The Great Data Escape: AI, Local-First, and the Cloud Exodus (solutionsreview.com)

Brian Pontarelli (FusionAuth) warns that three converging trends—AI agents, local-first computing, and repatriation—are eroding the cloud’s decade-long “data moat.” As organizations realize that their most valuable asset (and compliance burden) is often captive inside SaaS and PaaS silos, they’re pushing to regain control. Pontarelli contrasts SaaS/PaaS lock-in (hard-to-export logs, metrics, or email archives) with IaaS (where data is more retrievable), and cites industry signals—Thales: ~60% of corporate data in the cloud (doubling since 2015); Barclays: repatriation intent rising from 43% in 2020 to 83% in 2024. For AI/ML, this matters technically and strategically. Agentic models that learn and act locally (examples: tuning turbines, robotics, private personal assistants) require direct data access and low-latency inference; they’re fundamentally different from reactive cloud chatbots like ChatGPT. Recent local model launches (e.g., Meta’s Llama, China’s DeepSeek) and the rise of local-first apps—enabled by sync primitives such as Conflict-free Replicated Data Types (CRDTs)—make on-device learning and private inference feasible. Real-world moves include Linear’s sync engine, Git workflows, Ink&Switch prototypes, and enterprise repatriations (e.g., GEICO) driven by cost, speed, and compliance. The implication for ML teams: design pipelines and models for data locality, privacy-preserving fine-tuning, and intermittent connectivity—or risk losing control, agility, and margin to either hyperscalers or homegrown stacks.