The numbers are seductive. A $65 billion ARR—or was it $650 billion? The exact figure from the SemiAnalysis report on Anthropic barely matters because the magnitude itself is a lie. It’s a lie that every blockchain project with a liquidity mining program knows intimately: the difference between subsidized volume and organic value. In the silence of the bear, we heard the truth—that channel revenue is the new liquidity mining, and it’s making us all feel richer while we dilute our own sustainability.
Over the past seven days, I’ve been dissecting the channel model of Anthropic, the AI company that has become a case study for how not to measure success. More than 40% of its ARR comes from indirect channels—AWS Bedrock, Microsoft Foundry, Google Cloud. The cloud giants take a cut, and Anthropic pays for compute on top. The result? Every dollar earned through a channel is worth less than half of a direct sale. This is not a story about AI. It’s a story about trust, about the architecture of value, and about the blockchain industry’s own obsession with vanity metrics.
Context: The Protocol That Is Not a Protocol
Anthropic is not a blockchain project, but its business model is a mirror. It sells a model—Claude—through a layer of intermediaries who control the distribution, the pricing, and the customer relationship. Sound familiar? That’s exactly how most Layer 2 rollups operate: they depend on Ethereum for security, but they also depend on centralized sequencers for data availability and on aggregators for user acquisition. The channel is the verdict. My code was the covenant, not just the contract—but when the channel controls the covenant, the code becomes a commodity.
In blockchain, we call it “TVL farming.” In AI, they call it “channel revenue.” Both are ways to inflate top-line numbers while sacrificing bottom-line health. The SemiAnalysis report estimates that Anthropic’s channel revenue has a gross margin of 30–50%, compared to 70–80% for direct sales. That’s a 30%+ penalty for using a distribution layer. In DeFi, we see the same: a protocol that incentivizes LPs with 100% APY often sees 80% of that liquidity vanish when incentives stop. The channel is the incentive, and the channel is the leak.
Core: The Anatomy of a Dilution Engine
Let me walk through the mechanics. Anthropic pays the cloud provider a commission (typically 15–30%) plus the cost of GPU compute. The cloud provider then resells Claude to enterprise customers, often bundling it with other services. The customer pays a single bill, but the value chain is opaque. The same happens in blockchain: a rollup pays a data availability layer (like Celestia or EigenDA) for posting blobs, and the rollup’s sequencer charges users a fee. But the sequencer is often a single entity, and the channel—the user’s wallet, the frontend, the aggregator—takes another cut.
Based on my audit experience of over 20 DeFi protocols, I’ve seen this pattern repeat. The worst case was a lending protocol that reported $1.2 billion in TVL, but 90% of it came from a single whale using a leveraged loop through a third-party aggregator. When the aggregator changed its fee structure, the TVL dropped to $200 million in a week. The protocol’s “growth” was a channel illusion. Anthropic’s 40% channel dependency is the same dragon. Every broken token taught me how to hold value—and the lesson is that value is not in the volume but in the direct connection.
But let’s go deeper. The channel model creates a structural misalignment of incentives. The channel provider (cloud, frontend, aggregator) wants to maximize transaction volume, not user retention. They have no stake in the long-term health of the product. In blockchain, this manifests as MEV extraction, frontrunning, and spam transactions. In AI, it manifests as cache poisoning, model abuse, and hidden costs. The channel is a black box that separates the creator from the user.
Consider the data: Anthropic’s direct sales team is small, estimated at less than 50 people, while its channel partners employ thousands of enterprise sales reps. The cloud providers have the relationships, but they also have competing products. AWS sells Bedrock, but it also invests in its own AI models. Google Cloud offers Claude, but it also pushes Gemini. The channel is a Trojan horse—it brings growth, but it also brings competition. The same is true for blockchain: a rollup that uses a centralized sequencer is dependent on that sequencer’s incentives. If the sequencer decides to frontrun or censor, the rollup is powerless.
Contrarian: The Strategic Necessity of Channels
Now, the contrarian angle. I’m not saying channels are evil. In fact, for a startup, channels are often the only way to reach enterprise customers. Anthropic’s channel strategy allowed it to achieve a $3.6 billion run rate (the corrected number, not the fantasy $65 billion) within two years. Without AWS and Google, it would have taken five years. The same applies to blockchain: without centralized exchanges, DeFi protocols would have no liquidity. Without aggregators, retail users would have no access.
The key is to recognize the channel as a temporary scaffold, not a permanent foundation. The most successful blockchain projects—think Uniswap, MakerDAO—built their own direct channels over time. Uniswap’s frontend is a direct channel, and its liquidity is decentralized. MakerDAO has its own Oasis app. They used centralized channels early (like Coinbase listings) but then built self-sovereign distribution. The mistake is to treat the channel as the business model, rather than the on-ramp. In the silence of the bear, we heard the truth—that the channel is the first step, not the last.
Anthropic is now trying to build a direct sales force, but it’s expensive and slow. The same is true for blockchain: a protocol that relies on 100% of its revenue from a single aggregator is a protocol that will die. The contrarian insight is that the channel is not the problem—the lack of a channel exit strategy is. Every project should have a plan to reduce channel dependency by 10% per quarter. If you can’t, you’re not building a protocol; you’re building a feature for the channel.
Takeaway: The Vision Forward
The $65 billion ARR number is a mirage, but the mirage itself is data. It tells us that the market is desperate for a story of growth, even if the story is built on sand. In blockchain, we face the same desperation: protocols that inflate TVL with incentive programs, rollups that claim “decentralization” while using a single sequencer, DAOs that measure success by treasury size rather than user adoption. The channel is the new liquidity mining, and it’s making us all feel richer while we dilute our own sustainability.
My code was the covenant, not just the contract. The covenant is the direct relationship between creator and user. The channel is a contract, and contracts can be broken. The question is not whether to use channels—the question is whether you are building a cathedral or a tent. A tent is quick to set up but easily blown away. A cathedral takes decades, but it stands. The next bull market will not reward the biggest ARR; it will reward the most resilient value chains. And resilience starts with the courage to see through the channel illusion.
Every broken token taught me how to hold value. Now, I’m watching Anthropic’s broken ARR teach the same lesson to the AI world. The blockchain industry should listen—not because AI is our competitor, but because the channel is our common enemy.