Curated or open: the supply tradeoff every compute network eventually faces

Hand-pick every machine and guarantee quality, or open the doors and chase scale. One doesn't grow. The other fills with noise. The networks that matter are trying to do something harder than picking a side.

Disclosure: we're building a venture in this category. The analysis sticks to attributed figures, and we flag our own stake where it bears on the argument.

Every compute marketplace arrives at the same fork, usually around the time it starts caring about revenue. Down one road: curate. Vet every supplier, approve every machine, guarantee the quality. Down the other: open up. Let anyone plug in, chase the scale, sort it out later.

Both roads have a cliff at the end. The interesting work is in refusing to drive off either one.

What curation buys, and what it costs

Curate, and quality becomes a promise you can keep. A vetted pool of professionally-run machines behaves. It shows up, it runs the job, it stays busy. The proof is in the numbers: curated, enterprise-supplied networks report utilization in the 38 to 80% range (House of Chimera, Q1 report, Apr 2025; BlockEden, Mar 12 2026), against the roughly 2% that idle-aggregation pools manage on registered supply (per the flagship analysis). Vetting works.

The major marketplaces have institutionalized this. One runs a "Certified Data Center Partner" program: a minimum of five flagship-class GPU servers, a professionally managed facility, a registered business, identity verification, and a signed hosting agreement, reviewed in about two business days (a major GPU marketplace, data-center application, 2026). Clear the bar and you get a trust label, priority placement, and a path into a separately-certified "Secure Cloud" tier for compliance-sensitive work (a major GPU marketplace, 2026). It's a serious, well-built gate. Credit to them for it.

But a gate is a governor on growth. Every machine that needs a human to approve it is a machine that doesn't onboard at internet speed. Curation guarantees quality by capping the rate at which supply can arrive. That's the cost, and it's structural.

What openness buys, and what it costs

Open the doors and scale stops being the problem. Supply floods in from everywhere idle. The catch is that quality stops being guaranteed, and the noise can get loud.

This is where the cautionary tale lives. One large network's own review reported roughly 2,752 verified GPUs against a registered base it has put north of 327,000, which is under 1% verified (a leading decentralized GPU network, 2025 year-in-review, Jan 2026). And the open door has been pushed on: a 2024 incident saw a single RTX 4090 split to present as roughly 1.8 million virtual GPUs (per Messari's report, summarized in our competitive file). The machine count soared. The trustworthy supply didn't move.

To be fair, that same network responded with real defenses: proof-of-work checks, time-lock staking of its token per chip, a multi-hour stress test, uptime requirements, and slashing for bad behavior (that network's own documentation, 2026). Openness without guardrails is noise. Openness with guardrails is a harder, better engineering problem, and they're working it.

Curate and you guarantee quality but cap growth. Open up and you unlock scale but invite noise. The whole game is building a third thing: open supply that behaves like curated supply.

The third road, and why it's the only one that matters

The binary hides one move. The tradeoff between quality and scale is not a law of physics; it's a consequence of doing trust by hand. The moment you can verify a machine's work automatically, instead of vetting its operator personally, the fork starts to dissolve.

If a buyer can trust a result without trusting the person who returned it, a network can safely admit machines it never met. Verification becomes the thing that lets open supply behave like curated supply: the scale of the open road with the quality of the curated one. We walk that mechanism in detail in Proof you can't fake: how verifiable inference changes who you can rent compute from, and the privacy half of the same problem in What "isolated compute" actually means. The supply-quality gap that makes all of this urgent is the flagship, The sub-1% problem in decentralized compute.

That third road is the one we chose to build. Reputation-scored trust levels, from Bronze to Platinum, that gate a provider's earnings; statistical verification of the work itself; a network that stays open to new supply without surrendering quality. Not curated or open. Open, made to behave like curated. The harder thing, and the one we think is worth building. (That venture is Griddly, and our stake in it is the conflict we flagged up top.)

Nothing here is an offer to sell a security or investment advice.