A Decentralized Internet Part 3: The 2026 Edition — DePIN, Decentralized AI, and the Self-Hosted Index

Blockchain Domains

A Decentralized Internet Part 1.

Capture the FlagAxel Quack

Naming without ICANN — where this series started, in 2020.

Hedge Risk: P2P Data Storage w/ IPFS

A Decentralized Internet Part 2.

Capture the FlagAxel Quack

Content addressing and the distributed file system — the 2020 installment.

This series went quiet for six years, so let me reopen it honestly. In 2021 I pinned two essays comparing decentralized video platforms to a draft and never finished the piece. Re-reading that draft in 2026 is like opening a time capsule: half the projects it assumed would matter are dead, and the things that actually mattered weren't on the list at all. So this installment does what the series always should have done: it audits the thesis. What survived, what died, what pivoted — and what the decentralized internet actually became by 2026. Spoiler: it stopped trying to be a product, and became a layer.

Graveyard and survivors: the video platforms

The 2021 snapshot was obsessed with decentralized YouTube killers. The 2026 verdict:

• LBRY is the cautionary tale. The company behind the LBRY blockchain shut down after its legal battle with the SEC. Odysee, the video front-end built on it, survived — by abandoning the LBRY network and re-platforming, and it is still operating in 2026. The chain lost its flagship; the flagship survived by leaving the chain.
• DTube and most token-first video platforms faded into maintenance mode or silence.
• PeerTube quietly won. The free, open-source, ActivityPub-federated video platform — no token, no ICO, no company — runs thousands of self-hosted instances that federate, share content, and let followers subscribe across servers, with peer-to-peer delivery reducing load when videos spike. It is the most-used decentralized video platform on earth, and it never once needed a price chart.

The lesson is consistent across every category this series touches: federation outlives tokenization. Tokens attract capital and speculation; federation attracts operators. When the music stops, the operators are still standing.

The naming reckoning (Part 1, six years later)

Part 1 argued that blockchain domains replace revocation with ownership. The 2026 audit splits the category in two:

• Handshake is retreating. Namecheap — once its largest retail registration channel — ended Handshake TLD support effective June 10, 2026, after selling its Handshake marketplace Namebase in January and winding it down by early June; the ViaBTC mining pool closed around the same time. More than 12.7 million TLDs remain registered on-chain, but the retail layer is collapsing around them.
• ENS consolidated. Where naming had real network effects — wallet handles, profiles, logins inside a living ecosystem — it won outright, and is now scaling its resolution onto dedicated infrastructure. Where naming was pure speculation — buying root zones as digital land — it died.

Naming won where it was a handle, and lost where it was a lottery. The through-line from my identity essay this year: a name is only as sovereign as the ecosystem that resolves it.

Storage grows up (Part 2, six years later)

In 2020 I distributed my own portfolio site over IPFS and explained content addressing as a curiosity. By 2026, content addressing is simply how web3 stores things — the plumbing underneath NFTs, frontends, datasets, and increasingly AI artifacts. The landscape matured into distinct tools with real price tags:

The honest summary: decentralized storage stopped being an ideology and became an engineering decision — cost, latency, permanence, and retrieval guarantees, compared line by line against S3. That boring sentence is what winning looks like.

The physical turn: DePIN

Here is what neither earlier installment could have anticipated: decentralization went physical. DePIN — Decentralized Physical Infrastructure Networks — is a ~$15 billion sector in 2026, and for once the revenue is real: Solana's DePIN projects alone reported around $150 million in a single month (January 2026).

Two examples carry the thesis:

• Helium. I have to confess something: in 2020 I wrote Helium off in my notes as keine sinnvolle IoT Investition — not a sensible IoT investment. The 2026 scoreboard: over 400,000 hotspots, $12 million in Q1 revenue (up 45% year-on-year) after the network's move to Solana, and — the headline — Helium Mobile signed roaming partnerships with three tier-2 telecom carriers: the first decentralized network integrated into traditional telecom infrastructure. I was wrong, and the interesting part is why: the technology didn't change as much as the sequencing — the network found buyers (data, roaming, coverage) instead of waiting for them.
• Hivemapper. The most surprising project in this article. Mount a dashcam, drive normally, and your street-level imagery feeds a global map that mapping companies and autonomous-vehicle researchers pay for in HONEY tokens. In 2026 Hivemapper supplies navigation data to Lyft and Volkswagen. Google Maps, crowd-built, one windshield at a time — and the drivers own the network.

The pattern: DePIN works where there is a physical asset people already own (a car, a roof, a phone plan) and a buyer who already pays for the output (coverage, imagery, data). Where those two exist, decentralization is not a pitch — it is a procurement change.

The decentralized AI stack

The second thing 2021 could not have predicted: the decentralized stack became the AI stack. Decentralized AI infrastructure crossed $15 billion in aggregate market cap in early 2026, and the layers map cleanly:

• Bittensor runs the model layer: a network of 80+ specialized subnets, each an incentive market for a different AI task, at $1.8B+ market cap — with an ETF decision looming as of August 2026, which would be the sector's institutional crossing of the Rubicon.
• Akash is the open-source decentralized cloud — a GPU compute marketplace where capacity is bid, not priced by a hyperscaler's pricing page.
• Render pivoted from GPU rendering for graphics into AI compute; io.net aggregates GPUs from data centers, miners, and consumer devices into a single AI/ML cloud.
• Livepeer — the decentralized video infrastructure from my 2021 draft — completed the most telling pivot of all: from video transcoding to AI video inference on a permissionless GPU network. The decentralized video platforms of 2021 didn't beat YouTube; their infrastructure got absorbed into the AI economy instead.
• Data marketplaces (Vana, Ocean) close the loop: training data itself becomes a tradeable, consent-scoped asset.

Read through this blog's lens, the sovereignty argument is straightforward: the most concentrated resource of the decade is GPU compute, and a decentralized market for it is the only structural counterweight to three hyperscalers pricing intelligence itself. Whether the token incentives survive contact with real enterprise procurement is the open question — but the infrastructure, unlike the 2021 video platforms, is being paid for by workloads that exist regardless of the tokens.

The sovereign layer: self-hosted everything

The last layer needs no token at all, and it is the one I trust most — because it is the one you can run yourself.

• Bitmagnet is the emblem. A self-hosted BitTorrent indexer that crawls the DHT network directly (BEP51), classifies content with machine learning, and serves it through a web UI and GraphQL API with integration into the Servarr stack (Sonarr, Radarr, Prowlarr). No central tracker, no index website, no company to subpoena — discovery without a registry. It is the BitTorrent answer to the question Part 1 asked about DNS: what happens when the index itself becomes peer-to-peer?
• Bluesky's AT Protocol brought the same logic to social: past 30 million users in 2026, with self-hostable Personal Data Servers — your identity and posts live on a server you can run, and your account can migrate between providers while your handle stays bound to your own domain. Portable identity at mainstream scale is new, and it belongs in the same sentence as the keys-and-names argument of my identity essay this year.
• The supporting cast is mundane and magnificent: SimpleX for messaging with no user identifiers at all, Jellyfin for self-hosted streaming, SearXNG for metasearch without a profile of you (one runs in my own stack), The Graph as the "Bitmagnet of blockchains" — decentralized indexing for on-chain data.

This is also where the series lands personally: most of these pieces — indexers, search, local models, media — now run in my own lab, on the bare-metal control plane I documented in my Talos essay this year. The decentralized internet of 2026 is not a website you visit. It is a stack you operate.

Final Thoughts

Six years, three verdicts. The naming layer consolidated where it had network effects and died where it had speculation. The storage layer stopped being a manifesto and became a line item. And the frontier moved twice — first into the physical world (DePIN: Helium signing carrier deals, Hivemapper selling map data to Volkswagen), then into intelligence itself (Bittensor's subnets, decentralized GPU clouds pricing an alternative to the hyperscalers).

The 2021 draft assumed the decentralized internet would arrive as a better YouTube. It never does. It arrives as a layer: federation for the social, content addressing for the archive, DePIN for the physical, decentralized compute for the intelligence, and self-hosting for everything you refuse to rent. Platforms die; layers compound. Plant your flags accordingly.

The website and the information contained therein are not intended to be a source of advice or credit analysis with respect to the material presented, and the information and/or documents contained on this website do not constitute investment advice.