I have spent the past year watching Gulf states pour billions into sovereign AI initiatives, only to discover that sovereignty stops at the data center door. The compute layer, the networking fabric, the specialized silicon, all of it traces back to a handful of American providers. Databricks reaching $188 billion in valuation while marketing itself as the enterprise AI platform should make investors ask a simple question: what happens when the infrastructure beneath these platforms becomes contested geopolitical terrain?
The company's pivot from big data to AI looks smooth in pitch decks, but the technical reality tells a different story. Databricks runs on cloud infrastructure it does not control. Its Mosaic acquisition gave it model training capabilities, but training at scale still means negotiating for H100 clusters that Nvidia allocates according to priorities we do not fully understand. The company champions open-weight models as a cost advantage, which is true until you calculate the inference costs at enterprise scale. I have reviewed procurement documents from African financial institutions trying to deploy these solutions. The infrastructure bills do not decrease, they just get reclassified.
The market treats infrastructure as a solved commodity problem. It is not. I watched a West African government spend $200 million building a data center for AI workloads, only to discover that without direct relationships with hyperscalers, their effective compute costs remained 40% higher than competitors in Europe. Databricks' valuation assumes frictionless access to compute, storage, and networking at predictable costs. That assumption works until it does not. Energy availability, cooling capacity, and fiber connectivity matter more than elegant software architecture when you are trying to run production AI workloads in markets beyond the US and EU.
The comparison everyone makes is to Snowflake's trajectory, but Snowflake operates in a mature infrastructure environment where data warehouse workloads are predictable. AI inference patterns are not predictable. A single viral application can 10x your compute requirements in hours. Databricks is building a business on infrastructure it leases, in a market where demand volatility is the only constant. The company raised $19 billion in 18 months because the capital markets believe AI infrastructure is abundant. I have seen the queue times for GPU clusters. Abundance is not the word I would use.
Sovereign AI projects in the Gulf teach a clear lesson: you cannot build strategic autonomy on rented infrastructure. The same logic applies to enterprise AI platforms. Databricks may have excellent engineering and a compelling product vision, but its valuation assumes infrastructure remains cheap, available, and politically neutral. None of those assumptions hold under serious scrutiny. The market is pricing Databricks as if it owns the stack. It does not. It is a sophisticated tenant, and tenants face risks that landlords do not.