Hi everybody, this is Olivia Ulier live from Promenade 69 at Davos, Switzerland. We're in the Qualcomm House and as part of the AI summit of the official USA house, we are here to have conversations with leaders, with tech leaders, and it's my pleasure to welcome Andrew Feldman, CEO of Cerebras Systems. Andrew, thanks for having me. Thank you.
So, I heard it's your first Davos. How does that feel?
What an interesting place. It's sort of an outdoor trade show, right? Where the whole town has been converted into trade show booths, right? And there are interesting little talks and big talks and the people you meet, there's something to learn from, there's something to share. I've never been to anything like it.
It's overwhelming the first time, isn't it?
But before we dive into what you are doing with a company in Davos, I think for our global audience, it would be good to hear about what Cerebras Systems is doing, the story of the company and what your products are, what you're developing.
Sure. We build fast AI computers. So computers optimized for doing AI work and as a result they're the fastest in the world at AI work by not a little bit but by 10 or 20 or sometimes 30x over GPUs and what we did is we built...
Oh, you brought some tech in hardware, we have good show and tell.
And so what we did is we solved a problem that had been open in the computer industry for 75 years, it's how to build a very big chip.
And for a lot of work, you don't want a big chip. You don't want a big chip for a cell phone. You don't want a big chip for a car. But for AI work, big chips process information more quickly, and they deliver answers in less time. And so by solving this problem, we were able to build systems based on this chip, process AI queries, both inference and training much faster, and deliver extraordinary results to our customers.
This is my fifth startup. So I am a professional David in the battle with Goliaths.
You still consider Cerebras Systems as a startup.
We're 750 people. We have offices in the US, in Toronto, in the United Arab Emirates, in Bangalore, India. Yeah, I think we compete with Nvidia and they were one of the most valuable companies in the world and compared to them we are very much David fighting with Goliath.
Why did you decide to tackle that particular issue with this fifth startup?
So in 2015 the founders got together, that was Gary and Michael and JP and Sean and I and what we saw on the horizon was a new type of work.
And what excites computer architects is a problem that is different enough so that the existing computer architectures, the existing machines aren't very good at it. And then you ask yourself, can I build a machine that could make this work done in less time? Could we build a machine that's faster? And then you ask yourself the question, is the market big enough to care? And what we saw was that for AI, we came to believe that the existing GPU architecture had a set of fundamental limitations and that we could in fact build a chip architecture that would be orders of magnitude faster. And to do that we had to solve a problem that had been bedeviling the industry for three quarters of a century. How to build big chips and we've solved that problem and now we're delivering and we have an extraordinary collection of customers including OpenAI including...
Yeah, well there was some big announcement recently right with OpenAI.
Yeah, we announced together with OpenAI that they'd be doing fast inference with us, that we would be serving their inference work. And that was an opportunity that was reported in the tens of billions of dollars.
For the average user of ChatGPT, it delivers with quite good speed. The outcome is satisfactory or not. That depends on how you ask, etc. I don't want to go there. But what difference does it make for the end user that now you're partnering with OpenAI?
Well, I think what you will see is as AI moves from being a novelty to being a necessity, as it becomes part of a workflow that companies and individual developers use, the willingness to wait, the willingness to wait 10, 15, 20 seconds or for deep research 10, 15 minutes shrinks to zero. Nobody wants to wait. And so when we can deliver real time results, it will be embedded in the way you work. Remember years ago, Google learned that even millisecond delays...
Right. Cause them to lose the attention of their user. And it's the same for AI.
And create frustration and then that's what we... less engaged.
That's right. So with OpenAI, you got a product that basically very soon is going to reach a billion users.
When you start a company, that's your dream that you build a technology and a billion people use it on a given week or I mean that's why you start a company.
Yeah, I think that's a good definition of impact.
Can you share about other customers or partners that you have please?
Sure. We have a big partnership with G42, which is the UAE's national champion. We train models with them. We do inference with them across a range of use cases. We have a big partnership with GlaxoSmithKline where we train models and we do inference for GlaxoSmithKline. We have a big...
What we're trying to do there is genomic work, to predict efficacy of drugs based on genomic information and epigenomic information and accelerating drug development, improving patient outcomes and healthcare delivery.
That's the dream, right? I mean, healthcare in the US is 17% of the economy and we do a terrible job of it, right? It takes a long time to make drugs. Our delivery systems are weak. You know, if you're wealthy, there's great health care. If you're not wealthy, it's not so great. We can do better and I think AI is one of the real levers we have.
Oh, to improve healthcare delivery. I mean, the estimate is about 30% of the money invested in healthcare globally that is wasted. I used to run health and healthcare at the World Economic Forum and we're working and we launched a public private partnership on value based healthcare, the biggest in the world and the fact that we can track that, we can process that, we can improve patient outcome and also remove some of the misalignment thanks to a technology like yours is massive in terms of impact, it's going to save lives.
And how easy it is to motivate your team when that's what they can do, right? I mean that's what we want, when you start companies, when you hire engineers, what we want is to have an impact, we want the fruits of our ideas to impact the world and we want it for the better. And so everybody's fired up on those projects.
We've worked with Mayo Clinic. We have customers in the US government both for scientific work. We did some of the really early work in understanding genetic mutations in the coronavirus at Argonne National Laboratory. We have one of the leading coding AI companies called Cognition and they use our equipment.
And we're going to talk about academia very soon but it's very important what you mentioned that your staff, your team can see the impact of what they do. Reminds me, I think it's Bill George when he was the CEO of Medtronic once a week would bring a former patient or someone whose lives has been improved by what the team was doing and so that people could literally see the impact of their work.
I think that's really central to inspiring the engineering community. I think there are different ways to inspire people. I think engineers don't get inspired like football players. They don't get inspired by big speeches and sort of the way you see it on TV. I think they get inspired by showing them that the results of their thinking and of their work have real impact on people around the world.
And positive impact. I'm not going to name the tech titan that lost so many employees when they realized that some of the outcomes of the work the engineers were doing was basically putting some lives at risk.
I think we've done a bad job of thinking about, as in Silicon Valley, and thinking about some of the damage that the inventions have done to young men and to young women. I think it's okay if...
It's with a history of technology there. I mean people who build things like you, like us, we have the best intentions in the world but if there is something that is common to every invention throughout history is that it's been hijacked by evil people from their initial purpose and that you cannot prevent. However, the fact what we can prevent is the fact that when we develop, we think about safeguards. We think about how we can improve those positive outcomes. Right.
I think that's right. I think powerful technology comes both with responsibility and with an understanding that technology is neither good nor bad. It's technology and good people are going to use it to do amazing things and bad people are going to try and do it to do really bad things.
There is an example with the European regulation of AI at the moment that is very much centered on the process not the outcome. And if you think about it, a few years back in France, my home country, someone was stabbed with a scalpel, right? If you have the thinking of the EU regulation on AI, then you would ban the scalpel, right? Preventing the scalpel from being used to save lives in surgery rooms. And this is how AI is regulated in Europe at the moment instead of looking at how it's being used and what the outcome is. They're looking at the process. At the end of the day, we need outcome-driven regulation. That's my point.
But remember early in the rise of the internet, I mean, we were getting emails from Nigerian princes, right? There were scams there. There were all sorts...
Because you're famous. You never got those. But...
Come on. There was somebody who owed me money in Nigeria. They were scams and after a while people learned that they were scams and set them aside and that was a bad use of this cool technology. And they'll be the same. We'll have a problem with deep fakes. We'll have a problem with...
We got Nobel prizes because there was a hijacking of what Alfred Nobel invented.
That's right. That's right. And we as a community have to come together and say we have this powerful technology. It's our obligation to do extraordinary work with it. We're using a lot of resources. We have to do extraordinary work and at the same time we have to think about how to mitigate the damage it could do and those are the safeguards you're talking about and I think responsible companies are thinking those thoughts, are doing those things. A good example is recently Microsoft's five-point plan on data centers, right? They said that if we put a data center in your community, your taxes shouldn't go up.
That seems reasonable. Your power rates shouldn't go up. That seems reasonable too. These are, this is how you engage with the community, right? So that every stakeholder in the ecosystem wins.
That should be by design.
I mentioned we would go back to academia because there is actually a connection here. Not so long ago, I joined the Mohamed bin Zayed University of Artificial Intelligence in Abu Dhabi. And when I was told the various floors, the various buildings, I was told, 'Oh, here are employees from IBM and here are employees from Cerebras Systems.' So, I think I'd love for our audience to hear what you're doing with this extraordinary university in Abu Dhabi.
