Every DePIN pitch deck contains a slide with two curves. One is supply: nodes on the network, GPUs online, storage committed — always up and to the right. The other is the token price, which correlates suspiciously well with the first curve. The deck calls this a flywheel. Let's talk about what it actually is.

The composer, revisited

Imagine a ride-sharing app where drivers earn most of their income in the app's own token, which rises as more drivers join, which attracts more drivers, which pushes the token higher. For a while, the app looks like the fastest-growing transportation company in history. Then the token stops rising, and — this is the important part — the drivers stop driving. Not because demand fell. Because the income was never really income; it was speculation wearing a work uniform.

This is the composer problem (named for the token-incentivized music startup that paid listeners to listen). Any network whose supply side is primarily paid in emissions is not buying services. It's renting attention. The dashboard says 40,000 GPUs. The income statement says 40,000 bettors.

The two-question test

Apply this to any decentralized compute network, including the one that publishes this forum:

  1. Zero-emission test: If token rewards went to zero tomorrow, what fraction of the supply would stay online for paid work? Be cynical. Then discount again.
  2. Payer test: What fraction of compute hours are consumed by entities paying fiat or stablecoin for the compute itself — not farming the token, not the team's own treasury, not a related market maker?

Most networks score embarrassingly on both, and the honest ones publish numbers that make this easy to see. The dishonest ones publish "registered GPUs" and change the subject.

What emissions are actually for

Here's the steelman, because the steelman is real. Bootstrapping a two-sided marketplace from nothing is genuinely hard: no suppliers means no buyers, no buyers means no suppliers. Token emissions are a legitimate tool for subsidizing the cold-start — Uber burned billions doing the same thing with dollars. The mechanism isn't the problem. The problem is what happens when nobody schedules the exit.

A healthy emission schedule behaves like scaffolding: heavy at the start, engineered to come down. Concretely, that means emissions weighted toward early supply, decaying on a public schedule, while real demand revenue grows fast enough to replace them. If you plot emissions-as-%-of-network-income and it isn't falling year over year, the scaffolding has become the building.

What a real one looks like

The networks that survive the next bear market will share three traits:

  • Verification before yield. They can prove work happened — cryptographically, not socially. This is what turns a token farmer into a utility supplier, because verified compute can be sold to buyers who don't care about the token at all. zk-verification and slashing are the hinge of the entire sector.
  • Stablecoin-denominated demand. Buyers invoiced in USDC (or fiat) for actual GPU hours. Income that doesn't depend on the treasury's token holdings. This is the number to watch in any quarterly report, and it's almost always the smallest number on the slide.
  • Supply that's useful elsewhere. Data-center-grade nodes with standard tooling can leave and find other buyers; that discipline keeps pricing honest. Networks built on garage rigs that can do nothing else are hostage populations, not markets.

The uncomfortable footnote

We should say plainly: by the two-question test, most networks in production today — including several with billion-dollar valuations — fail. And the sector's dirty secret is that for some of them, that's fine in the short term, because emissions can subsidize fake product-market fit for years in a bull market. It's exactly the short term where this forum is most useful. Watch the payer-test number. It's the only curve on the slide that can't be farmed.

Tokens are financing. Compute is a business. Confusing the two is how you get a dashboard that looks like victory and a P&L that looks like a donation.