Name: Brainstorm AI London 2024: The Importance Of AI Chip Development
Uploaded: 2024-04-15
Duration: 5 min 37 s
Description: Presenter: Ami Badani, Chief Marketing Officer, Arm

Presenter: Ami Badani, Chief Marketing Officer, Arm

Brainstorm AI London 2024: The Importance Of AI Chip Development

Photography by Joe Maher / Fortune Brainstorm AI

Conferences

Brainstorm AI London 2024: The Importance Of Ai Chip Development

Hi there. I think we're all here because we believe that A I will have a profound impact, probably the most profound impact that will be in our lifetime. It'll be bigger than the internet, bigger than the smartphone revolution, bigger than decades of innovation prior like the invention of the light bulb, but it's only A I is only on its tipping point. In fact, you've seen use cases today like sports, financial services, entertainment and we're only getting started. If you think about how A I will become mainstream over the next decade, you think about chat G BT, it's at 200 million users in just over a year. That's the fastest growing application that we have ever seen. Secondly, you think about self driving cars, one in 10 cars by 2030 will be self driving cars and that will all be because of artificial intelligence. 3rd, 75% of office applications, office productivity applications will incorporate some form of generative A I in some way, shape or form. So as you can see A I will become mainstream, but we're on a collision course with reality and it's not sustainable. There are two things that we need to really advance artificial intelligence and that's compute and power. Let's talk about compute A I models are more hungry, more data hungry than they than they have ever been before. They're increasing in complexity. They're increasing in size and they're increasing in scope. And what that drives is the need for more compute chips. In fact, if you recall open A I had their latest, so a model that's their latest text to video model. It takes 100,000 A I chips working at full compute capacity and full power consumption in order to train. So a that's a huge amount of compute chips and that's just one large language model of any of many. Then you think about smaller language models that are going to grow in scale and be quantized and apply to things like smartphones, your thermostat in your home and other home appliances, your PC laptops, your inference in the cloud. And then you think about large language models, small language models and that's really driving the demand for compute chips, but it's not about driving the demand for compute chips. It's really about power efficient compute chips. And that's really gonna help us carry the advancements of artificial intelligence. Power efficiency is in arms DNA. In fact, 99% of smartphones use an arm power efficient chip. And it's the same design philosophy that we're applying to the advancements of artificial intelligence. In fact, if you think about a I it comes with a cost and that cost is unfortunately power. In fact, large language models have an insatiable demand for power. So without the need to address power, we won't have advancements in artificial intelligence. And if you think about the future of artificial intelligence, it's all going to be about power efficient chip designs. Now, if you look at this graph, what it really shows is data center, power consumption, most A I training and large language models are all done in the data center today. And that's 2% of the global electricity consumption, which is an enormous amount. But if I talk, you know, I spoke about earlier that A I will become mainstream. And when A I becomes mainstream, the biggest tax that we'll see is our power grid and we won't be able to continue the advancement of A I without addressing power. In fact, by 2030 artificial intelligence will have triple the load than what it looks like today. To put this into perspective, 25% of the US power grid will be because of artificial intelligence. So we really need to address this power problem head on. And if you look at chat G BT, it requires 15 times more energy than a traditional web search. So that's a huge amount of energy. And that's what arm is really focused on is driving power efficient compute chips. In fact, if you look at customers like Google, Microsoft, Amazon Oracle they are all designing power efficient chips because they know they have to solve the power problem first in order to solve and drive the advancements of artificial intelligence. So in order for companies to have a competitive advantage, they really need to think about power efficiency in solving the power problem. In order to advance artificial intelligence.