Conference Presentation, Keynote
DDN President, Paul Bloch: AI is the Pivot -- Data is the Power
- AI is expected to redefine the global economy as the fourth or fifth industrial revolution, transitioning from a first phase of training to a second phase focused on inference and digital transformation.
- While Nvidia remains the core provider for training, the speaker forecasts increased adoption of chip makers like Samba Nova and Cerebras as the market for inference expands.
- Financial services have shifted from initial reluctance to becoming major GPU implementers in high-frequency trading, with operations now commonly utilizing between 10,000 and 100,000 GPUs.
- Healthcare applications are projected to leverage AI for accelerating drug discovery, genomics modeling, and large-scale simulations.
- Over the next three to five years, the CAC 40 is expected to move beyond pilot programs to treat GPU infrastructure as mission critical.
- Europe is projected to host approximately 13 AI factories and 4 giga-factories, requiring investment rates of 100 to 300 million and 4 to 5 billion respectively.
- Capital requirements for building proper AI factories are predicted to reach 40 to 50 billion by the end of the current phase and 50 to 100 billion by next year.
- A race for AI capability is anticipated involving half a trillion in US announcements and the development of superintelligence.
- The industry is moving toward physical AI requiring embedded intelligence and core-to-edge communication.
- System farms of 50,000 GPUs are expected to face hardware failures where one unit fails every five minutes, necessitating frequent checkpointing.
- Data scaling requirements are projected to grow from 100 to 500 terabytes and petabytes up to multi-exabytes.
- Model sizes are predicted to advance to 500 billion parameters, followed by trillion-parameter models in a short timeframe.
- Infrastructure capacity expectations have doubled from 50,000 GPUs six months ago to 500,000 GPUs one year later, with plans for XAI and hyperscalers to deploy one million GPUs by early next year.
- Approximately ten companies worldwide are expected to possess one million GPUs next year.
- DDN's Infinia product is projected to deliver 22x faster time to results for RAG compared to standard AWS setups and save approximately 40% in GPU time and costs.
- XAI became DDN's largest customer in 2024, replacing NVIDIA, which held that position in 2021, 2022, and 2023.
- Google is expected to begin production use of DDN for AI workloads this month, with plans to offer a seamless 1P integration requiring only a few clicks.
- Increasing data volume, including real-time video and feeds, will be continuously fed into foundation models.