Rene Haas

But the level of compute that will be required for inference is actually going to be much larger. I think it'll be Better than half, maybe 80% over time would be inference. But the amount of inference cases that we'll need to run are far larger than what we have today.

So I think that's why you're seeing companies like CoreWeave and Oracle and people who are not traditionally in this space who are now running AI cloud. Well, why is that? Because... capacity. The traditional large hyperscalers, the Amazons, the Metas, the Googles, the Microsofts, there's just not enough capacity.

So I think that's why you're seeing companies like CoreWeave and Oracle and people who are not traditionally in this space who are now running AI cloud. Well, why is that? Because... capacity. The traditional large hyperscalers, the Amazons, the Metas, the Googles, the Microsofts, there's just not enough capacity.

So I think that's why you're seeing companies like CoreWeave and Oracle and people who are not traditionally in this space who are now running AI cloud. Well, why is that? Because... capacity. The traditional large hyperscalers, the Amazons, the Metas, the Googles, the Microsofts, there's just not enough capacity.

So what I think we'll continue to see is a changing of the landscape, maybe not a changing so much, but certainly opportunities for other players in terms of enabling and accessing this growth. And for ARM, it's very, very good because we've seen a very, very large increase in growth in market share for us in the data center, AWS at reInvent,

So what I think we'll continue to see is a changing of the landscape, maybe not a changing so much, but certainly opportunities for other players in terms of enabling and accessing this growth. And for ARM, it's very, very good because we've seen a very, very large increase in growth in market share for us in the data center, AWS at reInvent,

So what I think we'll continue to see is a changing of the landscape, maybe not a changing so much, but certainly opportunities for other players in terms of enabling and accessing this growth. And for ARM, it's very, very good because we've seen a very, very large increase in growth in market share for us in the data center, AWS at reInvent,

this week who build their general purpose devices, Graviton, based on Arm. They say that 50% of all new deployments are Graviton. So 50% of anything new at AWS is Arm. And that's not going to decrease. That number is just going to go up. And I think one of the things we're seeing, whether it's devices like Grace Blackwell from NVIDIA.

this week who build their general purpose devices, Graviton, based on Arm. They say that 50% of all new deployments are Graviton. So 50% of anything new at AWS is Arm. And that's not going to decrease. That number is just going to go up. And I think one of the things we're seeing, whether it's devices like Grace Blackwell from NVIDIA.

this week who build their general purpose devices, Graviton, based on Arm. They say that 50% of all new deployments are Graviton. So 50% of anything new at AWS is Arm. And that's not going to decrease. That number is just going to go up. And I think one of the things we're seeing, whether it's devices like Grace Blackwell from NVIDIA.

Grace, which is the CPU, and that's ARM, using an NVIDIA GPU, that's a big benefit for us because what happens is the AI cloud is now running a host node based on ARM. And if the data center now has an AI cluster where the general purpose compute is ARM, They naturally want to have as much of the general purpose compute that's not AI running on ARM.

Grace, which is the CPU, and that's ARM, using an NVIDIA GPU, that's a big benefit for us because what happens is the AI cloud is now running a host node based on ARM. And if the data center now has an AI cluster where the general purpose compute is ARM, They naturally want to have as much of the general purpose compute that's not AI running on ARM.

Grace, which is the CPU, and that's ARM, using an NVIDIA GPU, that's a big benefit for us because what happens is the AI cloud is now running a host node based on ARM. And if the data center now has an AI cluster where the general purpose compute is ARM, They naturally want to have as much of the general purpose compute that's not AI running on ARM.

So what we're seeing is just an acceleration for us in the data center, whether it's AI or inference or general purpose compute.

So what we're seeing is just an acceleration for us in the data center, whether it's AI or inference or general purpose compute.

So what we're seeing is just an acceleration for us in the data center, whether it's AI or inference or general purpose compute.

On one hand, it would be crazy to say that growth continues unabated, right? We've seen, obviously, that that is never really the case. I think what will get very interesting in this particular growth phase is to what level does real benefit come from AI that can augment and or replace certain levels of jobs? You know, some of the AI models and chatbots today are decent, but not great.

On one hand, it would be crazy to say that growth continues unabated, right? We've seen, obviously, that that is never really the case. I think what will get very interesting in this particular growth phase is to what level does real benefit come from AI that can augment and or replace certain levels of jobs? You know, some of the AI models and chatbots today are decent, but not great.

On one hand, it would be crazy to say that growth continues unabated, right? We've seen, obviously, that that is never really the case. I think what will get very interesting in this particular growth phase is to what level does real benefit come from AI that can augment and or replace certain levels of jobs? You know, some of the AI models and chatbots today are decent, but not great.

They supplement work, but they don't necessarily replace work. But if you start to get into agents that can do real level of work that can replace what people might need to do in terms of thinking and reasoning, then that gets fairly interesting. And then you say, well, how's that going to happen? Well, We're not there yet, so we need to train more models.