Panel
AI and Open Source: Shaping Each Other's Future | RAISE Summit 2026
RAISE SummitRaj Dutt, Mark Surman, Oskari Saarenmaa, Gideon Mendels, Charles Zedlewski, Olivier Huez
- The open-weight AI approach is expected to expand significantly, with increased transparency in training recipes and checkpoints resembling NVIDIA's Nemotron projects, driven by sovereign tech entities and the demand for inspectable training data and modifiable code.
- A distinction is anticipated between the software-defined open source stack using traditional licensing and "open washing" tactics where proprietary models are loosely labeled as open source.
- While open-weight models offer economic value through derivations, their production is currently concentrated in China due to subsidies, creating high resource barriers for individual developers to rebuild versions without equivalent backing.
- Usability and modification thresholds for most open-weight models are expected to be met even with incomplete upstream pipelines, generating substantial value through millions of derivatives on platforms like Hugging Face.
- Grassroots open source is predicted to merge with industrial resources from governments and companies to produce highly reliable digital industrial goods, while the definition of "open" may tighten to rally major industry players like NVIDIA around truly open models.
- Cloud hyperscalers have evolved to cooperate better with the community, though historical behaviors involved using permissive licenses to outpace contributors without reciprocation.
- Projects such as OpenSearch and Redis are cited as evidence that shifting to fully proprietary models is unsustainable, with these entities returning to open source licensing a couple of years after initial changes.
- Dedicated AI platforms are expected to offer higher quality and performance at lower costs than hyperscalers for recent models, as research insights decay within six to nine months, necessitating a continuous pipeline of new innovations to avoid commoditization.
- Successful open source companies must continuously balance community value and commercial capture on a monthly or quarterly basis, a challenge investors view as critical for sustaining systems over years and decades.
- AI agents will initially create manageable noise and automated PRs before becoming a net positive by enabling maintainers to filter output and scale development through testing environments, though individual developers face an identity crisis shifting from coding to reviewing agent contributions.
- A growing sense of entitlement from corporate code contributors is expected to cause burnout for non-commercial maintainers obligated to respond to issues, while the viability of creating custom forks is projected to increase, revitalizing the ecosystem.
- The "CQ" project is anticipated to serve as a cultural experiment in agentic coding, potentially fostering collaboration over isolation, while open source is predicted to become the primary discovery channel for developer infrastructure due to model training recommendations.
- Licensing strategies may shift away from Apache MI2 licenses as AI agents facilitate closing gaps, yet the ecosystem is expected to remain heterogeneous with a hybrid model where open source serves as cheap, modifiable infrastructure alongside closed solutions.
- Open source is predicted to become the worst enemy to large AI players regarding valuation economics despite being their best friend technologically, while frontier models will continue recommending open source solutions due to their reliance on open source training data, creating a self-reinforcing cycle.