Web3’s long‑awaited “TCP/IP moment” is still ahead of us. The next real breakthrough in crypto will not come from yet another marginally faster L1 or a slightly more efficient rollup. It will come from something deeper: a new, decentralized internet layer that does for web3 what TCP/IP did for web2 — but without sacrificing the core values of blockchains.
Right now, the industry is trying to scale an internet of value on top of networking assumptions from the 1980s. That made sense when the goal was delivering emails and web pages through trusted intermediaries. It fails spectacularly when the task is to support global settlement, DeFi at trillion‑dollar volumes, billions of autonomous AI agents, and a worldwide mesh of DePIN and sensors.
No amount of bigger blocks, higher TPS claims, or more creative rollup architectures can fully compensate for a broken foundation. Web3 doesn’t just need “faster chains”; it needs its own equivalent of TCP/IP: a decentralized Internet Protocol designed from first principles for trustless, adversarial, permissionless environments — and engineered to scale.
—
Why TCP/IP mattered — and why web3 never got its version
Before the Internet Protocol existed, networks were isolated islands. Mainframes, terminals, and local networks could talk inside their own bubble, but there was no common standard for routing packets between them. TCP/IP changed that, creating a universal fabric for data exchange between any two points on the planet. It turned fragmented systems into a coherent internet and became one of the three foundational pillars of web infrastructure, alongside computation and storage.
Every mainstream web2 application — from email to streaming platforms — rides on TCP/IP. It doesn’t care which ISP, data center, or country sits in the middle. The protocol abstracts the complexity away and guarantees that packets find their way.
Web3 never got such a universal fabric. Instead, each blockchain designed its own bespoke networking stack: gossip protocols here, Turbine or Narwhal there, custom mempools and data availability sampling elsewhere. None of these approaches is standardized, and none of them is designed as a global, shared layer. They are local hacks, not an internet.
The result: blockchains scale “locally” — inside their own networks — but stumble and fragment when you try to scale them “globally,” across chains, geographies, and use cases. This is the exact opposite of what TCP/IP achieved.
—
The repeated mistake: optimizing execution while ignoring networking
The industry is obsessed with execution throughput: more transactions per second, more parallelization, more efficient virtual machines, more rollups, more shards. These are valuable improvements, but they all assume that the underlying network will magically keep up.
In practice, the network is already the choke point:
– Block propagation slows or stalls at peak load.
– Mempools and gossip layers become congested and inconsistent.
– RPC endpoints centralize access because node networking is inefficient.
– DA layers struggle to deliver data fast enough to every participant.
All of this mirrors the pre‑TCP/IP era, when scaling meant building ever larger, more specialized, closed systems. That worked, up to a point. Then the internet changed the game by offering a universal, open fabric that everyone could build on.
Web3 is still living in the “before” phase. To move past it, we need a networking layer that is:
– Fast enough to support global, real‑time applications.
– Trustless so that no intermediary can censor or rewrite data.
– Fault‑tolerant under attacks, failures, and partitions.
– Modular so it can be plugged beneath any chain or rollup without a full redesign.
– Universal so it becomes the common backbone every chain can rely on.
This is what a decentralized Internet Protocol should provide.
—
Two decades of research: can decentralization match centralized performance?
A central question of modern systems research is whether decentralized architectures can ever match the speed and reliability of centralized services — and whether this can be proven mathematically, not just claimed in whitepapers.
Over the past twenty years, one line of work has married two fields that historically lived apart:
– Network coding theory, which optimizes how information moves through networks by encoding data, not just forwarding it.
– Distributed algorithms and fault tolerance, drawing on foundational work in consensus and Byzantine fault-tolerant systems.
The conclusion from this research is striking: decentralized systems can, in principle, hit performance levels comparable to highly optimized centralized infrastructures. But to do so, they must rethink data movement from the ground up. Simply grafting blockchains onto legacy networking models is not enough.
A key piece of this puzzle is Random Linear Network Coding (RLNC) — a technique that encodes data into linear combinations as it flows through the network. Instead of each node just forwarding raw packets, nodes send coded combinations that collectively carry the same information. Any receiver that collects enough of these coded packets can reconstruct the original data.
Mathematically, RLNC has been shown to be optimal (or close to optimal) in many settings for throughput, robustness, and reliability across decentralized networks. It turns the network from a set of fragile routes into a coded fabric where information flows more like an electrical field than a sequence of packets along narrow pipes.
Once blockchains appeared, the implication was obvious: if we want a decentralized web that genuinely scales, we can’t keep relying on the internet stack built for banks, CDNs, and social networks. We need a new layer, rooted in coding theory and distributed algorithms, that is explicitly designed to survive faults, attacks, and adversarial behavior — while still performing at global scale.
—
Not a replacement, but an enabler — like TCP/IP
A decentralized Internet Protocol is not meant to rip out everything that exists. TCP/IP did not replace data centers or storage systems; it made them interoperable and massively more useful. Likewise, a new web3-native protocol would sit beneath existing and future blockchains, acting as a shared bandwidth grid.
This protocol would:
– Accelerate block propagation across heterogeneous chains.
– Improve DA retrieval and state access without altering consensus rules.
– Distribute load intelligently so local congestion becomes rare.
– Allow chains to benefit from global capacity, not just their own nodes.
When it works, it should be nearly invisible — much like people rarely think about TCP or IP today. But its impact would be enormous: it would shift web3 from an archipelago of islands to a true network.
—
The four macro trends exposing web3’s bottlenecks
The timing is critical. Decentralized systems are hitting their limits just as the world is demanding far more from them. Several macro trends all converge on the same bottleneck: networking assumptions inherited from centralized systems.
1. Fragmented L1s and L2s: local scale, global failure
We now have hundreds of L1s and a growing ecosystem of L2s and rollups. Each chain can tune its consensus, block size, and execution environment to hit impressive benchmarks under lab conditions. But these networks do not exist in isolation. Assets, users, and applications span multiple chains. Liquidity is fragmented. Interoperability is messy and often constrained by bridges, oracles, and trusted middle layers.
The minute these chains must coordinate in real time — for cross‑chain swaps, shared liquidity pools, unified identity, or global settlement — the underlying networking constraints surface. Propagation delays create inconsistencies. Global ordering becomes complicated. Security assumptions fracture.
Without a universal networking fabric, blockchains remain like national power grids that can’t interconnect: each might function locally, but you can’t rely on them as one coherent system.
2. Tokenization and DeFi at trillion‑dollar scale
Tokenization is moving beyond experiments into traditional finance: real‑world assets, government bonds, securities, and payment rails are beginning to migrate on‑chain. DeFi is no longer just about yield farming; it is evolving toward full‑fledged market infrastructure.
But global capital markets cannot operate on networks where:
– Block propagation lags under bursts of volume.
– High contention leads to MEV games and unpredictable execution.
– RPC bottlenecks mean that most users and institutions effectively depend on a small cluster of centralized providers.
For DeFi and tokenization to handle trillions in value, the underlying network must behave more like a global utility and less like a congested local bus. A shared coded network that spans chains could absorb traffic spikes, smoothing out local surges by redistributing load across a broader pool of bandwidth and nodes.
Traditional finance handles peak load by building massive data centers and private networks. Web3 cannot depend on that playbook without re‑centralizing control. It must achieve resilience and throughput through coded distribution, not megacenters.
3. DePIN at planetary scale
Decentralized physical infrastructure networks (DePIN) aim to coordinate real‑world resources: sensors, connectivity, energy, storage, mobility, and more. At small scale, this is manageable. At global scale, the challenges multiply: billions of devices, intermittent connectivity, heterogeneous hardware, and adversarial actors trying to game incentives.
Legacy networking stacks assume relatively stable, centrally managed infrastructure. A global DePIN mesh is the opposite: chaotic, dynamic, and only loosely controlled.
A decentralized Internet Protocol designed with coding at its core would allow data from millions or billions of nodes to flow reliably, even when individual connections drop, misbehave, or attack the system. Rather than each project reinventing a brittle custom overlay network, a common fabric could provide robust routing and aggregation from the edge to the blockchain and back.
4. Decentralized AI and swarms of agents
The rise of AI agents and autonomous systems introduces a new class of web3 workload. Imagine millions — or billions — of agents executing transactions, accessing oracles, negotiating contracts, and interacting with DeFi and DePIN systems in real time.
Centralized AI platforms can handle this by building proprietary, centralized backbones. A decentralized AI ecosystem cannot. It needs a network that can:
– Serve and verify model outputs in a trustless way.
– Coordinate agent communication across many chains and environments.
– Tolerate high churn and adversarial behavior.
The current patchwork of blockchain networking is already strained by human‑driven activity. It is nowhere near ready for planetary‑scale machine‑driven demand. Without a new protocol layer, decentralized AI will be forced into centralized choke points — undermining the promise of openness and trustlessness.
—
What a decentralized Internet Protocol could look like in practice
Conceptually, a web3‑native Internet Protocol would sit beneath consensus and execution, acting as a programmable, coded bandwidth grid:
– Universal fabric: Every participating node, regardless of which chain it serves, becomes part of a shared coded network. Data moves not as raw packets, but as coded combinations optimized for throughput and resilience.
– Consensus‑agnostic: The protocol does not replace consensus; it accelerates and secures the flow of data that consensus algorithms depend on.
– Composable for chains and rollups: L1s, L2s, and rollups can plug into this fabric via standardized interfaces, much like applications connect to TCP/IP sockets today.
– Adaptive routing: Instead of static paths, the network uses coding to dynamically optimize information flow around congestion, failures, or attacks.
– Security‑aware: The protocol is designed with adversaries in mind, aligning with Byzantine fault‑tolerant assumptions rather than assuming honest majority in the network layer.
When block propagation, DA retrieval, and state access are all powered by this kind of coded backbone, the performance ceiling of decentralized systems shifts dramatically upward. It becomes possible to imagine web3 infrastructure rivaling centralized cloud backbones not just in theory, but in deployed practice.
—
Why incremental improvements are no longer enough
There will always be room for better consensus algorithms, smarter virtual machines, improved rollup architectures, and more efficient data availability layers. These are important research and engineering directions. But treating them as the core scaling story misses the elephant in the room.
The current stack is fundamentally constrained by a networking layer never built for decentralized, adversarial, permissionless systems. We are trying to optimize engines while leaving the roads in a 1980s state — narrow, congested, and fragile under stress.
A true TCP/IP‑style moment for web3 means accepting that no amount of patching at the application or execution layer will fully unlock the next era. The foundation itself must evolve.
—
The opportunity: a new layer of the internet stack
If web3 succeeds in building and deploying a decentralized Internet Protocol, several things change:
– Blockchains become infrastructure peers, not isolated silos. They plug into a common grid of bandwidth, similar to how data centers plug into a power grid.
– DeFi and tokenization can safely climb into the trillions. Markets can assume predictable, robust data propagation even under extreme conditions.
– DePIN emerges as a genuine planetary nervous system. Sensors, devices, and physical resources connect into a shared, resilient fabric rather than fragile bespoke overlays.
– Decentralized AI gets a native home. Autonomous agents can transact and coordinate at massive scale without surrendering to centralized gatekeepers.
Just as TCP/IP quietly turned a handful of networks into the internet, a decentralized Internet Protocol can turn a collection of blockchains into a unified, global web3 infrastructure.
—
Web3’s real leap is still ahead
The industry often talks as if web3 has already had its internet moment: “We’re just waiting for the killer app.” In reality, we are much closer to the late 1970s in networking history than the late 1990s. We have powerful ideas, working prototypes, and pockets of scale — but we lack the unifying layer that makes everything click.
The next leap will not be the tenth “fastest L1,” nor yet another incremental rollup. It will be the emergence of a decentralized Internet Protocol that:
– Encodes data instead of merely forwarding it.
– Treats failure and attack as default conditions.
– Lets blockchains, DePIN, and decentralized AI share the same global fabric.
– Preserves the core ethos of web3: trustlessness, censorship resistance, and open participation.
When that layer arrives and is widely adopted, it will be obvious in hindsight that this was web3’s real TCP/IP moment. Everything built on top — from global finance to planetary sensor meshes and agent economies — will finally have a foundation that can carry their weight.