Fireside Chat, Interview
Fireside Chat with William Falcon & Ozan Kaya of Lightning AI | RAISE Summit 2026
- Anticipates a market shift by late 2025 from initial Gen AI companies and resellers to direct enterprise clients seeking open source models and full-stack ownership to avoid censorship and dependency on global providers.
- Plans to expand physical power capacity from an initial 25 megawatts to 70 megawatts by leveraging six existing data center relationships and a recently signed lease in Canada.
- Targets the 5 to 20 megawatt power requirements of Gen AI and enterprise customers, explicitly excluding "Mag7" hyperscalers who require gigawatt-scale infrastructure.
- Intends to evolve from a compute provider to a full-stack solution offering GPUs, software tools, and fine-tuning capabilities to enable 95% of use cases while maintaining customer flexibility and preventing vendor lock-in.
- Expects hardware costs to commoditize and models to become smaller and more efficient, eventually allowing inference on CPUs and personal devices, while older model workloads may shift to alternative chips like IPUs and GraphCore.
- Focuses on a "Neocloud" strategy that democratizes access to compute, providing "four-star" or "white glove" service for non-PhD engineers to build, train, and deploy proprietary models without external dependencies.
- Predicts that owning intelligence will become the primary economic value as customers seek to own their model weights and data, with strategic mergers (e.g., with Lightning) facilitating deeper enterprise integration and increased compute cluster sales.
- Maintains a hardware-agnostic approach, leveraging partnerships with NVIDIA and Dell while diversifying infrastructure to ensure customers can choose the best technology for their specific use cases.
- Foresees a historical trajectory where today's expensive technology becomes accessible to all, mirroring the transition from mainframes to consumer devices, driven by the trend of open source prevalence and the ability to run models locally.