(Benefitting from a few hindsights)

1. Capital markets reopen a critical IPO window for AI-native companies. Anthropic and SpaceX appear the most IPO-ready. Private market valuations for AI-native companies will continue rising and peak around the time of Anthropic’s IPO despite the recent SaaS selloffs. However, this will not last forever as corrections (2022 moment) will come eventually.

2. AI agents break traditional database architectures. This shift will accelerate adoption of low-latency, highly parallel platforms like Databricks (Lakebase) and MotherDuck (serverless DuckDB).

3. As agent context grows, DRAM capacity and memory bandwidth will become binding constraints. Infrastructure that predicts data access, prefetches context, offloads cold state to cheaper storage, and deduplicates embeddings will define the next wave of performance and cost optimization.

4. Agent-driven workloads have disrupted traditional SaaS models as we have observed. As agents become the primary users of software, seat-based SaaS and UI-heavy workflows will continue to give way to API-first, proactive execution models that automate workflows rather than relying on reactive chat-based human interactions.

5. The AI inference market is likely to expand by >10× over the next two years. Multiple winners will coexist across hyperscalers, neoclouds, and a small number of specialized startups, each differentiated by workload profile, cost structure, or performance niche.

6. Multi-modality will become the default interaction model for AI applications. As agents are embedded into everyday workflows, applications will be built to natively handle text, voice, images, and structured data in a single experience rather than treating modalities as add-ons.

7. AI security will expand beyond data protection to agent identity, permissions, behavior validation, and AI-generated code review to address verification and control at production scale.