Consensus is broken.
The AI industry is obsessed with frontier models. GPT-5. Claude 4. Gemini Ultra. The narrative is that the model is the moat. But the market is lying. Databricks just raised $5 billion at a $190 billion valuation, and its revenue run rate is $7 billion, growing 80% year-over-year. That is not a model company. That is a data infrastructure company. And the market is pricing it higher than most model labs.
This is a structural shift. The AI value chain is migrating from the model layer to the data and infrastructure layer. I have seen this pattern before. In 2017, during the Ethereum scalability debate, I spent weeks modeling gas price volatility against transaction throughput. The consensus was that bigger blocks solved scaling. I argued the real bottleneck was computational complexity. The market was wrong then. It is wrong now.
Context: The Databricks Playbook
Databricks is a data platform. It started as a Lakehouse architecture—combining data lakes and warehouses. But the three products they are betting on tell a different story. Unity AI Gateway is a multi-model router with cost control. Lakebase is a serverless Postgres database. Genie is a natural-language interface to enterprise data. None of these are foundation model innovations. They are engineering-level, combinatorial innovations that sit between the models and the data.
What Databricks is selling is not AI. It is the middle layer—the control plane for enterprise AI consumption. In a world where every company will use multiple models (OpenAI, Anthropic, open-source), the real value is in the router that decides which model to call, tracks spending, and enforces data governance. That is Unity AI Gateway. It is the equivalent of a cross-chain router in DeFi—like a decentralized exchange aggregator, but for AI inference.
Core: The Macro Liquidity Map
Let me stress-test this. The $5 billion is not for model training. Databricks has already pivoted away from training its own frontier model (DBRX). The money is for AI infrastructure products, hiring, and acquisitions. This is a capital allocation decision that confirms a macro trend: the bottleneck in AI is not intelligence—it is data context, cost, and governance.
Consider the revenue run rate: $7 billion. At 80% growth, the valuation is 27x revenue. That is at the upper bound of reasonable for AI infrastructure. Compare to Snowflake at 15x. The market is pricing Databricks as a premium asset because it captures the FinOps of AI—the cost control narrative. Every CFO is now auditing AI spending. Databricks sells the cost-saving story.
Lakebase is hitting $100 million in revenue run rate. That is a product-market fit signal. It is a serverless Postgres database running on top of the Lakehouse. This is not just a tool—it is a wedge into the transactional database market. It directly competes with Neon, CockroachDB, and Supabase. But more importantly, it positions Databricks against Snowflake’s transactional ambitions. The database market is consolidating around the data platform.
Contrarian: The Decoupling Thesis
Yields are traps.
The market is still treating AI models as the high-growth asset class. But the decoupling is happening. Model performance improvements are slowing. The marginal gain from GPT-4 to GPT-5 is diminishing. Meanwhile, the cost of inference is dropping. The real value accrues to the infrastructure that lowers the cost of deploying models.
This is exactly what happened in crypto. In 2021, everyone thought NFTs were the future. I audited 50 NFT collections and found only 4% had true interoperability. The narrative was a trap. The real value was in the data layer—like Ethereum’s settlement layer or decentralized storage networks. Today, AI models are the NFTs of 2025. The hype is real, but the moat is an illusion.
Scale kills decentralization. In AI, the same principle applies: the big platforms (Databricks, Snowflake, cloud providers) will absorb the value because they own the data pipe. Frontier models are commoditizing. The model providers are fighting over a shrinking margin. The data infrastructure providers are building the rails.
Takeaway: Cycle Positioning
Where does this leave the crypto-AI narrative? The market is currently obsessed with decentralized compute networks (Render, Akash) and AI agents (Virtuals, Fetch). But the real macro play is in decentralized data infrastructure. Projects like Filecoin, Arweave, and even EigenDA for data availability are the analogues to Databricks. They provide the data layer that AI models need to operate.
Based on my 2022 Terra/Luna analysis, where I reverse-engineered the death spiral against global M2 expansion, I learned that the market often misprices the underlying liquidity driver. The Federal Reserve’s tightening cycle killed Luna. The next cycle will be driven by AI data demand. The winners will be the ones who control the pipeline, not the ones who build the fancy models.
Consensus is broken. Rethink your positioning.