Conference Presentation, Panel
From Research to Reality: Why Open Source Is the Engine of AI | RAISE Summit 2026
RAISE SummitIon Stoica, Thomas Wolf, Dillon Rolnick, Eiso Kant, Robin Rombach, Mark Porter, Dylan Rolnick
- Iso Kent anticipates a dystopian scenario by 2030 where economic and scientific value is centralized in only two to four companies if intelligence remains proprietary, though he predicts a future where intelligence becomes a commodity available to many rather than few.
- Scaling laws are considered real with no current end to the utility of increased compute, leading to infrastructure companies aligning incentives to commoditize models and reduce value to the cost of land, energy, and server silicon.
- Thomas Wolf asserts that all new AI techniques have historically been open source, yet expresses concern that innovations in attention mechanisms from Chinese labs and the lack of open infrastructure could restrict global scientific research in fields like bio, materials, and fusion.
- Robin notes the world model field is early and fluid with no closed-source leader, predicting that agents will soon become the primary interface for AI while warning that overfitting to benchmarks harms the open source field's progress.
- Thomas Wolf forecasts impressive non-LLM developments next year in robotics, physics, material research, and fusion, describing the next few years as "insanely stressful" but filled with potential for physical AI.
- Economic viability requires a monetization strategy, as the immense capital needed for scaling (billions to tens of billions) makes open sourcing models without a business model unreasonable, whereas software layers have near-zero reproduction costs and high incentives for open sourcing.
- Open source distribution is predicted to provide ecosystem extensibility, tools, and trust, enabling small companies to potentially beat intelligence frontiers and avoid a future where only four or five entities control all intelligence.
- Robin warns that staying within a single ecosystem creates dependency on that ecosystem's decisions, and suggests that a lack of open models could cause massive disruptions if specific classes of models, such as those from China, become unavailable due to export controls.
- Thomas Wolf highlights a lack of major security events in the past four years aside from deepfakes and urges Europe to move beyond fear-based regulation to have an "adult and emotional discussion" on security.
- The open source track for the event was funded by Commit, Hugging Face, and Mozilla, while Thomas Wolf expresses hope that Europe's manufacturing base will help retain resources for physical AI and world models.