Welcome to the Convergence Thesis Weekly. This week, we are looking at the critical bottleneck shifting from digital intelligence to physical infrastructure.
Thesis of the Week: The Physical Bottleneck
For the past three years, the primary constraint on technological progress was algorithmic—building better models. As of May 2026, the paradigm has definitively shifted. We have abundant, agentic intelligence. The new constraints are entirely physical: power generation, datacenter cooling, and physical embodiment (robotics). The capital flows are reflecting this reality, pivoting heavily from pure software to “hard tech” infrastructure. If you don’t have the watts or the actuators, your intelligence is stranded.
1. AI: The Rise of “Agentic Swarms”
We are moving beyond conversational LLMs into the era of Agentic Swarms. The latest models from leading labs are no longer just answering prompts; they are orchestrating complex, multi-step workflows across disparate APIs.
- Signal: The focus is no longer on context window size, but on reliability in execution over long time horizons.
- Portfolio Move: Look toward companies building the monitoring, orchestration, and security layers for autonomous AI agents.
2. Energy: Tech Giants Become Utilities
The energy demands of next-generation AI data centers have forced tech giants into the energy sector. We’re seeing unprecedented capital injected into nuclear energy, specifically Small Modular Reactors (SMRs).
- Signal: Hyperscalers aren’t just buying green credits anymore; they are directly financing the construction of nuclear facilities to guarantee gigawatt-level power for training clusters.
- Portfolio Move: SMR manufacturers and uranium suppliers are becoming proxy plays on AI compute growth.
3. Robotics: VLMs Hit the Factory Floor
Humanoid robotics are making the leap from highly curated demo videos to actual, messy factory floors. Vision-Language-Action Models (VLMs) are allowing robots from companies like Figure and Tesla to generalize tasks on the fly, rather than relying on brittle, hard-coded routines.
- Signal: The shift from “programmed automation” to “learned autonomy.” Robots are learning by watching humans, drastically reducing setup time.
- Portfolio Move: The value capture is shifting toward the VLM software stack and the proprietary datasets of real-world physical interactions.
4. Crypto: DePIN as the Compute Relief Valve
With centralized GPU access heavily constrained and expensive, Decentralized Physical Infrastructure Networks (DePIN) are experiencing a renaissance. Networks aggregating idle GPU cycles globally are finally finding product-market fit among mid-tier AI developers.
- Signal: DePIN is moving from a speculative crypto narrative to a pragmatic infrastructure solution for AI startups priced out of tier-1 cloud providers.
- Portfolio Move: Monitor the utilization rates of leading decentralized compute networks; tokens with actual buy-and-burn mechanics tied to compute usage.
5. BCIs & Longevity: Moving Beyond the Lab
Endovascular Brain-Computer Interfaces (BCIs) are expanding clinical trials, offering minimally invasive read/write access to the brain. In longevity, CRISPR-based therapies targeting specific hallmarks of aging (like cellular senescence) are showing robust long-term data in phase 2 trials.
- Signal: The regulatory pathways for “enhancement” technologies are slowly being paved by initial therapeutic use-cases (e.g., curing paralysis or blindness).
- Portfolio Move: Focus on the “picks and shovels” of biotech—the specialized delivery mechanisms (like advanced LNPs) required to get CRISPR therapies into specific organ tissues.
Until next week, Stay ahead of the curve.