DePIN is a $10 billion-plus market cap sector. 423 tracked projects. 41.8 million devices deployed across the globe. Over $1.7 billion in venture capital poured in across 2024 and 2025 alone. By every supply-side metric, it is thriving.
But here is a number nobody wants to sit with: by 2024, only two projects launched before 2023 had crossed $1 million in annualized non-token revenue. Two. Storj, with $13M ARR from enterprise storage contracts paid in actual dollars. And Helium Mobile, with roughly $30M ARR estimated from subscriber fees. Out of hundreds of projects that raised money, shipped hardware, and celebrated node counts on Twitter.
The rest were burning tokens to pay for infrastructure that nobody was buying.
The biggest infrastructure buildout in crypto history had almost no customers.
The playbook that broke
To understand how we got here, one needs to understand the model that dominated from roughly 2017 to 2022. The “hardware-first, demand-later” playbook, which at the time looked completely rational.
The logic went like this: use token incentives to bootstrap a massive physical network. Storage, wireless coverage, compute, bandwidth or whatever the resource is. Get enough supply deployed, and demand will follow. Tokens appreciate because the network is growing. More providers join because tokens are appreciating. Flywheel spins.
Elegant on a whiteboard, yet devastating in practice.
Across the sector until 2022-23, 65 to 70 percent of token supply went to supply-side incentives, like mining rewards, hardware subsidies, staking for operators. Filecoin allocated 70% to storage miners: 55% through block rewards, another 15% in a mining reserve. Helium split roughly 65% across Proof-of-Coverage and data transfer rewards. Livepeer routed 100% of its inflation to orchestrators and delegators.
That leaves 30-35% for everything else. Protocol development. Business development. Enterprise sales. Product UX. Developer experience. The functions that actually generate paying customers. They were structurally starved from the beginning.
And how did projects measure success? Node counts. Capacity deployed. Coverage maps. Helium celebrated crossing 100,000 hotspots, then raced toward a million. Filecoin marketed 17 million terabytes of available storage capacity. These numbers looked spectacular in blog posts. They got retweeted. They got funded.
Nobody was asking the question that mattered: who is paying for this?
The numbers that should have been embarrassing
Retrospectively, the old model collapses under its own weight.
Filecoin, 2022. Over 17 million terabytes of storage capacity deployed across the network. Actual verified client data stored: over 400 petabytes. That is roughly 3% utilization. The rest was empty racks, subsidized deals, or self-stored data. I saw this firsthand. I was a Startup Operator in the Filecoin ecosystem in 2021-2022, working with 100+ teams joining the network. The technology worked. The storage proofs were real. But getting enterprises to actually store data on the network was a different problem entirely. The world’s largest decentralized storage network, and 97% of it was sitting idle.
Helium, June 2021 to August 2022. At its peak, nearly 975,000 hotspots deployed globally. A genuinely impressive physical network covering cities, suburbs, rural areas. Total monthly revenue from actual IoT data usage via Data Credit burns averaged roughly $6,500 per month over this period and was falling, dropping below $2,000 per month by late 2022. Not $6,500 per hotspot. $6,500 total. Across the entire network. That revenue would cover the electricity costs of maybe seven hotspots. The other 974,993 were running at a loss, subsidized entirely by HNT token emissions.
Livepeer, Q4 2021. The network paid out roughly $17 million in LPT rewards to orchestrators and delegators during the quarter. Transcoding fees collected from actual customers: $154,000. Fees were 0.9% of rewards. For every dollar the network earned from real usage, it paid out over $110 in incentives.
Akash, 2022. Significant compute oversupply. Leases were clearing at approximately 90% discounts relative to provider costs. When your marketplace needs to discount by 90% to find any takers, the demand signal should have been alarming.
Early Helium marketing told operators they could recoup their $450 hotspot investment in under a week. The median realized payback by 2022 was around 409 days at $10 HNT if the token held. At $3 HNT during the bear market, payback stretched to 3.7 years. And that is before you account for the fact that the network grew from 15,000 hotspots to nearly a million over this period, diluting individual rewards to near-zero for late entrants.
The analysts caught on. Messari, Delphi Digital, and multiple VC post-mortems during 2022-2023 all converged on the same conclusion: the DePIN 1.0 model created unsustainable sell pressure. Operators had fiat-denominated costs: electricity, hardware, internet. And the only way to cover them was to sell tokens. With no demand-side burn to absorb that selling, token prices fell. Falling token prices meant lower operator rewards. Lower rewards meant operators either dumped harder or switched off. The flywheel didn’t just stall. It ran in reverse.
The DePIN graveyard
When the 2022 bear market stripped away the token price appreciation that had been masking these dynamics, the weak projects had nowhere to hide.
Meson Network is the cleanest case study. They built exactly what the DePIN 1.0 playbook prescribed. Over 100,000 nodes globally, a real CDN infrastructure. But the tokenomics rewarded node presence and uptime, not paid usage. After major exchange listings in early 2024, MSN’s token price collapsed. Public development updates stopped. Social channels went quiet. A hundred thousand nodes, and no verifiable revenue to sustain them. Capacity is a vanity metric without customers.
PKT Network originally ran a Layer-1 blockchain with a bandwidth-hard Proof-of-Work algorithm. Pure supply-side incentives. In August 2024, they executed a full pivot, migrated to Base, adopted a Stake-to-Earn model designed to incentivize useful network operations rather than abstract hashing. When you abandon your entire consensus mechanism and migrate chains, you are admitting the original model was broken.
SONM tried to build a fog computing marketplace but could not reliably verify that off-chain compute was actually being performed. No verification means no enterprise trust, and no enterprise trust means no revenue. They pivoted to NFTs. That tells you everything.
The pattern repeats. Substratum burned through $13 million in ICO funds in two years and went inactive. Gladius settled with the SEC over an unregistered ICO and shut down nine months later. Skycoin sold overpriced proprietary hardware with no functional demand loop. Bluzelle found enterprise adoption too slow and pivoted to GameFi.
These are not isolated incidents. Of approximately 1,170 projects tracked by late 2024, only around 20% showed meaningful demand-side revenue, meaning at least $50,000 in annualized fees from verifiable usage. By 2025, the curated list had narrowed to 423 active projects, and the MDR share improved to roughly 30%. Better, but still: the majority of surviving DePIN projects were running on pure supply-side token incentives, with no external customers paying real money for their services.
The sector generated an estimated $72 million in total on-chain revenue in FY25. That sounds like progress until you realize it is concentrated in a handful of networks. A few winners, a lot of empty infrastructure.
Building DePIN the 2022 way wasn’t working anymore
The projects that survived did not survive by doing the same thing harder. They survived by becoming different businesses entirely.
Helium stopped being a network of IoT hotspots and became a phone company. $20/month nationwide unlimited plan, launched December 2023. T-Mobile for 5G fallback, then AT&T and Telefonica. The subscriber buys a mobile plan with a credit card and never touches a wallet. By Q4 2024, roughly 124,000 people were paying for Helium Mobile like it was any other carrier. And the demand kept accelerating. By Q4 2025, average daily Data Credit burns had grown 83.6% quarter-over-quarter to nearly $57,000 per day, with 1.6 million average daily users.
Storj killed its old Tardigrade model (crowdsourced hard drives, clunky onboarding, crypto-native everything) and rebuilt as Storj DCS. S3-compatible enterprise cloud storage. Drop-in AWS replacement with USD pricing and a 99.95% availability SLA. $13M ARR from Web2 customers who probably do not know or care that their data sits on a decentralized network.
Filecoin launched the Filecoin Virtual Machine in March 2023, later announced “Onchain Cloud.” The shift: from “rent space from miners” to “build applications on provable data.” Still early, but the direction is clear: they want to be a platform, not a marketplace.
Akash had heavy CPU oversupply and almost no demand. Then they pivoted hard into GPU compute for AI. Mainnet 6 in August 2023 enabled high-end GPU support. AkashML launched for managed inference. USDC payments, credit card top-ups via Stripe. By late 2025: 3.1 million deployments, 466% year-over-year growth. A different company than it was two years prior.
Render went from niche 3D rendering to a dual compute platform, rendering and AI workloads on the same network. Jobs priced in USD, equivalent RENDER burned on completion. They powered visuals for the Las Vegas Sphere and the Super Bowl. The studios paying for those jobs are working in fiat. The token layer is invisible to them.
What drove this? Three things.
The bear market was the push. Token prices crashed in 2022 (and looks like something similar happens right now), and the subsidized growth model collapsed with them. Projects that needed token appreciation to keep operator economics alive had to find revenue or shut down. Many shut down.
The AI boom was the pull. Starting in 2023, the generative AI wave created urgent, non-speculative demand for GPU compute. Not crypto-native speculation: actual startups and researchers who needed H100s and could not get them from AWS at any price. io.net claims it cut Wondera’s AI training costs by 75%, saving roughly $2.5 million versus AWS. Aethir claims it cut Respeecher’s voice synthesis costs by 66%. As of March 2026, AI-related projects account for 59% of DePIN market cap, up from 48% a year earlier.
And deliberate maturity was the mechanism. The survivors did not just wait for demand to arrive. They built proper cloud architecture, S3-compatible APIs, MVNO mobile plans, fiat billing rails, managed inference endpoints. They made decentralized infrastructure look like a normal cloud vendor to a procurement team. Most hid the crypto.
The KPIs shifted with them. Reports from Messari, Grayscale, Gate.io, DePINscan, and the projects themselves moved from celebrating node counts and capacity to reporting on-chain revenue, ARR, paying users, and enterprise contracts. Messari reported roughly 26% average quarter-over-quarter demand growth for DePINs in Q4 2024.
A caveat, because the data demands honesty: not all of these numbers are equally reliable. Aethir’s reported $147-166M ARR comes from unaudited corporate blog posts with some internal inconsistencies. io.net’s $22.8M is gross merchandise volume, not protocol revenue. Render does not disclose a clean USD equivalent for its burn-based metrics. Akash’s USD-denominated revenue fell 46% in Q4 2025, but that was a token price drop, not a usage drop. Native AKT-denominated revenue grew 16% in the same quarter. The sector needs standardized, audited revenue reporting. We are not there yet.
Still. The direction is clear. The projects that are winning are not the ones with the most nodes. They are the ones with the most real and dominantly enterprise customers.
So what are we actually looking at?
Here is what I personally keep coming back to.
Helium sells phone plans. Storj sells S3-compatible cloud storage with SLAs. Akash sells managed AI inference with credit card billing. Render processes 3D and AI workloads with fiat quotes. None of their paying customers need a wallet. Most do not know they are using crypto infrastructure. The token layer is invisible.
These are not DePIN projects in the way we understood the term three years ago. DePIN described a model for incentivizing hardware deployment. What these companies are doing now is delivering services: end-to-end, SLA-backed, fiat-billed, that happen to run on decentralized infrastructure underneath. It is like calling both a cell tower installer and T-Mobile “telecom infrastructure.” Technically related. Economically, completely different games.
And that difference matters. When cloud computing made the same shift, from raw infrastructure provision (IaaS) to managed services (SaaS), the service layer grew to capture a larger market at higher margins. SaaS spending hit $247 billion in 2024. The infrastructure underneath it was worth less. That pattern has repeated in telecom, in data, in every IT infrastructure sector and now AI. The service layer always eats the value.
That is the question worth sitting with. Not “what happened to DePIN 1.0”, because that we now know. The question is whether we are watching the same shift happen in real time inside crypto. And if the projects that figured out service delivery are accessing a structurally larger market than the ones still running the hardware playbook, what does that mean for where capital goes, what gets built, and who wins the next five years?
I think we are.
And in Part 2, I am going to lay out exactly what separates the projects that made the jump from everyone else still running the old playbook, and why that line is going to define this sector.