How AI Killed "Database Startups"

It feels like the database startup world, which saw tons of new options pop up recently, is actually shrinking now. Big players like PlanetScale, Fauna, and Neon have faced significant challenges – PlanetScale killed their free tier, Fauna shut down, and most recently, Neon was acquired by Databricks. This acquisition is a big deal, raising questions about the future of database innovation and what these market shifts mean if you're trying to pick a database.

Here's a breakdown of what's going on:

• The database market for startups feels rough right now, despite a boom in new companies trying to innovate.
• Specific examples of this difficulty include PlanetScale removing their free tier to focus on enterprise and profitability, Fauna shutting down entirely, and Neon being acquired by Databricks.
• Neon was known as a "serverless Postgres" provider, aiming to rethink how Postgres works for modern applications, especially in serverless environments.
• Postgres traditionally struggles with serverless environments because it's connection-limited. Each concurrent request typically requires a separate connection, and managing lots of ephemeral connections from serverless functions (like AWS Lambda, Vercel, Cloudflare) is difficult and hits database limits quickly.
• Companies like Supabase, which also rely on Postgres, have had to build their own connection pooling solutions, but they still face limitations.
• Other companies, like PlanetScale, tackled this by offering an HTTP layer for database interaction instead of direct connections, which works very well for serverless use cases. PlanetScale was found to be a phenomenal solution for this reason.
• Neon also aimed to solve the serverless connection problem but focused heavily on improving the developer experience (DX).
• A key DX feature Neon championed was preview environments and database branching. This allows developers to create temporary copies of their database schema for testing new features in a pull request, similar to how preview environments work for frontend code. This feature is incredibly useful once you have it.
• Traditional Postgres makes branching expensive because each database instance is tightly coupled and costly to spin up frequently.
• Neon attempted to make individual database deployments much cheaper and enable easy branching. They did this by rewriting Postgres from scratch in Rust, aiming for Postgres compatibility but with a serverless, cheap-to-run architecture. They also offered an open-source version and their own cloud hosting.
• Neon's cloud offering was historically quite cheap, with generous free and low-cost tiers ($0 and $19/month).
• Despite their technical efforts, Neon had significant reliability issues, including a lot of downtime in their early years, although this has reportedly improved. They haven't had the embarrassing data loss or exposure incidents seen with some competitors like Terso.
• A major concern with Neon from the presenter's perspective was their rapid and seemingly excessive hiring. They had 130 employees relatively early on, with less traffic than companies like PlanetScale which had significantly fewer employees (~40).
• High employee costs, estimated at over $11 million per year just for salaries based on a $90k average, made profitability incredibly difficult for Neon, especially with a large number of users on free or low-paying tiers.
• PlanetScale, facing similar market pressures, chose a different path: eliminating their free tier to become profitable and ensure long-term viability, even if it meant initially upsetting some users. This strategic pivot, influenced by CEO Sam Lambert's background in venture capital, allowed them to thrive and invest in high-performance features like "Metal."
• Contrast Neon's costs and revenue with companies like Terso (SQLite-based). Terso's per-database cost is "hilariously cheap" (estimated 5 cents/month) due to storing data in cold storage like S3, but their revenue is likely low as few people pay, and their reliability has been poor, with instances of data loss and exposure.
• Another contrast is PlanetScale. While their minimum cost per database is higher ($20/month), every user is paying something, and they can charge much more for enterprise features ($500+/month), giving them better margins and long-term upside.
• The presenter believes it's almost impossible to build a successful business targeting developers with only a $10/month price point; you need higher tiers or a different model.
• Neon had raised a massive amount of funding ($129.6 million). This large amount of money in the bank was the only way they could sustain their high costs, but it also puts immense pressure to show growth and a path to a higher valuation for future fundraising rounds or an exit.
• Given their high costs, likely low revenue from paid users, and dwindling money in the bank, Neon was in a difficult position to raise more funding, especially at a higher valuation. Venture capitalists look for strong growth and profitability potential, which Neon seemed to lack relative to their spending.
• The presenter suspects Neon's CEO sought more funding, highlighting their growth in databases created by AI agents (a recent trend), but potential investors likely saw the high costs, low paying customer conversion, and difficult path to profitability/high valuation, leading to a down round or inability to raise.
• An acquisition becomes the most viable option when a company can't raise more funds without significant dilution or a down round, and profitability isn't increasing fast enough to cover costs.
• Databricks is a large, enterprise-focused data analytics and AI platform company with lots of money and enterprise customers but lacks a traditional database offering and struggles to get into the small-to-medium business space.
• Neon, conversely, had significant growth in the small-to-medium and developer space, especially with the rise of AI, but had no money and no enterprise presence.
• The acquisition is seen as a "match made in heaven" in a sense: Databricks gains a developer-focused database product and a pathway into the early-stage market and AI development space, while Neon gets the stability, funding, and potential enterprise reach it desperately needed to survive.
• The Databricks announcement explicitly mentions Neon's value for "developers and AI agents," noting that AI agents create four times more databases than humans on Neon.
• The presenter observes a trend where companies that built tools focused on junior developers or simple workflows (like Neon's branching/cheap DBs, Convex's infrastructure-as-code via text files, Cloudflare Workers' cheap compute for inference) accidentally created products that are exceptionally well-suited for AI development and agents because AI is good with text, simple interfaces, and often just waits for results.
• These companies, by building niche or developer-friendly tools, have become significantly more valuable due to the AI boom, even if their original market wasn't massive. Neon is one such example, but their cost structure was unsustainable without external funding.
• Databricks, recognizing that its traditional enterprise analytics platform isn't built for this new wave of small, fast-moving AI teams and projects (even within large companies), acquired Neon to gain access to this market and the specific benefits Neon's architecture provides for AI agents.
• This shift towards large companies having internal "small startups" using modern, fast tooling is becoming more common.
• While the acquisition is positive for Neon's survival and its users' data stability, it also potentially signifies the end of the recent wave of independent database startups trying to fundamentally rethink data interaction. The market seems to be consolidating or shifting focus dramatically towards AI use cases.
• The state of the database market is complex, and understanding these dynamics is crucial when choosing a database platform for your project.