I spent the first two weeks of January in a fluorescence-lit room in Chengdu, staring at a fork of the MakerDAO governance dashboard. We were debating whether to adjust the stability fee by 0.25%—a decision that would affect 12,000 collateral holders. The debate lasted 72 hours, produced 43 on-chain votes, and left me exhausted but certain: decentralized governance is slow, messy, and painfully human. But it is honest.
Then, on January 15, I read the news: Databricks had closed a $5 billion funding round at a $190 billion valuation. The round was led by Morgan Stanley, with participation from MGX, the Abu Dhabi sovereign wealth fund. The company’s revenue run rate had crossed $7 billion, growing at 80% year-over-year. The CEO, Ali Ghodsi, declared that “AGI has already arrived, using the pre-2022 definition.”
My first reaction was not awe—it was a chill. Not because of the numbers, but because of the narrative. Databricks is not a model company. It is a data-platform company that has repositioned itself as the “AI governance layer” for enterprises. It sells control over model routing, token costs, and data context. It is, in essence, building a centralized equivalent of what we in crypto call a “governance treasury with multisig and spending limits.”
And the market is paying 27 times revenue for it.
As a DAO Governance Architect who has spent the last six years designing the very same mechanisms—routing, cost control, permissions—I felt a strange kinship. Databricks’ Unity AI Gateway is a multisig wallet for AI tokens. Its Lakebase is a unified ledger for transactional data, much like a blockchain’s state machine. Its Genie product is a natural-language interface to query that ledger, akin to a blockchain explorer with a RAG layer.
But there is a fundamental difference. Databricks is a centralized, permissioned system. Its governance is opaque, decided by a board and a CEO. Its users are tenants, not sovereigns. The $5 billion it raised is a bet that enterprises will pay for that centralized control—that they prefer a trusted third party to manage their AI spending and data access.
In crypto, we call that a “fiat bridge.” And we know that bridges eventually leak.
Let me trace the three products through the lens of blockchain governance.
Unity AI Gateway: The Multisig for Model Spend
Databricks describes Unity AI Gateway as “cross-model routing and spend control.” It allows enterprises to route requests to different LLM providers—OpenAI, Anthropic, Mistral, or their own fine-tuned models—and set budgets per team, project, or user. It integrates with Unity Catalog, the company’s data governance layer, so that routing decisions respect data lineage and access policies.
This is exactly what a DAO treasury does with a multisig. You have a set of signers (the governance committee), a set of spending limits (the budget), and a routing logic (which token to use for which transaction). In a DAO, the multisig is a smart contract, audited and immutable. In Databricks, it is a proprietary service, controlled by a single entity.
The question is not whether the technology works—it does. The question is whether the governance model is sustainable. When a DAO’s multisig fails, the community can fork. When Databricks’ gateway fails—say, a routing bug that sends sensitive data to a third-party model—who is accountable? The company’s SLA? The enterprise’s legal team? The answer is not written in code, but in a contract. And contracts can be renegotiated, litigated, or broken.
Lakebase: The Unified Ledger with Postgres Compatibility
Lakebase is a serverless Postgres database that runs on Databricks’ Lakehouse. It has already reached $100 million in annualized revenue. The technical implication is clear: Databricks is moving from analytical workloads (OLAP) to transactional workloads (OLTP). It wants to be the single source of truth for both analytics and operations.
In blockchain terms, Lakebase is a state machine that supports the Postgres protocol. It is like a blockchain that speaks SQL instead of Solidity. The ambition is to absorb the existing Postgres ecosystem—millions of databases, thousands of applications—into Databricks’ walled garden.
But here is the hidden cost: by moving to Lakebase, enterprises lose the ability to self-host, to audit the code, to fork the data. They become tenants on a platform that can change its pricing, its terms, or its features at any time. This is the same trade-off that enterprises made when they moved from on-premise to cloud—and we all know how that ended. The cloud oligopoly now extracts 30-40% margins on storage and compute.
In crypto, we call this “rent-seeking.” And we have built a whole industry around minimizing it through open protocols, permissionless access, and programmable money.
Genie: The Natural-Language Interface to the State Machine
Genie allows users to ask questions in plain English and get answers from the enterprise data warehouse. It is a combination of text-to-SQL, semantic layer, and RAG. The goal is to democratize data access—to let non-technical employees query the lakehouse without writing SQL.
This is reminiscent of the “universal blockchain explorer” dream—a chat interface that can query any on-chain data. But in practice, blockchain explorers like Etherscan and Dune are already doing this, with the added benefit of transparency. Every query on Dune is public, reproducible, and auditable. In Databricks, every query is private, governed by enterprise policies, and invisible to the outside world.
Transparency is not just a feature; it is a governance primitive. When a DAO’s treasury is visible to all members, trust is built into the system. When a Databricks query is hidden behind a corporate firewall, trust is delegated to the administrator. That works until the administrator makes a mistake—or worse, a decision that benefits the company at the expense of the user.
The Contrarian Angle: Centralized AI Infrastructure as a “Governance Trap”
I have seen this pattern before. In 2020, I wrote a dissenting essay on MakerDAO’s risk parameters, arguing that algorithmic neutrality often masks systemic bias. The whales won, and the smallholders lost. I learned that governance is not about the code; it is about who writes the code and who can change it.
Databricks is writing the code for enterprise AI governance. It is building the tools that will decide which models run, how much they cost, and what data they can access. It is doing this with a centralized team, a board of directors, and a profit motive. That is not evil—it is business. But it is a governance trap, because once enterprises are locked into Databricks’ stack, their ability to switch to a decentralized alternative becomes prohibitively expensive.
We saw this in the 2010s with cloud computing. We saw it in the 2020s with social media platforms. Now we are seeing it with AI infrastructure. The same pattern: a centralized platform offers convenience and control, grows to dominance, and then extracts rents. The only difference is that this time, the platform is not just managing data—it is managing the intelligence that interprets that data. That is a power that should not be concentrated.
The Crypto Lens: What Decentralized AI Infrastructure Could Look Like
If Databricks is the centralized walled garden, what is the decentralized alternative? I believe it is a combination of three things:
- A permissionless model routing layer—like a blockchain-based marketplace where any model can be listed, and any user can query any model, with fees and governance handled by a DAO. Projects like Bittensor and Render are moving in this direction.
- A sovereign data layer—where data is stored on a decentralized storage network (Arweave, Filecoin) and governed by user-owned access controls. This is the promise of “data DAOs.”
- A transparent query interface—where every query is logged on-chain, and the results are verifiable. This is the vision of “on-chain AI.”
These three components together form a decentralized AI infrastructure stack that is open, auditable, and resistant to rent-seeking. It is not as fast or as cheap as a centralized platform—yet. But it is more resilient, more equitable, and more aligned with the values of self-sovereignty.
The Takeaway: A Fork in the Road
The $5 billion that Databricks raised is a signal that the market believes in centralized AI governance. It is a bet that enterprises will pay a premium for convenience, control, and compliance. And they will, for a while.
But the history of technology is a history of centralization and decentralization cycles. The mainframe gave way to the PC. The PC gave way to the cloud. The cloud is now yielding to the blockchain. AI infrastructure will follow the same arc.
Databricks is the mainframe of AI. Crypto is the blockchain. The question is not which one is better today—it is which one will be better in ten years.
I know which one I am building for.
