DePIN's Real Numbers: Akash's Verifiable Spend vs. Grass

The broader crypto market is stuck in fear territory. Altcoins have suffered their worst drawdown of the current cycle, with a large share of tokens trading well below their prior highs. But DePIN, Decentralized Physical Infrastructure Networks, is treated as the sector where usage is finally outrunning speculation.

The sector generated $72 million in on-chain revenue in 2025, and that report's own framing matters as much as the total: growth came from "utility and efficiencies, not subsidies," a contrast with 2021, when high token inflation subsidized usage that never paid for itself. That distinction, subsidized activity versus paid demand, is the one most DePIN coverage skips. It is also the one that separates Akash's numbers from Grass's, and it is what separates a verifiable onchain usage figure from a marketing claim wearing the same label.

The AI Compute Crisis Meets Decentralized Solutions

Global AI data center capital expenditure is projected to reach $400-450 billion in 2026. Inference workloads, the computational cost of running trained AI models, now account for two-thirds of all compute demand, up from one-third in 2023. That figure describes total AI infrastructure spending, not DePIN's share of it: the relevant question is which decentralized networks can show actual customers paying for a slice of it, not just token holders being rewarded for showing up.

DePIN protocols pitch compute, storage, and bandwidth at a lower cost than centralized cloud providers, positioning themselves against the capacity constraints implied by that $400-450 billion spending figure. But "offering" it and "selling" it are different claims, and the gap between them is exactly what the next two sections measure.

Akash: The Numbers Are Onchain

Akash Network crossed $5 million in all-time compute spend in the first 90 days of 2026, according to the network's own Q1 2026 report. That figure is a cumulative lease total pulled from onchain deployment data, the same ledger anyone can query independently; it is not a token-reward figure. The report also notes that AkashML, the network's inference marketplace, was processing 1.7 billion tokens per day through OpenRouter by the end of the quarter, a usage metric distinct from spend and one that is growing the addressable demand for the network's GPUs.

The bigger structural change: Akash's Burn-Mint Equilibrium (BME) went live on March 23, 2026, roughly two weeks after this article originally published, following onchain Proposal #318 and a two-week incentivized testnet. Under BME, tenants burn AKT to mint ACT, a non-transferable, USD-pegged compute credit; providers get paid in stable terms regardless of market swings. When AKT's price rises between the moment a tenant burns tokens and the moment a provider is settled, fewer AKT are minted back out than were burned, and the difference leaves circulating supply permanently. Akash's own explainer walks through a worked example: if AKT moves from $1.14 to $1.50 during a lease, 210 AKT are burned net from that single transaction. The network has not published a network-wide monthly burn forecast, so a total deflation estimate is not something this article can responsibly cite; the mechanism itself, tying token scarcity directly to paid compute volume rather than to a fixed schedule, is the verifiable part.

Render Network: Enterprise GPU Power

Render has expanded from visual effects rendering into AI compute. The project's own network dashboard reports roughly 5,600 nodes and over 78 million frames processed since inception, with enterprise GPU support that now includes NVIDIA H200 and AMD MI300X chips. The dashboard does not publish a node-use rate; where Render's own marketing cites a high self-reported figure, that number has no independently queryable counterpart the way Akash's lease ledger does, worth remembering when comparing the two.

Render's pivot from rendering to AI inference, if the self-reported use rate holds up to independent scrutiny, points to a revenue model that scales with AI demand rather than with speculative token flows.

Grass: A Real Business Model, Numbers You Cannot Audit Yet

Grass converts idle internet bandwidth into structured web data for AI model training, a plausible business model. The company's own site claims it is "trusted by over 8.5 million users worldwide", and Grass's own July 2026 network call reports $17-18 million in H1 2026 revenue (the two figures appear in different sections of the same page, an inconsistency Grass has not corrected), with the company projecting $65-75 million in full-year 2026 revenue from training data alone.

Neither figure is independently verifiable the way Akash's lease data is. Grass's user count is a registered-account total, not a count of currently active, unique bandwidth-sharing nodes, so it says nothing about how much duplication, dormancy, or multi-accounting sits inside it. DefiLlama, the tracker we checked for onchain protocol fees across the DePIN sector, lists Grass with no fee or revenue module at all (DefiLlama fees overview, checked August 17, 2026, 2,578 protocols tracked), meaning the revenue figures trace back to the project's own reporting rather than to that tracker. That does not make Grass's business fake. It makes its DePIN revenue claim exactly as trustworthy as the company disclosing it, and no more verifiable than any other self-reported DePIN revenue figure, which is a categorically different thing from a number a buyer can verify onchain.

This is the pattern worth generalizing: when a DePIN project reports "revenue," ask whether the figure comes from a payments ledger a third party can query, or from the project's own dashboard. Akash's lease ledger is auditable in the first sense today. Filecoin's Solstice mechanism (below) is designed the same way, tying its burn rate directly to revenue, but it is still a proposal that has not shipped, so for now it is a commitment to future auditability rather than a number anyone can check yet. Grass's headline numbers, as published today, are not independently verified by the sources checked for this article.

Which DePIN Revenue Claims Are Actually Verifiable

The 2025 market reset eliminated some low-utility DePIN projects. Whether the sector as a whole shifted toward paying customers, independently measurable usage, and genuine cost advantages over centralized alternatives is a hypothesis this article cannot confirm across every protocol below: Akash has onchain lease data a third party can query and Helium's hotspot count is verifiable on its onchain registry, though Helium's revenue figure is report-derived rather than independently queryable; Aethir's and Grass's figures are self-reported with no independent tracker, and Render discloses no revenue or spend figure at all. The table separates which protocols have individually evidenced the shift and which have only claimed it.

Revenue multiples across the sector vary widely by protocol and are not something this article can responsibly compress into one range without a dated, sourced valuation report. What is verifiable is the underlying asymmetry: a multiple on an on-chain-verifiable number is a different bet than the same multiple on a number nobody outside the project can check.

For a deeper breakdown of which DePIN projects actually generate sustainable revenue, see our DePIN Revenue Reality analysis.

DePIN vs. Pure AI Narrative Tokens

Not all "AI crypto" is created equal. DePIN protocols differ from pure AI narrative tokens in fundamental ways.

DePIN projects like Akash and Render coordinate physical infrastructure that individual providers own and contribute to the network, rather than owning that infrastructure themselves. Akash's compute leases are paid for on an onchain ledger a third party can query; Render reports usage (78M+ frames rendered, 5,600 nodes since inception) but, as the table above shows, has not published a revenue or spend figure, so customer income for Render is not yet something this article can independently verify. A token without a comparable usage or revenue ledger, like Akash's onchain lease record, or without even a registered-infrastructure count, like Helium's onchain hotspot registry, has no equivalent evidence to fall back on if the market corrects; whether that gap actually predicts which tokens lose their premium is a hypothesis this article does not test.

The distinction matters for investors weighing that gap. As we explored in our analysis of AI agents as a breakout narrative, the untested hypothesis above points the same direction: real infrastructure underneath the narrative, and increasingly, real infrastructure that can prove its own usage.

This also connects to the broader trend of AI agent payment infrastructure: autonomous agents need actual compute resources, and DePIN protocols are positioned to provide them, provided their revenue claims can withstand the same scrutiny their token charts get.

What to Watch: Catalysts Ahead

Several developments will keep testing which DePIN revenue claims hold up.

Resolved March 23, 2026: Akash's BME launch went live as planned, moving compute payments onto a burn-mint mechanism tied directly to paid usage rather than to a fixed emissions schedule. Whether the resulting AKT burn becomes large enough to matter now depends on two things: how much paid compute volume the network can sustain, a number that will keep showing up in Akash's own quarterly reports, and how AKT's price moves between burn and settlement, since a rising price means fewer AKT are minted back out than were burned.

Filecoin's Solstice proposal: rather than a vague "pivot to monetization," Filecoin has a specific, published mechanism: its Solstice proposal would ramp the network's consensus-reward share from 95% down to a 50% floor over roughly nine quarters, redirecting the freed issuance into a service stream that pays providers who bring in paying customers. The proposal states plainly that under the new model, "the burn rate is directly tied to revenue," the same emissions-follow-demand principle behind Akash's BME, applied to storage instead of compute.

Ongoing: The broader blockchain-in-telecom market is projected to grow from $1.19 billion in 2024 to $19.96 billion by 2030, a 60% CAGR, according to TechSci Research.

Growing: Render's AI inference capabilities and enterprise GPU integration. Bittensor's 28% surge shows one AI-compute token's price moving sharply around a real-technology narrative; whether decentralized AI compute narratives carry that kind of market appeal more broadly is an untested hypothesis this single episode does not establish, and appeal and auditability remain separate questions regardless.

The Bottom Line

DePIN's real story in early 2026, on the one in-quarter example this article has, was Akash putting paid-usage mechanics live with its March 23 BME launch; its $5M figure is cumulative since launch, not a period split, and whether the turn is a sector-wide pattern is unconfirmed, not a measured market-beating return and not yet a confirmed sector-wide pattern: Akash's onchain lease ledger lets a third party check its usage figures, though the cumulative total contains pre-BME activity and cannot isolate the shift's outcome, and Helium's onchain hotspot registry backs its registered-infrastructure count, though that registry does not by itself establish usage, and Helium's revenue figure is report-derived rather than independently queryable; Aethir's and Grass's growth claims and Render's usage figures do not yet have that kind of independent backing, whatever the underlying businesses turn out to be worth. Some projects are redesigning their own token emissions to follow paying demand rather than raw supply growth: Akash's BME already does this live, and Filecoin has proposed the same principle for Solstice, though that one has not shipped yet. Live or proposed, it is as strong a signal of intent to prove real usage as a token chart can offer.

Others are still asking to be taken at their word. Grass's bandwidth-to-training-data model serves a real-sounding use case, and 8.5 million self-reported registered accounts is a real number worth taking seriously, but it is not a verified count of active users, and DefiLlama, the tracker we checked, does not track a Grass module, so no independent source among those checked verifies that count. Until both the user count and the revenue can be checked against a source independent of the company, its self-reported figures and projections belong in the same category as any other unaudited claim: plausible, not proven.

For investors weighing DePIN against the current fear cycle, the filter is simple: ask whether a project's DePIN revenue number can be verified independently or exists only on a dashboard the project itself controls. Decentralized compute is competing for a slice of a large addressable market: AI infrastructure buildout is projected at $400-450 billion in capital expenditure for 2026 alone, a total-spending figure that says nothing on its own about DePIN's actual share of it. The open question is which protocols can prove their share of it, not just claim it.

Disclaimer: Nothing in this piece is financial advice. Cryptocurrency investments carry significant risk. Always conduct your own research and consult with a qualified financial advisor before making investment decisions.