DePIN Compute: io.net, Render, and the Structural Question
Both networks have real GPU supply and real customers. The structural question — whether decentralized compute is a meaningful price-improvement over hyperscaler rentals at scale — has not been answered. Until it is, neither network has an unassailable moat.
I get asked the same question every week: io.net or Render? The answer is neither, until one of them answers the structural question.
The structural question
Decentralized compute networks have to clear a hurdle their competition does not — the hurdle of being meaningfully cheaper or meaningfully more available than AWS, GCP, and the Coreweave-style centralized GPU rental shops. Decentralized supply has overhead: orchestration, settlement, fragmentation, trust. That overhead has to be paid for somewhere, either by the supply side accepting lower yield or the demand side accepting higher complexity.
Right now, both networks compete on availability during AI capacity crunches and on pricing in steady state. That is a real wedge. AI training workloads have spiky demand. When OpenAI takes 80% of the H100 supply for a training run, the spillover goes somewhere — and decentralized networks have caught a lot of that.
The question is whether that wedge is structural or cyclical. If hyperscaler capacity catches up — and Coreweave and Lambda are racing to make sure it does — the spillover disappears. At that point, the only customers left on decentralized networks are the ones who explicitly want crypto-native settlement, censorship resistance, or geographic distribution. That is a real market, but it is much smaller than the current capacity-crunch market.
What separates the two networks
io.net leans on aggregation. They onboard supply from anywhere — consumer GPUs, mining-pivot operators, small data centers. Variety is the moat. The downside is heterogeneity makes orchestration harder and quality-of-service less predictable for serious training workloads.
Render leans on specialization. They started as a rendering network (Otoy origin) and pivoted to AI compute. Their supply is more concentrated, the use cases more vertically specific (rendering still pays bills), and the brand legacy gives institutional comfort. The downside is they are not racing on the same wedge as io.net — and the AI compute opportunity may not be where their best supply economics live.
The tier read
Both are Conviction on Lookout right now. Both have real revenue, real supply, and the structural question is a 2027 question, not a 2026 question.
The tier framework calls for resolution by Q4 2026:
- If hyperscaler capacity catches up to demand and the spillover dries up, both should drop to Watching
- If a third entrant materializes with meaningfully better orchestration, both incumbents lose moat to that entrant
- If institutional AI workloads move to decentralized compute (i.e. an AI lab publicly trains on io.net or Render), the structural question gets answered in favor of decentralized — and tier holds
Full sector dive coming. Until then, both stay on the tracker.