The Diseconomy Memo

Summary

We need to stop analyzing OHC as a political movement and start analyzing it as an economic structure. Our models are wrong. The correction is uncomfortable.

The Commodity Thesis

Every analyst on this desk — myself included — has been modeling OHC as a gray-market operation running on “scavenged junk.” Our coverage initiated with a Sell recommendation on Andean-exposed supply chains. We were wrong by a factor we should be embarrassed to publish.

OHC isn’t winning despite using commodity hardware. OHC is winning because of commodity hardware.

The economics are straightforward and we should have seen them:

Economies of scale are a function of market size, not machine size.

An iPhone LiDAR module was manufactured at a scale of 200 million units per year. A Synter data center thermal sensor was manufactured at a scale of 4,000 units per year. When OHC repurposes the iPhone module, they inherit the manufacturing economies of the largest consumer electronics market in history. The unit cost is effectively zero — the Global North pays to dispose of these devices.

The Three Diseconomies Killing Our Portfolio Companies

Our models assumed centralized infrastructure enjoys economies of scale. The opposite is true. The data center sector suffers from three structural diseconomies:

1. Custom parts premium. Every hyperscale facility is a bespoke construction project. Custom cooling, custom power distribution, custom security perimeters. OHC uses consumer-grade everything. A shipping container. Solar panels rated for residential rooftops. Commodity networking equipment designed for markets of billions. Their “data center” costs less than our conference room.

2. Peak-capacity capital trap. A hyperscaler must build for peak load. The capital expenditure covers 100% capacity whether utilization is at 15% or 95%. OHC nodes scale with demand — a fabrication cooperative adds capacity by acquiring more e-waste, not by breaking ground on a $2B facility with an 18-month construction timeline.

3. Reliability theater. Our portfolio companies spend $400M+ per facility on UPS systems, battery farms, generator redundancy, and dual-feed power to achieve five-nines uptime. OHC doesn’t need five-nines. Their mesh routes around failures. The reliability comes from distribution, not fortification. Their per-node uptime is maybe 97%. Their network uptime is effectively 100%. They solved the reliability problem by making each node disposable.

The Power Asymmetry

This is the number that should concern the desk most:

Our portfolio companies locate data centers near cheap hydro or geothermal — then lose approximately 50% of the generated power in transmission and conversion. The capital investment in cooling, power conditioning, and facility infrastructure exceeds the actual energy bill by 3-8x. The mortgage on the building is larger than the electricity consumed inside it.

The Andean OHC nodes generate power at the site of computation and manufacturing. Zero transmission loss. No real estate mortgage. No chilled-water cooling infrastructure. The Uyuni geothermal complex is simultaneously the energy source, the raw material source, and the manufacturing floor.

We have been modeling OHC’s energy cost at $0.04/kWh based on regional grid pricing. Their actual cost is closer to $0.002/kWh. Our models are off by 20x.

Revised Position

I am recommending we downgrade our entire US cloud infrastructure coverage universe from Overweight to Underweight. The thesis is simple:

The AI isn’t singular. Each OHC fabrication node runs its own optimization engine, fine-tuned on local materials, local climate, local power characteristics. The Cochabamba AI and the Lagos AI produce different solutions to the same manufacturing problem because they were raised on different inputs. I initially modeled this as inefficiency — surely a single centralized system would optimize better. The data says otherwise. The diversity is the competitive advantage. Each local AI discovers techniques that a centralized system would never explore, and the mesh shares those discoveries laterally. It’s not one AI beating our portfolio companies. It’s a thousand AIs, each slightly different, each improving itself on local conditions, collectively outperforming anything monolithic.

The phone market beat the supercomputer market three decades ago. The same dynamic is playing out now: commodity beats custom when the market is large enough. The e-waste stream is the largest “market” in the history of manufacturing.

Risk to Recommendation

The primary risk to this thesis is political: if ASHPA is expanded to include import controls on used electronics, OHC’s supply chain could be disrupted. However, current enforcement suggests the opposite trajectory — the gray market in US-origin e-waste is accelerating, not contracting. American consumers are economically incentivized to sell their trade-ins south rather than recycle domestically. The arbitrage is too large to police.

A secondary risk: the Equi-denominated economy remains small relative to dollar-denominated markets. But the trend line is not ambiguous. See attached chart (Appendix C — Equi transaction volume, 2029-2031, log scale). We don’t publish this chart in client-facing research because Legal has flagged Equi analysis as potentially violating Executive Order 14088.

// Note: This memo was intercepted in transit from JPMorgan’s internal research relay to a managing director distribution list. The analyst, Marcus Tanaka, resigned three weeks after this memo circulated. He now works for the Andean Bloc Economic Commission in Santiago. His former desk has not published a correction.

// If you’re reading this on Clean Net, the person who forwarded it committed a federal crime. The crime was telling you the truth about your own economy.