Decentralized Physical Infrastructure Networks — DePIN — was one of the most discussed crypto narratives of 2023 and 2024, encompassing wireless networks (Helium), distributed compute (IO.net, Akash, Render), distributed storage (Filecoin, Arweave), mapping and geospatial data (Hivemapper, GEODNET), and several other categories where blockchain-coordinated hardware deployment was proposed as an alternative to centralised infrastructure provision. The premise was elegant: use token incentives to coordinate distributed hardware deployment at lower cost than centralised providers, capture the network effects on-chain, and create commodity infrastructure markets that competed with established providers.
By 2026, the DePIN category has been operating long enough to separate genuine revenue-generating networks from networks that exist primarily as token emission schemes with limited end-user demand. The differences between these two outcomes are visible in the data, and understanding which DePIN projects fall into which category is the analytical work that distinguishes informed crypto investing from narrative-driven speculation in this segment.
What DePIN Was Supposed to Solve
The economic argument for DePIN rests on a specific observation about infrastructure markets: building physical infrastructure (cell towers, server farms, mapping fleets) requires substantial upfront capital, and the centralised providers that dominate these markets capture the economic returns from this investment. If hardware operators in a DePIN network can be incentivised through tokens to deploy infrastructure that aggregates into a competitive network, the result could be lower-cost infrastructure provision, distributed ownership of infrastructure economics, and reduced dependence on incumbents who may behave as gatekeepers.
The challenge that the DePIN model has to solve, in every category it attempts, is that the network must provide a service that end users will pay for at a price that covers the hardware operators’ costs and returns. Token incentives can bootstrap initial supply by paying operators in tokens for their deployment, but a network that depends permanently on token emissions to maintain operator participation is not building toward sustainable economics — it is creating a token distribution mechanism that funds infrastructure deployment in the short term while accruing structural pressure on the token from continued issuance.
The successful DePIN networks are those that convert from token-emission-dependent operator economics to fee-revenue-dependent operator economics over time. The unsuccessful ones are those that fail to attract enough end-user demand to support operator economics through fees, requiring permanent token issuance to keep operators participating.
What Is Actually Working: Compute and Storage
The DePIN categories with the most demonstrable end-user demand in 2026 are distributed compute and distributed storage, both of which benefit from intersection with the AI compute buildout and the broader infrastructure shortage that AI training and inference have created.
IO.net, Akash Network, and Render Network have collectively built distributed GPU compute marketplaces that serve a real demand: AI developers and researchers who need GPU access at prices below the established hyperscaler rates and who are willing to operate in distributed compute environments with the operational tradeoffs that decentralised compute involves. The structural shortage of leading-edge AI compute has created pricing power for any supplier of GPU access, and distributed compute networks have captured a meaningful share of demand from researchers and smaller AI labs who cannot secure hyperscaler capacity at acceptable terms.
The honest assessment of this segment is that the demand is real but the unit economics are still developing. The operational complexity of distributed compute — orchestrating workloads across globally distributed hardware with variable availability, network latency, and trust assumptions — is genuinely higher than centralised compute, and the price premium that customers will accept for distributed alternatives has limits. The current pricing premium for hyperscaler GPU access creates the opportunity for distributed compute to compete; if AI compute supply normalises over the next several years, the competitive dynamic tightens significantly.
Filecoin and Arweave in distributed storage represent a more mature DePIN category with longer operational history. Filecoin’s deal flow with major enterprises — including significant data archiving contracts — provides revenue that is more clearly fee-based than token-emission-based. Arweave’s permanent storage proposition has found niche demand in NFT metadata storage, decentralised application data, and use cases where the immutability guarantee is genuinely valuable. Both networks have had to evolve their incentive structures and operator economics over time as the realities of running storage infrastructure at scale became clearer.
What Is Working But Is Narrower Than Expected: Wireless and Mapping
Helium pivoted from being primarily a LoRaWAN network for IoT devices to building a mobile carrier service (Helium Mobile) on 5G infrastructure that hotspot operators deploy. The Helium Mobile service has attracted meaningful subscriber growth — hundreds of thousands of subscribers by 2026 — by offering competitive mobile service pricing with a coverage map that combines Helium’s deployed hotspots with roaming agreements with major US carriers. This is a genuine consumer business with subscription revenue, not just a token emission mechanism.
The honest assessment of Helium’s progress is that the IoT-focused original vision did not produce the demand the network’s hotspot deployment had anticipated, but the mobile carrier pivot has found a real product-market fit at scale that justifies a meaningful portion of the network’s continued operation. The hotspot deployment that was originally framed as an IoT network has effectively become a coverage augmentation for the mobile carrier service, which is a different business model from the original DePIN vision but a functional one.
Mapping and geospatial DePIN projects — Hivemapper for street-level mapping, GEODNET for precise positioning — have built infrastructure that competes with established providers (Google Street View, professional GNSS networks) at lower cost. The customer base for these services is more specialised than the consumer mobile carrier business, which limits the absolute revenue scale, but the use cases are real and the unit economics for operators have stabilised at levels that support sustained deployment.
What Is Not Working: Tokens Without Demand
The DePIN category includes a substantial number of projects that raised capital during the 2023-2024 narrative peak, deployed hardware to operators, and have not subsequently developed end-user demand sufficient to justify the operator economics. These networks continue to operate primarily because token emissions continue to pay operators despite limited utilisation, but the trajectory is unsustainable: as token emissions decline (as they do mechanically in most DePIN tokenomics), operator participation falls if it is not replaced by fee revenue from end users.
Identifying which networks are in this category requires looking at usage metrics — actual queries, transactions, or data served — rather than at hardware deployment counts. A DePIN network with thousands of deployed nodes but minimal end-user activity is a network where token emissions are funding hardware deployment without creating the demand that justifies the infrastructure. The token’s value in such a network is structurally pressured because there is no fee revenue to support it once emissions decline.
The reluctance to identify specific underperforming networks by name in this analysis is intentional: the relevant signal is the methodology for evaluating DePIN projects, not specific predictions about which projects will fail. Investors who apply usage-to-token-emission analysis to any DePIN project can determine for themselves whether the network is on a path to fee-based sustainability or is operating as a token distribution mechanism without corresponding utility.
The AI-DePIN Intersection
The most genuinely promising development for DePIN in 2026 is the intersection with AI infrastructure demand. The AI compute buildout has created a structural shortage of GPU access at every tier, and DePIN compute networks have credible value propositions for:
AI training workloads that can tolerate the operational complexity of distributed compute in exchange for lower costs than hyperscaler rates. Distributed inference for AI applications that need globally distributed serving infrastructure (low latency to end users in many geographies) at scale. Specialised AI workloads — fine-tuning, model evaluation, RLHF data collection — that benefit from elastic GPU access without the long-term commitment requirements that hyperscaler enterprise contracts typically involve.
The economic moat for AI-DePIN is the price differential to hyperscalers and the supply availability when hyperscalers are capacity-constrained. The economic limitation is that AI workload sophistication and complexity tend to push toward managed services rather than distributed compute, and that the most demanding AI training workloads (frontier model training) require coordination and reliability characteristics that distributed compute cannot match.
For investors and developers evaluating DePIN in 2026: the category is real and growing but narrower than the 2023-2024 narrative implied. The networks that have built genuine end-user demand — primarily in AI-adjacent compute and storage, secondarily in consumer mobile and specialised mapping — are operating as real businesses with token economics that are increasingly fee-revenue-dependent. The networks that exist primarily as token emission mechanisms without corresponding utility face a slower-motion erosion that the data is already beginning to show. Distinguishing these two categories on a project-by-project basis is the analytical work that determines investment outcomes in the DePIN space.
Things That Don’t Scale Yet: The Signals That Separate Real DePIN Traction From Narrative
Paul Graham’s most useful framing for early-stage companies is about doing things that do not scale. The point is not that non-scalable activity is good. The point is that the willingness to do it is diagnostic. A founder personally onboarding contributors, manually verifying coverage maps, and handling support tickets is doing something important: they are learning what real demand looks like before optimising for growth. A founder who has automated everything before finding genuine usage is optimising a speculation.
Applied to DePIN in 2026, this framework separates the meaningful projects from the extractive ones. The compute and storage networks that show real traction do so in the form of actual paying customers who would be upset if the service stopped. The token holders are often not those customers. The customers are AI developers who need GPU time at a price point that centralised providers cannot efficiently serve at the long tail of demand. That is a real market. The DePIN layer captures value because it is solving something specific, not because the tokenomics are well-designed.
Wireless DePIN networks have a harder version of the same problem. Helium demonstrated that community-deployed coverage networks can work at scale. It also demonstrated the difficulty of converting coverage into paying enterprise customers on a timeline that sustains token holder confidence. AI data center power grid buildout creates demand for edge compute that wireless networks are theoretically positioned to serve. But the specific use cases that need low-latency edge inference are still being validated. The network exists. The application layer that makes it indispensable does not yet.
Physical infrastructure with high capital requirements and long deployment cycles presents the hardest version of the problem. Energy networks, sensor grids, and mapping infrastructure share the problem of all capital-intensive infrastructure businesses: the economics only work at scale, and getting to scale requires absorbing losses that many token-funded models cannot sustain. China deflationary transition matters here directly. The hardware deployed in physical DePIN networks is largely manufactured in China, and deflationary pressure on Chinese industrial goods cuts both ways: it lowers the cost of building the network, and it lowers the barriers for competitors to replicate it.
The AI-DePIN intersection is where the most capital is currently flowing and where the most caution is warranted. Perplexity AI valuation analysis in the centralised AI stack are already pricing aggressive market share assumptions. DePIN compute networks are being valued on top of those assumptions, implying that not only will AI demand remain high, but that a meaningful fraction will prefer decentralised supply. The second assumption is doing heavy lifting. Inference at scale requires reliability guarantees that decentralised networks have not yet demonstrated at required SLAs.
The Bitcoin Layer 2 ecosystem development path offers one useful reference. Both bets depend on building utility on top of a trust layer rather than replacing it. DePIN projects building durable coverage do so because the underlying physical infrastructure is genuinely useful independent of the token price. A DePIN network that loses 80% of its hardware providers when the token drops 50% is not a network. It is a yield farming operation that happened to involve antennas.
The projects worth monitoring are those where operators would continue operating even if token incentives were temporarily removed. That is not a large set. It is, however, the only set that actually matters.
