WDCWestern Digital Corporation

“One leading hyperscalers LLM processes over 16 billion tokens per minute via direct API used by their customers, while another AI company processes over 2.5 billion prompts every single day from 900 million active users.”

“the demand for storage is increasing because storage is persistent, right? If you look at it, if you talk about inferencing, the resources that are used in inferencing, whether it's compute or whether it's memory, they can get recycled. But the data that's getting generated for inferencing is not being recycled. All that data that is getting generated is getting stored and that storage data that's being stored is persistent. And that is consistent with what we see with all our top 4 customers, whether their business model is in search or whether their business models in advertising or in the enterprise software space, it's pretty consistent. It's really this ongoing data storage requirements to support training, improvements in training to support the demands of inference and to support synthetic data being driven by physical AI.”

“The third driver that we see for data storage for HDDs is obviously physical AI. As we've highlighted before, physical AI with the limited data sets that it has, whether it's autonomous vehicles or robotics is using AI to generate a lot of synthetic data to further train and enable physical AI as well. And obviously, that needs to -- any data that gets generated out of that gets stored and feeds that whole training and synthetic data development loop as well. So those are the 3 big drivers of growth that we see going forward, Erik. That's why we have the confidence to see exabyte growth growing beyond 25% CAGR going forward.”

“These forces are not additive. They are a compounding loop. Inference creates data, agents consume and generate more data, physical AI creates data and train synthetic models that create more data. And ultimately, the loop accelerates. We are truly seeing that the AI-driven data economy is creating an unprecedented demand for high-capacity, reliable and high-performance storage on HDDs. This reinforces our conviction that long-term data storage growth will be greater than 25% CAGR.”

“As AI workloads extend from training to large-scale inferencing, data generation is at an inflection point. This year, inference is expected to account for roughly 2/3 of all AI compute. This larger focus on inference increases the amount of data generated, which in turn increases the need for data storage.”

“we see the rise of agentic AI, the next wave and arguably the biggest yet. What we are seeing with agentic AI frameworks represents a structural shift from AI that answers questions to AI that continuously executes workflows. That transition materially increases data generation and extends data retention cycles. Every hour of autonomous agent work and every action an agent takes creates data that must be stored. As a result, we expect agentic AI to drive a step function increase in capacity-oriented storage demand, particularly in cloud and enterprise environments.”

“We really see 3 core drivers of HDD growth going forward. One that we've seen for quite a while right now, which is the ongoing storage requirements that's associated to training. So that's not going to end. Training will continue. relearning, reinforcement, retraining is going to happen. And what we are seeing from our customers as they retrain and reinforce learning with these models, the quality of the model results that they get are improving. So they continue to store all the data they're generating to enable improved quality of the model.”

“Synthetic data, the primary fuel for physical AI is by design orders of magnitude larger than real-world inputs that seed it. Across industries, physical AI data factory frameworks are being designed to transform limited training data into larger synthetic data sets at scale for robotics, autonomous vehicles and vision AI. And physical AI itself, robots, industrial systems, autonomous fleets generates continuous streams of video, sensor and motion data that must be stored, versioned and fed back into training loops.”

“The second driver that we're seeing is obviously the rise of agentic AI and inferencing. With every inference that happens, new data is getting generated. And what's happening is that all that new data that's getting generated is getting stored as well to both feed back into training models and be stored to support future inference references as well.”

“even in inferencing, right, we saw a symbiotic relationship. The new data that's created from inferencing typically will get stored on HDDs. The vectoring data that's required for inferencing, that's actually stored on flash. So it's a very symbiotic relationship.”

“Our dual pivot technology is being built specifically for new AI workloads with an open API approach aimed at simplifying deployment at scale.”

“it is an exciting time to be part of WD, a focused HDD company and a strategic partner to hyperscalers and cloud service providers in this AI-driven data economy.”