Thesis. The 2026 AI cycle shifts from conversational models to “digital workers” (agents) and physical AI (robots, driving, inspection); value migrates from the model layer to operational integration and automated real assets, converging directly with smart cities and tokenizable infrastructure.
Key Data. ~71% of companies are expected to integrate AI agents into key functions (finance, risk) by the end of 2026; humanoid robots in production testing (Boston Dynamics’ Atlas at Hyundai, January 2026); industrialized “physical AI” stack (10-billion-parameter vision-language-action models, simulators reducing validation variance by ~83%); digital twins of factories (NVIDIA × Siemens); paradigm shift from “bigger is better” to specialized models with near-zero marginal cost (e.g., Falcon-H1R, elite performance with 7× fewer parameters). On the biotech side: AI-driven protein design moves from prediction to creation; brain-machine interfaces in clinical trials (speech, robotic movement).
Analysis. Three margin shifts: (1) from model to orchestration, where agents at $167/day operating cost for 4,000 executions (real measured case) destroy the cost of middle-office processes: value is captured by whoever owns the workflow and proprietary data, not the LLM; (2) from digital to physical, the shortage of skilled labor makes physical AI monetizable (manufacturing, logistics, construction sites): robots become flow assets, leasable, insurable, financeable, and therefore tokenizable; (3) from growth to rent, smart cities: sensor networks, distributed energy, autonomous mobility = infrastructure with recurring revenue indexed to usage, exactly the “cash-flow RWA” profile of the new paradigm. Biotech (MANBRIC) remains the most convex class: long cycles, but massive optionality on longevity and computational health.
Investment Implications. Barbell: Agent orchestration infrastructure (short-term, SaaS revenue) + fundable automated physical assets (robot fleets, distributed energy, edge data centers) structured as tokenized yield vehicles (medium-term); convex computational biotech pocket (long-term). Avoid “wrapper-model” without proprietary data.
Risks. Intense AI capex cycle (risk of digestion if agent revenues disappoint); regulation of autonomous machines (liability); biotech valuations sensitive to long-term rates in a “higher for longer” regime.
Skip to content
Skip to sidebar
Skip to footer
