05:00.440 --> 0:05:03.720
<v Speaker 1>So our architecture has allowed us to deliver the fastest

0:05:03.760 --> 0:05:06.920
<v Speaker 1>inference in the world by an order of magnitude while

0:05:06.960 --> 0:05:14.520
<v Speaker 1>avoiding the main supply supply chain constraints faced by others

0:05:14.520 --> 0:05:15.040
<v Speaker 1>in the field.

0:05:16.120 --> 0:05:19.080
<v Speaker 2>Can you say, hand on heart, not just winning business,

0:05:19.080 --> 0:05:20.800
<v Speaker 2>but have you actually been able to go to a

0:05:20.839 --> 0:05:23.760
<v Speaker 2>customer and say we can get this compute online faster

0:05:23.800 --> 0:05:26.520
<v Speaker 2>than others for those reasons you just outlined and then

0:05:26.560 --> 0:05:27.640
<v Speaker 2>actually gone and done it.

0:05:28.279 --> 0:05:28.919
<v Speaker 3>Oh for sure.

0:05:29.880 --> 0:05:33.040
<v Speaker 1>I mean in any case I signed, we signed. For example,

0:05:33.040 --> 0:05:39.200
<v Speaker 1>we signed our contract with open Ai on December twenty

0:05:39.279 --> 0:05:44.520
<v Speaker 1>fourth and had them in for production on February first.

0:05:45.960 --> 0:05:46.839
<v Speaker 1>That's unheard of.

0:05:46.839 --> 0:05:47.600
<v Speaker 3>Stay, that's quick.

0:05:48.440 --> 0:05:51.600
<v Speaker 2>I went really quickly, Andrew this morning. Open Ai is

0:05:51.640 --> 0:05:57.479
<v Speaker 2>out with Kalapino intelligence processor. But everyone is compute constrained, right,

0:05:57.520 --> 0:06:00.839
<v Speaker 2>How do you interpret the Frontier Labs going to custom

0:06:00.880 --> 0:06:04.200
<v Speaker 2>a six alongside their other compute options.

0:06:04.839 --> 0:06:06.520
<v Speaker 1>And one of the things we've been saying all along

0:06:06.640 --> 0:06:10.599
<v Speaker 1>is that this market is enormous and is going to

0:06:10.640 --> 0:06:16.400
<v Speaker 1>be met with a heterogeneous collection of architectures for hardware.

0:06:16.440 --> 0:06:20.400
<v Speaker 1>This market will not be consolidated around GPUs. There will

0:06:20.440 --> 0:06:23.360
<v Speaker 1>be a six, there will be a six from Hyperscalers,

0:06:23.360 --> 0:06:26.920
<v Speaker 1>there will be a six from from Labs, and then

0:06:26.960 --> 0:06:32.080
<v Speaker 1>there'll be companies like Cerebras with pioneering architectures who all

0:06:32.120 --> 0:06:35.040
<v Speaker 1>of us will will will take big bites of this

0:06:35.279 --> 0:06:38.120
<v Speaker 1>enormous market. I think one of the things that's difficult

0:06:38.160 --> 0:06:40.800
<v Speaker 1>to get your head around is just how big the

0:06:40.880 --> 0:06:43.520
<v Speaker 1>compute market is right now that One of the things

0:06:43.560 --> 0:06:48.760
<v Speaker 1>AI does is it makes tractable for compute much of

0:06:48.800 --> 0:06:51.320
<v Speaker 1>the world around us, and that really wasn't the case

0:06:51.360 --> 0:06:52.039
<v Speaker 1>prior to AI.

0:06:53.839 --> 0:06:56.640
<v Speaker 2>Andrew's if the world is simple and it's money that

0:06:56.720 --> 0:06:59.640
<v Speaker 2>greases the wheels to get your industry going, can we

0:06:59.760 --> 0:07:02.320
<v Speaker 2>see you come back to capital markets in some form

0:07:02.440 --> 0:07:03.520
<v Speaker 2>just so you can move quicker.

0:07:04.839 --> 0:07:07.279
<v Speaker 1>We have more than nine billion on the balance sheet,

0:07:07.480 --> 0:07:12.840
<v Speaker 1>I think, so we're really pleased with our position there.

0:07:12.880 --> 0:07:17.240
<v Speaker 1>But of course we are always scanning both the capital markets,

0:07:17.560 --> 0:07:21.040
<v Speaker 1>both equity and debt for ways to accelerate our growth.

0:07:23.240 --> 0:07:26.320
<v Speaker 2>Finally, how are you judging success yourself? What is the

0:07:26.320 --> 0:07:29.640
<v Speaker 2>milestone that you'd pitch to the market to keep closest attention.

0:07:29.400 --> 0:07:33.440
<v Speaker 1>To Well, I think I think if my mother's proud

0:07:33.440 --> 0:07:37.160
<v Speaker 1>of me, I think that's the biggest thing you can

0:07:37.280 --> 0:07:42.800
<v Speaker 1>you can ask for. I think for markets, I think

0:07:43.160 --> 0:07:46.720
<v Speaker 1>when you lay out a plan that is aggressive and

0:07:46.880 --> 0:07:50.120
<v Speaker 1>you crush it, Uh, that's how you feel good about

0:07:50.520 --> 0:07:53.800
<v Speaker 1>both your ability to execute in your ability to predict

0:07:53.840 --> 0:07:59.320
<v Speaker 1>your own execution. And so you know, we announced yesterday

0:07:59.360 --> 0:08:02.760
<v Speaker 1>that we would beat our full year We gave full

0:08:02.840 --> 0:08:06.520
<v Speaker 1>year guidance that was ten gross margin points above consensus.

0:08:08.200 --> 0:08:10.000
<v Speaker 1>You ought to be proud of that, and we got

0:08:10.040 --> 0:08:13.200
<v Speaker 1>to continue to execute and continue to set extraordinarily high

0:08:13.280 --> 0:08:15.360
<v Speaker 1>bars and then continue to.

0:08:17.120 --> 0:08:19.160
<v Speaker 3>Beat them.

0:08:19.360 --> 0:08:22.440
<v Speaker 2>Cerebrasy Andrew Feldman, thank you very much. Indeed,