DePIN is Dead. Long Live Infrastructure-as-a-Service. · Konstantin TkachukSkip to content<br>ContentsEvery infrastructure sector in modern history developed using the same playbook. Someone builds raw capacity ahead of demand. The raw resource commoditizes and gets cheaper forever. And the money moves to whoever sells guaranteed, contracted service on top of it. So far, cloud ran it, telecom ran it, even chip makers did it to an extent. Crypto infrastructure is attempting to repeat it right now.<br>Motivation to continue this research came from a news article a couple of weeks ago. Storj, one of the actually working service businesses in decentralized storage, went into Chapter 11 Bankruptcy in July while its own restructuring announcement called the business underneath &ldquo;strong and right-sized,&rdquo; held back by &ldquo;legacy obligations from an earlier chapter,&rdquo; and said it expects to keep operating without interruption [1]. A crypto company restructuring like a normal business, to keep serving customers like a normal business. And that is not a single case. At least three different companies from legacy DePIN space were doing some form of restructuring, selling or splitting the business to get decoupled from the token component just this summer [1].<br>That&rsquo;s what I want to talk about in Part 2 of a series. Part 1 argued the old DePIN model is dead. DePIN thesis was in building Networks that incentivise ordinary people to crowdsource hardware, bandwidth, and storage, trying to turn resources none of us can negotiate with individually into services someone will actually buy. Part 2 is the evidence of the OG model stopping to work. The pattern, the winners, and where we are on the curve.<br>I. The Pattern<br>The cloud ran the script<br>Amazon launched EC2 in 2006 selling raw compute by the hour. Rent a server, configure it yourself. Table 1 shows the infrastructure layer financials for past twelve years. The elephant in the room is the fact that AWS is 18% of Amazon&rsquo;s revenue but 57% of its operating income, and Amazon is about to rebuild the infrastructure layer all over again with roughly $200B of CapEx in 2026, driven by AI [2][3].<br>Table 1. AWS by year: the infrastructure layer compounding.<br>YearAWS net revenueAWS operating incomeAWS share of Amazon operating income2013$3.1Bn/an/a2015$7.9B$1.9B~84%2017$17.5B$4.3B>100% (international losses exceeded North America retail profit)2019$35.0B$9.2B63%2021$62.2B$18.5B74%2023$90.8B$24.6B67%2024$107.6B$39.8B58%2025$128.7B$45.6B57%Source: Amazon 10-K segment reporting, 2013-2025 [2][3]<br>The services built on top are the ones getting the value accumulated from the infrastructure, not the infrastructure companies themselves. Netflix pays Amazon an estimated $1B+ a year for effectively all its computing, about 2 to 3 cents of every dollar Netflix earns (an estimate; Netflix delivers video over its own CDN, AWS runs the compute and storage) [4][5]. Snowflake books $4.5B in product revenue and owns zero data centers. Datadog $3.4B, Zoom $4.9B, Airbnb $12.2B, all running on rented infrastructure their customers never see [6][7]. The whole market is renting raw servers by the hour (Infrastructure as a Service or IaaS) against finished software delivered over the internet (Software as a Service SaaS). Total public cloud spending hit $595.7B in 2024, is forecast at $723.4B for 2025, and passes $1 trillion in 2027 on Gartner&rsquo;s projection [8].<br>Diagram 1. The service layer and its foundation: worldwide IaaS vs SaaS spending, $B. IaaS is Gartner's vendor-revenue actuals series through 2024; SaaS is end-user spending, with the 2025 forecast dashed. Different Gartner methodologies, so compare shapes, not exact levels [8].The service layer is structurally larger and the absolute gap keeps widening, but the infrastructure layer is currently the faster-growing one. GPUs and power are scarce again, so profits are flowing down to the infrastructure. Likely, if we would be able to isolate the GPU demand growth, we would see overall cloud commoditisation even further. Also if the trend continues, GPUs and power will be an example of the same commoditisation in a couple of years.<br>Telecom runs the same script, just different wires<br>T-Mobile paid up to $1.35 billion for the parent company of Mint Mobile, a phone brand that owns no towers, no spectrum, no radios, but with Ryan Reynolds [9]. Over 2,100 such carriers operate in about 100 countries, an $89-99B global market of phone companies that rent everything (the industry calls them MVNOs) [10].<br>Table 2. Towers vs wireless services, US, 2010-2024.<br>YearBig-3 tower operators combined revenue (AMT + CCI + SBA)US wireless service revenue (CTIA)ARPU (CTIA)Ratio (service : towers)2010$4.5B$159.9B$47.5336x2015$10.1B$191.9B$44.6519x2020$16.0B$189.9B$35.3112x2024$19.4B$224.9B$33.3612xSources: company annual reports [11][12][13]; CTIA Annual Wireless Industry Survey [14]. Tower revenues include international operations, CTIA is US-only;...