When Models Get Cheaper, Who’s Making Money?

Databricks announced a $3 billion financing round, valuing the company at $188 billion with a 35× price‑to‑sales (PS) multiple, far above Snowflake’s 18× and Microsoft Azure’s ~12×. The author interprets this as evidence that data‑centric platforms command higher valuations than pure model providers.

Key insight: As model costs fall, the sole profitable layer becomes data readiness . Anthropic’s internal experiment showed that raw data yielded only 21% accuracy, while adding a full data‑engineering stack (metadata, quality monitoring, access control) lifted accuracy to 95% – a 74‑point gain.

Data readiness is defined by three capabilities: discoverability, trustworthy quality, and controlled access. Without them, even the most powerful models act like blindfolded agents.

The article outlines four profit‑generating roles:

Companies that prepare enterprise data for AI (e.g., Databricks, ClickHouse) are earning strong revenue growth (Databricks AI revenue grew from $14 billion to $17 billion in four months).

Providers of the interface between models and data (e.g., Databricks Unity AI Gateway, Palantir AIP) charge for security, audit, and data‑masking services.

Businesses that embed AI into core operations by turning private data into model‑usable context (e.g., Databricks Genie, Snowflake Semantic Layer).

European compliance infrastructure vendors, driven by GDPR and the AI Act, are creating a new market for sovereign data‑governance tools.

Because models are now commoditized, the choice of model becomes a procurement decision rather than a technical differentiator. OpenRouter data shows Chinese models capture ~30% of traffic at a fraction of the cost of GPT‑5.5 (e.g., $0.09 / M tokens vs. $5 / M tokens), further emphasizing the shift toward data‑layer competition.

Cloud providers (AWS, Azure, GCP) pose a strategic risk: they bundle compute, storage, and model services, potentially capturing the data‑readiness value chain. Databricks counters this by promoting Unity Catalog as a cross‑cloud, neutral data‑governance layer.

In Europe, higher compliance costs (≈20% more than in the US) and data‑sovereignty requirements limit the adoption of US‑based SaaS data platforms, fostering a nascent market for local compliance middleware.

Overall, the profit pool is moving from model revenue to the “data‑to‑model” pipeline: security/control rents and business‑logic rents. Whoever controls data flow into models—through governance, routing, or semantic layers—captures the bulk of AI‑related profits.