Private company notes · August 2026
While AI continues to advance at a rapid pace, the real-world infrastructure supporting it is hitting practical limits.
Building and running advanced AI now faces major real-world hurdles: local power grids cannot supply enough electricity, microchips are generating too much heat, general-purpose chips are becoming too expensive for routine tasks, training systems are running out of high-quality human data, and software still lacks physical hands to do manual work.
Here are five companies we cover that are developing solutions to these operational bottlenecks.
1. Panthalassa
- The Chokepoint — Power Grid & Water Cooling Limits. Modern data centers require massive amounts of continuous power and water for cooling, leading to multi-year utility delays and local grid congestion.
- The Solution. Panthalassa builds modular, ocean-based power platforms that use wave energy to power off-grid computing pods, utilizing deep ocean water for natural heat dissipation.
- Company Stage: Growth Stage / Series B ($140M Series B closed)
- Website: panthalassa.com
2. Lightmatter
- The Chokepoint — Chip-to-Chip Wiring and Heat. In large AI server clusters, transmitting data between thousands of microchips over traditional copper wires generates excessive heat and slows down processing speeds.
- The Solution. Lightmatter develops silicon photonics systems that route data between processors using pulses of light instead of electricity, significantly reducing heat and latency.
- Company Stage: Late Stage ($5.37B valuation benchmark)
- Website: lightmatter.co
3. Positron AI
- The Chokepoint — GPU Cost and Memory Limits. AI models running on standard graphics chips (GPUs) consume high levels of electricity and face memory bottlenecks.
- The Solution. Positron designs specialized hardware built specifically for running transformer-based AI models, lowering the ongoing cost and energy required per user interaction.
- Company Stage: Commercial Stage ($4.5B valuation benchmark)
- Website: positron.ai
4. Mercor
- The Chokepoint — High-Level Human Expertise. AI models have largely exhausted general web text and now require specialized, professional human knowledge to improve reasoning in complex fields.
- The Solution. Mercor operates an AI-assisted hiring and evaluation network that matches domain specialists (such as doctors, lawyers, and engineers) with AI labs to generate advanced training data.
- Company Stage: Growth Stage ($10B valuation benchmark)
- Website: mercor.com
5. 1X Technologies
- The Chokepoint — Physical Labor. Digital AI models operate solely on screens and cannot physically move inventory, package goods, or handle dynamic industrial labor.
- The Solution. 1X builds humanoid and wheeled androids designed with safe, human-like actuators to work alongside human staff in warehouses and commercial settings.
- Company Stage: Commercial Deployment (Series B)
- Website: 1x.tech