When China Software International (CSI) announced it had signed a token revenue sharing agreement with Moonshot AI—developer of the Kimi large language model—the market reacted with a surge of optimism. The partnership, dubbed the "Moon Landing Project," aims to integrate Moonshot AI’s K2.7 Code and K3 models into CSI’s AllMeta platform, targeting enterprise clients in energy, power, and finance. But here’s what caught my attention: the revenue model is based on token consumption, not fixed licensing fees. In a world where blockchain advocates argue for programmable money and transparent incentive alignment, this deal feels like a crypto-native concept wrapped in a traditional IT services contract. Yet, as I dug deeper, I found a story that is less about decentralized innovation and more about a pragmatic pivot to recurring revenue—with all the familiar risks of centralized control.
Context
CSI is a publicly traded IT services giant (stock code 00354.HK) with deep ties to state-owned enterprises in critical infrastructure sectors. Moonshot AI, backed by Alibaba and other VCs, rose to prominence with its Kimi chatbot and long-context capabilities. The core of the deal is straightforward: CSI will act as a system integrator, embedding Moonshot AI’s models into its AllMeta platform for agentic AI applications. Instead of paying upfront for model access, CSI will share a portion of the token consumption revenue generated by enterprise usage. "Token" here refers to the unit of model computation—each prompt or response consumes tokens, akin to how gas fees work on Ethereum. This mirrors the Web3 concept of paying per action, but with a crucial difference: the tokens are not on-chain, and the revenue split is governed by a traditional contract.
Core
From a blockchain architecture perspective, this is fascinating—and troubling. The token revenue sharing model aligns incentives between an AI model provider and a system integrator, much like how a DeFi protocol distributes trading fees to liquidity providers. Both parties have a direct stake in maximizing usage. CSI’s sales team is now motivated to push deep AI integration into enterprise workflows because their compensation scales with token burn. Moonshot AI, in turn, gets a predictable revenue stream without bearing the cost of an enterprise sales force.
But here’s where the crypto lens reveals flaws. In a decentralized protocol, token distribution is governed by smart contracts—transparent, immutable, and auditable by anyone. The CSI-Moonshot agreement, by contrast, is a black box. I’ve spent years auditing smart contracts and DAO treasuries, and I can tell you: without on-chain settlement, the trust model relies entirely on legal enforcement and bilateral audits. Enterprises in energy and finance are already skeptical of sharing sensitive data with AI models; now they must trust that their token consumption is accurately tracked and fairly billed. The AllMeta platform becomes a centralized oracle—a single point of failure that can manipulate usage metrics. This is the same vulnerability I flagged in 2017 when I audited a DAO governance contract with a reentrancy flaw: the system appears closed-loop, but the loop has a hidden key.
Furthermore, the token itself is not a digital asset. It cannot be traded, staked, or used outside of the CSI-Moonshot ecosystem. This is a far cry from the vision of decentralized AI marketplaces like Bittensor or Render Network, where token holders govern resource allocation. Moonshot AI’s K3 model may be cutting-edge, but its integration into enterprise workflows will require extensive fine-tuning on proprietary data. That fine-tuning is done on central servers, with no accountability to the community. The "token" is just a billing unit—a clever contract term, not a cryptographic primitive.
Contrarian
Is this really a step toward decentralized AI, or is it crypto-washing disguised as innovation? Consider this: Moonshot AI and CSI are building a walled garden. The AllMeta platform will likely lock in customers by integrating deeply with their ERP and CRM systems. Switching to a different model provider would require re-engineering the entire agent workflow—a cost most enterprises will not bear. The token revenue model, while innovative, actually strengthens vendor lock-in. It creates a financial incentive for CSI to maximize token consumption, not necessarily to deliver optimal outcomes. This is the classic agency problem that DAOs try to solve with transparent treasury management. Here, the “DAO” is a bilateral contract between two companies, and the “treasury” is their shared revenue pool, managed by a private ledger.
From my experience during the 2022 bear market, I watched protocols collapse because they prioritized revenue sharing over governance resilience. When the market turned, those with transparent on-chain revenue streams survived; those with opaque partnership agreements—like the one CSI just signed—faced liquidity crises when partners reneged. Moonshot AI may have a stellar model today, but if K3 loses its competitive edge, CSI has every incentive to renegotiate the split or switch to another provider. The token revenue sharing creates a fragile equilibrium, not a trust-minimized foundation.
Takeaway
The CSI-Moonshot deal is a signal that large enterprises are experimenting with crypto-native business models, even if they avoid the baggage of actual cryptocurrencies. But as of 2026, token revenue sharing remains a centralized instrument—a contract, not a protocol. For the blockchain community, this should be a call to action: we need on-chain AI agent frameworks that offer transparent, programmable revenue distribution without relying on trusted intermediaries. The technology exists—ERC-20 tokens, smart contracts, and decentralized oracles—but adoption lags because enterprises fear losing control. We must show them that decentralization can be more reliable, not less. Proof is binary; meaning is fluid. In a world of ledgers, who holds the memory? The answer should not be a single company’s database.