Conference Presentation, Keynote
DDN President, Paul Bloch: AI is the Pivot -- Data is the Power
- Paul Bloch, President and Co-founder of DDN (founded 25 years ago), presented at the Race Summit in Paris, highlighting the company's scale of 1,200 employees, 800 engineers, and role in powering large-scale AI systems.
- The presentation characterized the current moment as the start of a fifth industrial revolution, noting that while training is currently dominant, inference is in its "infant" stage and will rapidly expand.
- Nvidia remains the core hardware provider for training, though competitors like SambaNova and Cerebras are emerging for inference workloads.
- France is positioning itself as a hub for sovereign AI, leveraging foundation models like Mistral (which recently launched a sovereign French cloud) and infrastructure providers such as ScaleWay, OVH, and Fluidstack.
- AI deployment is becoming mission-critical across verticals:
- Financial Services: High-frequency trading now utilizes 10,000 to 100,000 GPUs, reversing previous resistance to GPU adoption.
- Healthcare: AI is accelerating drug discovery, genomics modeling, and large-scale simulations.
- Automotive: Expansion includes autonomous driving (Tesla, BYD) and digital twin factory automation.
- Energy & Public Sector: Sovereign control over data is driving adoption in defense and energy sectors, with major DDN clients including CEA, EDF, Idris, CNRS, and Total Energy.
- European investment in "AI factories" is scaling rapidly, with plans for approximately 13 AI factories (valued at €100M–€300M each) and 4 "giga-factories" (valued at €4–€5 billion each).
- Building a proper AI factory requires capital of €40–€50 billion today, potentially rising to €50–€100 billion next year, while US announcements indicate potential investments up to $500 billion.
- GPU deployment targets are accelerating exponentially; XAI plans to deploy 1 million GPUs by early next year, a target the speaker predicts only about ten companies globally will meet.
- DDN's core value proposition focuses on the data layer required for massive scale, specifically managing data ingestion, preparation, indexing, and checkpointing for farms of 50,000 to 200,000+ GPUs.
- Checkpointing is identified as critical for stability in large GPU farms where individual unit failures occur frequently (e.g., one GPU failure every five minutes in a 50,000-GPU farm).
- Future infrastructure must support real-time, multi-modal data feeds (video, live feeds) to feed foundation models as parameter counts approach 500 billion and potentially trillions.
- DDN launched a new product, Infinia, designed to manage data and accelerate workloads on-prem and in the cloud, addressing low-latency inference and multi-tenancy requirements.
- A benchmark test of Infinia on AWS against a standard S3 setup demonstrated a 22x reduction in time-to-result and a 40% savings in GPU compute costs.
- Customer milestones include:
- DDN was Nvidia's largest customer in 2021–2023.
- XAI became DDN's largest customer in 2024, using DDN products (Exascaler and Infinia) for Grok training and inference.
- Google announced in 2024 that DDN will manage AI workloads in the cloud; DDN is now in production with Google Cloud Platform (GCP) as a 1P offering with full security and encryption.
- Sovereign AI and data locality are becoming paramount for regulatory compliance, ethical AI, and control over data value in a geopolitical landscape characterized by increasing instability.
- DDN is internally implementing Generative and Agentic AI for its own engineering and customer support functions, reporting initial success despite implementation challenges.
- The speaker emphasized that while AI technology evolves rapidly, human expertise remains essential for managing the idiosyncrasies of massive HPC and AI infrastructure.