# Ideas - atoms and substrate

← [Ideas](ideas.md) · [Startups](README.md) · [Index](../../../README.md)

---

Software founding got cheap. **Watts, rights-of-way, manipulation data, and last-metre handling did not.** These ideas sit where the corpus says the 2030s growth rate is decided ([robotics](../../physical/robotics/), [energy](../../../01-substrate/energy.md), [2032–2040](../../../04-timelines/2032-2040.md)).

## 1. Powered land and interconnection as a product

**Why now.** Energy and permits bind before capital does ([energy](../../../01-substrate/energy.md), [U2](../../../06-uncertainties/power-permitting.md)). Hyperscalers and neoclouds bid for megawatts and queue position; most founders still pitch "AI software."

**Build this.** Develop, entitle, and lease AI-ready sites: land + interconnection rights + shell + cooling path, with transparent queue and curtailment terms. Software for queue navigation is a wedge; the asset is the site.

**Own this.** Land, interconnection agreements, and long-term offtake contracts.

**Not this.** A SaaS dashboard of public queue PDFs.

**Dead if.** Permitting reform and surplus transmission make powered sites abundant in your geography - the rent moves.

## 2. Behind-the-meter and interruptible load services

**Why now.** Grid connection lags; on-site generation, storage, and flexible load are how campuses actually turn on ([energy](../../../01-substrate/energy.md)).

**Build this.** Integrated behind-the-meter systems + offtake structuring for AI and industrial loads: gas, storage, renewables, controls, and utility negotiation as one product.

**Own this.** Project rights, long-term service contracts, and operational data on real load shapes.

**Not this.** A chatbot for "energy strategy decks."

**Dead if.** Utilities productize the same bundle faster with balance-sheet advantage - partner or niche by geography.

## 3. Manipulation data factories

**Why now.** There is no internet of manipulation ([robotics data](../../physical/robotics/data-problem.md)). Whoever produces dense, rights-cleared physical interaction data trains the robots that matter.

**Build this.** Facilities and fleets that generate teleop and autonomous trajectories in target domains (warehouse, apparel, food, construction), with clean labels and licensing to model and robot firms. Monetize data + benchmark suites, not only hardware.

**Own this.** The dataset, the collection ops, and preferred deployment partners.

**Not this.** A foundation model trained on public videos of people cooking.

**Dead if.** Cross-embodiment transfer or simulation closes the data gap so physical collection stops being binding - watch [B12](../../../07-indicators/diffusion/capability.md) and teleop ratios.

## 4. Operating system for field work (humans + robots + agents)

**Why now.** Most labor is not at a desk. Dispatch software from the 2000s cannot route jobs across agents that quote, robots that execute, and humans with wearables (YC RFS: physical-world OS). The corpus adds: end-to-end work data is a robotics and insurance complement.

**Build this.** Vertical OS for construction, maintenance, or fleet: scheduling across three worker types, safety rules, billing, and a data plane of how work actually happened.

**Own this.** Workflow default in a trade + the longitudinal job graph.

**Not this.** Another FSM mobile app for timesheets.

**Dead if.** OEMs and robot vendors own the OS layer with the machine and freeze you out - integrate early or pick a vertical they ignore.

## 5. Last-metre logistics autonomy (structured first)

**Why now.** Planning is solved; handling is not ([logistics](../../physical/logistics.md)). Structured environments run years ahead of open world.

**Build this.** Autonomy + ops for yards, cross-docks, or apartment lockers where structure is enforceable - sell delivered cost per move, not demos.

**Own this.** Site deployments, ops playbooks, and proprietary edge cases data.

**Not this.** A general humanoid video for Twitter.

**Dead if.** Your unit economics never beat local wages fully loaded ([cost curves](../../physical/robotics/cost-curves.md)) after real utilization, not pilot utilization.

## 6. Automated wet-lab as a service

**Why now.** Science and drug discovery are gated on experimental throughput, not on hypothesis generation ([science](../science.md), [drug discovery](../../physical/medicine/drug-discovery.md)).

**Build this.** Cloud labs or modular automation that cut cycle time for defined assay classes, with APIs agents can drive and clear chain-of-custody for regulated work.

**Own this.** Lab capacity, SOPs, quality systems, and customer protocol libraries.

**Not this.** An LLM that "designs experiments" with no bench.

**Dead if.** Big pharma and national labs open enough internal capacity that external demand is residual - still nicheable by modality.

## 7. Dense real-world sensing for industries that run on intuition

**Why now.** Energy, agriculture, construction, and logistics still operate on sparse sensors and folklore. Cheap sensors + models make "measure it densely" a company (YC RFS: data for the real world). Control follows measurement.

**Build this.** Domain-specific sensing + models with a closed loop to action (inspection robots, atmospheric platforms, in-field ag sensing) sold as risk reduction or yield, not dashboards.

**Own this.** Sensor network positions, historical fields, and the action channel (who remediates).

**Not this.** A climate newsletter with satellite screenshots.

**Dead if.** Public goods sensing (government or hyperscaler) covers your resolution and refresh for free.

## 8. Care infrastructure for aging

**Why now.** Demography is already decided ([demography](../../../09-macro/demography.md)); caregiver supply is not. Voice-native interfaces, coordination for family caregivers, and eventual home robotics are large, underserved, and unglamorous (YC RFS: AI for aging).

**Build this.** Start with coordination + monitoring + clinical escalation that seniors and families actually use; add physical assistance only when unit economics clear. Design for accessibility as the product, not a theme.

**Own this.** Relationships with families, facilities, and payers; longitudinal care data with consent; licensed clinical partners where required.

**Not this.** A youth-UI chatbot that reminds dad to take pills once.

**Dead if.** Health systems and payers vertically integrate the same stack and shut out third parties - or robotics stays too expensive for home use through the 2030s and pure software coordination commoditizes.

## 9. Skilled-trades capacity (training + matching + tooling)

**Why now.** Trades are the labor complement that still pays ([assets](../../../09-macro/assets.md)); energy build and reindustrialization increase demand; apprenticeship systems exist but do not scale with AI-era needs.

**Build this.** Accelerated training with AI feedback *grounded in real jobs*, employer offtake agreements, and tooling that raises journeyman productivity without erasing apprenticeship.

**Own this.** Employer contracts, credential recognition in the trade, and placement density.

**Not this.** A MOOC certificate in "EV charger basics."

**Dead if.** Unions and community colleges absorb the method and the employer relationships without you - partner or become infrastructure they license.

---

**Related:** [Energy](../../../01-substrate/energy.md) · [Robotics](../../physical/robotics/) · [Logistics](../../physical/logistics.md) · [Demography](../../../09-macro/demography.md) · [2032–2040](../../../04-timelines/2032-2040.md)

**Next:** [Institutions](ideas-institutions.md)