“Spoczywaj w pokoju.” Rest in peace. The phrase comes from a Polish entertainment site announcing the death of actor Bernard Verley — but it fits the mood around vector databases in 2026 better than anyone expected. Because something once untouchable is now being gently lowered into the ground by the industry itself.
Not the technology. The hype. The $18 billion Oracle data center freeze, a 16-year-old walking through Microsoft’s JWT checks, and a Postgres-shaped consolidation wave all point the same direction. The standalone vector database, once crowned the center of the AI stack, is looking less like a king and more like a ceremonial figurehead.
TL;DR: Vector databases were supposed to own the AI stack. Instead, Oracle halted payments on an $18 billion data center project, a teen exposed 17 trillion Microsoft records, and pgvector-style consolidation keeps winning. The hype cycle is cooling, and the boring relational database is collecting the spoils.
What sparked the joke about vector databases resting in peace?
The joke borrows its phrasing from a Polish obituary headline: “Spoczywaj w pokoju, Bernardzie” — Rest in peace, Bernard — published after the death of actor Bernard Verley. Somewhere between that headline and a wave of database consolidation news, the tech community found its metaphor. Vector databases, the darlings of the AI boom, started getting farewell posts.
Why now? Look at the money. Databricks bought serverless database startup Neon for a reported $1 billion (siliconangle.com, May 2025). It then acquired Electric, adding embeddable PostgreSQL to its stack (TechTarget, August 2026). PgDog raised $5.5 million in seed funding to advance PostgreSQL infrastructure. Supabase executives were so impressed with Dreambase that they invested in its $3.7 million round.
Notice a pattern? Every major data-infrastructure deal routes through Postgres, not through a dedicated vector store. The funny part is that the announcement of a death was, like many obituaries, slightly premature. Vector indexes still work. The joke is really about the business model — and about who gets to collect rent on the AI data layer.
Is this a eulogy or a roast? Honestly, it’s both.
Did vector databases actually die, or just the hype around them?
Just the hype — and even that is a gradual cooling, not a sudden collapse. The technology inside vector databases remains genuinely useful: embeddings still need somewhere to live, and similarity search still needs indexes. What’s dying is the idea that this functionality justifies a separate, standalone product with its own billing plan, its own ops burden, and its own sales deck.
The consolidation evidence is hard to ignore. Databricks acquired Neon for a reported $1 billion, then picked up Electric to add embeddable PostgreSQL (TechTarget). EDB, a Postgres company, reached an estimated $161 million in ARR (GetLatka, 2025). PgDog secured $5.5 million in seed funding specifically to advance PostgreSQL infrastructure (Fenado AI, June 2026). Even Supabase’s leadership put personal money into Dreambase’s $3.7 million round (Crunchbase News, April 2026).
Each of those deals strengthens the relational database ecosystem that vector functionality increasingly lives inside. When the general-purpose database absorbs your feature set, your standalone product becomes a rounding error. That’s not death. It’s something quieter: irrelevance as a category.
Can a feature kill a company? When the feature ships for free in a database customers already run — sources suggest it can come very close.
Why is Oracle pulling out of an $18 billion data center project?
Oracle halted payments connected to the construction of the “Jupiter” data center in New Mexico, a project valued at $18 billion — roughly 65.5 billion Polish zloty (Business Insider Polska). The decision triggered a sharp reaction from investors and once again drew attention to the economics of AI infrastructure.
The reported reasons are blunt: not enough energy and not enough money for AI data centers. That’s the headline of the source coverage — a shortage of power and capital hitting a project of enormous scale. Halting payments on a build of this size isn’t a scheduling tweak. It’s a company telling the market that the cost curve of AI infrastructure no longer matches its appetite.
What does this have to do with vector databases? Everything, indirectly. The standalone vector database boom was built on the assumption of unlimited AI infrastructure spending: unlimited GPUs, unlimited data centers, unlimited budgets for anything with “AI” attached. When one of the world’s largest enterprise software companies stops writing checks on an $18 billion facility, the assumption underneath that boom takes the hit too.
What does Oracle’s retreat say about the AI infrastructure bubble?
It says the bubble’s pressure points are physical. Sources point to two shortages: energy and money. You can raise venture rounds and print revenue projections, but you cannot conjure megawatts for a New Mexico data center campus, and you cannot endlessly defer the moment when capex needs a return.
The investor reaction to Oracle’s payment freeze was described as violent — a sharp market response — and coverage notes it refocused attention on the broader problem. If a company with Oracle’s balance sheet hesitates at $18 billion, every smaller player in the AI data stack has to ask harder questions. Including, yes, the vector database startups that assumed the infrastructure wave would carry them forever.
There’s a financial echo in the consolidation data. EDB sits at an estimated $161 million ARR. Neon went for a reported $1 billion. PgDog raised $5.5 million — a modest seed by AI-boom standards. Money hasn’t left the database market entirely. It has moved toward boring, proven, Postgres-shaped infrastructure and away from speculative capacity. Capital is voting for consolidation.
The bubble isn’t popping with a bang. It’s deflating through halted payments and redirected checks.
How did a 16-year-old expose 17 trillion Microsoft database records?
A 16-year-old researcher got access to 17 trillion records from Microsoft databases — because Microsoft wasn’t verifying the JWT signature (sekurak.pl). That’s the core of it. The very mechanism designed to prove a token’s authenticity was skipped, and the gates opened.
The researcher’s path started with a service spotted from the internet that supposedly required login — access was marked as VPN-required. On the site’s page was the logo of a service referred to as “Titan,” which is where the investigation began, according to the sekurak.pl report. From there, the missing signature validation turned a supposedly protected environment into an open door.
Seventeen trillion records is a number that resists intuition. If the databases involved included vector embeddings — and coverage of the incident doesn’t confirm what data types were inside — the point would remain the same: the fanciest data layer in the world is worthless if token validation is optional. Security review doesn’t care about your database category.
Why does this belong in a vector database obituary? Because it punctures the same mystique. The AI data stack was sold as sophisticated. A teenager just demonstrated that some of it was held together with a single unchecked signature.
Is consolidation around Postgres the real killer?
The pattern across 2025 and 2026 tells a consistent story. Databricks acquired Neon for a reported $1 billion (siliconangle.com, May 2025). Months later it bought Electric, adding embeddable PostgreSQL (TechTarget, August 2026). PgDog raised $5.5 million in seed funding for PostgreSQL infrastructure (Fenado AI, June 2026). Supabase executives invested in Dreambase’s $3.7 million round (Crunchbase News, April 2026). EDB reached an estimated $161 million ARR (GetLatka).
Five data points, one direction: Postgres is absorbing the data infrastructure market. And vector search is increasingly a feature of Postgres ecosystems rather than a reason to buy a separate database. The standalone vector database didn’t get murdered. It got outcompeted by a platform that ships the same capability inside a tool teams already operate, back up, and understand.
That’s the real punchline of the “rest in peace” joke. The category isn’t dying of a technical flaw. It’s dying of being a product when it should have stayed a feature.
Can the vector database category recover its independence?
Possibly — but the burden of proof has shifted. For a standalone product to survive, it needs to offer something the consolidated Postgres world cannot: scale, latency, or operational simplicity at a level that justifies a second database in the stack. The funding climate makes that pitch harder than it was. Oracle’s halted $18 billion payments in New Mexico signal that infrastructure buyers now demand discipline, and seed rounds like PgDog’s $5.5 million show capital concentrating on Postgres infrastructure rather than on new standalone stores.
Recovery would also require vendors to stop selling magic and start selling operations. The Microsoft incident is a reminder that buyers are learning to ask unglamorous questions: how are tokens validated, who can reach the data, what happens when a single check is skipped? A vector database that answers those questions convincingly has a path. One that sells embeddings-as-destiny does not.
The honest answer for 2026: the category can survive as a niche, but its days as the center of the AI universe look over. Farewell speeches, it turns out, write themselves.
Why is PostgreSQL quietly winning the database wars?
PostgreSQL has become the default answer for almost every new database workload, and the money proves it. EDB, the largest enterprise PostgreSQL company, reached an estimated $161 million in ARR in 2025 on $64.9 million raised (GetLatka, 2026). PgDog, a startup building infrastructure on top of PostgreSQL, just secured $5.5 million in seed funding (June 2026). Supabase, a PostgreSQL platform, is active enough as a company that its executives became investors in a $3.7 million round for Dreambase.
The pattern is consistent. Startups keep raising money to extend PostgreSQL rather than replace it. Enterprise vendors keep growing revenue on PostgreSQL support. Platform companies keep building whole businesses on a single open-source database engine.
Why does this matter for vector databases? Because PostgreSQL gained vector extension support, and most teams discovered that “good enough” similarity search inside their existing database beats a dedicated system that introduces a second operational burden. The specialized engines promised speed. PostgreSQL promises one less thing to run. For the majority of applications, that trade is not even close.
What do Databricks’ acquisitions of Neon and Electric tell us?
Databricks has been on a PostgreSQL shopping spree, and the spending is not subtle. In May 2025, Databricks bought serverless database startup Neon for a reported $1 billion (SiliconANGLE). Then in August 2026, its acquisition of Electric added embeddable PostgreSQL to the portfolio (TechTarget).
Two acquisitions, one direction. A company whose core business is large-scale analytics — the kind of workloads vector databases claim to serve — decided the future of its transactional layer is PostgreSQL. Not a proprietary engine. Not a niche vector-first store. PostgreSQL, in serverless form with Neon and in embeddable form with Electric.
When the biggest players in AI data infrastructure consolidate around a general-purpose database, the writing is on the wall for standalone specialized stores. Why maintain a separate vector database when the platform you already run treats PostgreSQL as the substrate? The acquisitions suggest Databricks sees vector search as a feature of the database, not a category of product.
Where should developers actually store embeddings in 2026?
The honest answer from the funding data and the acquisition trail: in the database you already have, which for most teams means PostgreSQL. The ecosystem around PostgreSQL has matured to the point where entire companies are built on extending it. PgDog raised $5.5 million specifically to advance PostgreSQL infrastructure. Electric became valuable enough for Databricks to acquire it for its embeddable PostgreSQL. Supabase has grown so comfortably on PostgreSQL that its executives are investing in adjacent tooling like Dreambase’s $3.7 million round.
Here is a simple decision guide:
- Already running PostgreSQL? Store embeddings there with a vector extension.
- Building a serverless architecture? Neon’s model — now part of Databricks — points that way.
- Need an embeddable database? Electric covers the PostgreSQL side of that story.
- Want managed everything? Supabase-style PostgreSQL platforms exist for exactly this.
- Running analytics at scale? Databricks is embedding PostgreSQL into its own stack.
- Considering a dedicated vector database? Ask what it does that your current engine cannot.
- Worried about scale? EDB’s $161 million ARR shows enterprises trust PostgreSQL at serious size.
- Still unsure? The consolidation trend around PostgreSQL is the strongest signal available.
Is the era of specialized AI databases over?
“Over” is too strong. “Diminished” is accurate. The tongue-in-cheek funeral for vector databases reflects a real shift: the infrastructure market is rewarding consolidation, not fragmentation. Databricks spent a reported $1 billion on Neon and followed it with the Electric acquisition — both PostgreSQL plays. Venture money continues flowing to PostgreSQL infrastructure companies like PgDog with its $5.5 million seed round.
Meanwhile, the broader AI infrastructure picture is wobbling at the foundation layer. Oracle halted payments for the $18 billion “Jupiter” data center project in New Mexico, citing — according to coverage — shortages of energy and money for AI data centers, a decision that triggered a sharp investor reaction. When even hyperscale data center plans get paused, asking teams to run yet another specialized database for embeddings feels tone-deaf.
Specialized engines still exist and still serve demanding workloads. But the default trajectory has changed. The general-purpose database won the argument for the vast majority of use cases, and the capital markets agree.
What should teams do with their vector database deployments now?
Nobody is saying rip everything out tomorrow. But a pragmatic checklist is in order, because the market signals are hard to ignore:
- Audit your current vector database spend, including the operational cost of running it.
- Test whether PostgreSQL with a vector extension meets your latency and recall requirements.
- If you are on Databricks, watch how the Neon and Electric integrations evolve — they signal the platform’s roadmap.
- Evaluate managed PostgreSQL platforms like Supabase before signing another specialized vendor contract.
- Consider the maintenance burden: one database to patch, back up, and monitor beats two.
- Factor in infrastructure risk: Oracle’s paused $18 billion data center project shows capacity assumptions can change fast.
- Document your migration path now, before you need it in a hurry.
- Revisit the decision quarterly, because this market is moving quickly.
The lesson of the past two years is that convenience features in mature databases erode standalone products faster than anyone predicts. Plan accordingly.
Frequently Asked Questions
Did vector databases really become obsolete?
The “obituary” is tongue-in-cheek, but the consolidation is real. Databricks acquired Neon for a reported $1 billion and then bought Electric for its embeddable PostgreSQL, both bets on general-purpose PostgreSQL rather than specialized stores. The strong trend toward PostgreSQL as the substrate for AI-era data work means dedicated vector databases face shrinking market space.
How much was Oracle’s halted data center project worth?
Oracle halted payments tied to the construction of the $18 billion (approximately 65.5 billion PLN) “Jupiter” data center in New Mexico. Coverage cited shortages of energy and money for AI data centers as the context. The decision triggered a sharp reaction from investors.
How many records did the Microsoft database incident expose?
A 16-year-old researcher gained access to 17 trillion records from Microsoft databases, according to the security write-up of the incident. The cause was straightforward: Microsoft failed to verify the JWT signature on an authentication endpoint. The researcher located an internet-facing service that appeared to require login, associated with the “Titan” logo, and the missing signature check opened the door.
Why are companies consolidating around PostgreSQL?
Follow the money. EDB reached an estimated $161 million ARR in 2025, PgDog raised $5.5 million in seed funding for PostgreSQL infrastructure, and Databricks spent a reported $1 billion on Neon plus acquired Electric for embeddable PostgreSQL. When capital, acquisitions, and enterprise revenue all converge on one engine, consolidation follows naturally — and features like vector search become attributes of PostgreSQL rather than reasons for separate products.
Summary
The farewell is a joke. The market shift behind it is not.
- Databricks bought Neon for a reported $1 billion and acquired Electric for embeddable PostgreSQL — the biggest AI data platform is betting on a general-purpose engine.
- PostgreSQL companies keep attracting capital: PgDog’s $5.5 million seed round and EDB’s estimated $161 million ARR in 2025 show sustained momentum.
- AI infrastructure itself is shakier than assumed: Oracle halted payments on its $18 billion “Jupiter” data center in New Mexico, spooking investors.
- Security remains the unglamorous baseline: a missing JWT signature check let a 16-year-old reach 17 trillion Microsoft database records.
If you are running a standalone vector database for a workload that fits comfortably in PostgreSQL, this is the moment to benchmark the switch. Your future self, patching one database instead of two, will say thank you.