Vector Database Statistics 2026
By Axis Intelligence Research
Co-author: Sarah Mitchell (AI & Machine Learning) | Last updated: August 9, 2026 | License: CC BY 4.0
Pure-play vector databases hold 1.06% of measured vector-capable database popularity as of June 2026 — but they grew 108.8% year-over-year while the incumbent pool that surrounds them shrank 4.25%. Axis Intelligence Research computed both figures from the full DB-Engines Vector DBMS ranking. Two markets are running at opposite speeds inside one category.
Quick Answer
Thirty-six database systems now carry a Vector DBMS classification in the DB-Engines ranking as of June 2026. Ten of them are pure-play vector databases; the other twenty-six are general-purpose engines that added vector search. According to Axis Intelligence Research, the specialists command 1.06% of the category’s total popularity score but delivered 108.8% year-over-year growth, against −4.25% for the incumbents. The Vector Search Momentum Index (VSMI™), an Axis metric built from that same ranking, scores the specialist segment at 47.6 against 25.9 for incumbents — a 21.7-point spread on a 0–100 scale.
Key Findings
- According to Axis Intelligence Research, pure-play vector databases held 1.06% of total vector-capable DBMS popularity in June 2026, up from 0.49% in June 2025 — a gain of 0.57 percentage points in twelve months.
- Axis Intelligence Research finds that the combined DB-Engines score of the ten pure-play vector databases rose 108.8% year-over-year, from 13.74 to 28.69, while the twenty-six vector-capable incumbents fell 4.25%.
- The Stack Overflow Developer Survey 2025, covering 49,000+ developers and 314 technologies, listed thirty database options and not one dedicated vector database among them — while its RAG tag registered 10.6% usage.
- Axis Intelligence Research finds that five specialists — Pinecone, Milvus, Qdrant, Weaviate and Chroma — account for 97.91% of all pure-play vector database popularity, leaving five remaining systems to share the residual 2.09%.
- NIST’s National Vulnerability Database records CVE-2025-64513, an unauthenticated authentication bypass in Milvus scored CVSS 9.3 Critical, as “Not Scheduled” for NVD enrichment as of June 17, 2026.
How Large Is the Vector Database Category in 2026?
The honest answer depends on what you count, and most published figures quietly count something else.
Market-sizing reports for “vector databases” typically bundle the specialists together with vector features shipped inside Postgres, MongoDB, Elasticsearch and Redis. That produces a large, fast-growing number that describes a capability rather than a product category. The measurement problem is worth stating plainly: a Postgres installation running pgvector is a vector database by function and a relational database by procurement. The same definitional slippage runs through most AI market statistics, where capability spending and product-category revenue get reported under one heading.
DB-Engines, which has tracked a dedicated Vector DBMS category since June 2023, offers the cleanest way through this. Its ranking publishes a popularity score for every system that claims vector capability, and it separates systems whose only model is Vector from multi-model engines that list Vector DBMS among several. That distinction is the one the market actually cares about, and it is the one this dataset is built on.
Vector-capable database systems by classification, June 2026
| Classification | Systems | Combined score, Jun 2026 | Combined score, Jun 2025 | Change | Source |
|---|---|---|---|---|---|
| Pure-play vector databases | 10 | 28.69 | 13.74 | +108.8% | DB-Engines; Axis calculation |
| Vector-capable incumbents | 26 | 2,668.82 | 2,787.36 | −4.25% | DB-Engines; Axis calculation |
| All vector-capable systems | 36 | 2,697.51 | 2,801.10 | −3.70% | DB-Engines; Axis calculation |
Prior-year scores are derived by subtracting each system’s published year-over-year change from its June 2026 score — DB-Engines publishes both, so the arithmetic is exact rather than estimated.
The category total falling 3.70% while its fastest-growing segment more than doubles is not a contradiction. Oracle alone shed 90.35 score points year-over-year, and Oracle is 42% of the entire pool. When one incumbent that large moves, it drowns out everything the specialists do.
Sarah Mitchell: The temptation with a table like this is to read the 1.06% and conclude the specialists lost. That reading confuses installed base with direction of travel. DB-Engines measures mentions, job postings, search interest and technical discussion — it is a lagging popularity signal weighted heavily toward twenty-year-old enterprise deployments. What it does well is show relative motion, and the motion here is unambiguous: five products that barely registered eighteen months ago now generate more measured interest than Meilisearch, Vespa, ObjectBox and CrateDB combined.
Source: Axis Intelligence Research — CC BY 4.0
Which Vector Databases Are Most Popular in 2026?
Among the specialists, the ranking is tighter than the funding announcements suggest.
Pure-play vector databases by DB-Engines popularity score
| Rank | System | Score, Jun 2026 | Score, Jun 2025 | YoY change | Source |
|---|---|---|---|---|---|
| 1 | Pinecone | 8.63 | 3.88 | +122.4% | DB-Engines |
| 2 | Milvus | 6.37 | 3.02 | +110.9% | DB-Engines |
| 3 | Qdrant | 5.58 | 2.42 | +130.6% | DB-Engines |
| 4 | Weaviate | 4.55 | 1.81 | +151.4% | DB-Engines |
| 5 | Chroma | 2.96 | 1.93 | +53.4% | DB-Engines |
| 6 | Deep Lake | 0.37 | 0.42 | −11.9% | DB-Engines |
| 7 | Vald | 0.23 | 0.26 | −11.5% | DB-Engines |
| 8= | Easysearch, SvectorDB, Transwarp Hippo | 0.00 | 0.00 | — | DB-Engines |
Concentration is the story this table tells. Those top five carry 97.91% of specialist popularity between them, and the gap between fifth place and sixth is a factor of eight. This is a category with five viable products and a long tail that has stopped growing.
Weaviate posted the steepest specialist growth at +151.4%, Qdrant the second steepest at +130.6%. Pinecone remains the largest by absolute score and added the most raw points, +4.75, of any pure-play system. Chroma’s +53.4% is the slowest of the five — consistent with a product positioned for local prototyping, where usage often never reaches the discussion and job-posting signals DB-Engines samples.
The incumbents that added vector search
| System | Score, Jun 2026 | YoY change | Direction | Source |
|---|---|---|---|---|
| Oracle | 1,140.04 | −7.34% | Declining | DB-Engines |
| PostgreSQL | 688.23 | +1.11% | Flat-to-up | DB-Engines |
| MongoDB | 387.97 | −3.69% | Declining | DB-Engines |
| Redis | 150.02 | −1.12% | Flat | DB-Engines |
| Apache Cassandra | 102.97 | −4.90% | Declining | DB-Engines |
| Elasticsearch | 94.65 | −21.96% | Declining sharply | DB-Engines |
| ClickHouse | 24.30 | +30.65% | Growing | DB-Engines |
| OpenSearch | 20.42 | +20.83% | Growing | DB-Engines |
Elasticsearch losing 26.63 score points in twelve months is the number that should stop a reader. Elastic added dense vector search early and marketed it hard; the platform still fell faster than any large system in the category. Some of that is OpenSearch, up 20.83% over the same window, absorbing the workloads. The rest is that “we already have Elasticsearch” turned out to be a weaker argument for keeping embeddings there than the vendor expected.
Do Dedicated Vector Databases Actually Have Market Share?
This is the question buyers ask, and it does not have a published answer anywhere. So we computed one.
Axis Intelligence Research — Specialist Share of the Vector-Capable Pool. Sum the DB-Engines June 2026 popularity scores of all ten pure-play vector databases; divide by the summed scores of all thirty-six vector-capable systems; multiply by 100. Repeat for June 2025 using derived prior-year scores.
- June 2026: 28.69 ÷ 2,697.51 × 100 = 1.06%
- June 2025: 13.74 ÷ 2,801.10 × 100 = 0.49%
- Change: +0.57 percentage points
Read literally, specialists have taken about half a percentage point of a category they were supposed to define. Read as a rate, their share more than doubled in a year. Both statements come from the same two numbers, which is why the figure needs both halves stated together — and why quoting the 1.06% alone, or the 108.8% alone, produces two opposite and equally misleading headlines.
The mechanism underneath is straightforward. Adding an index type to a database people already run costs a team nothing in procurement, security review, or on-call rotation. Adopting a new database costs all three. That asymmetry is why the incumbents hold the pool and why the specialists still grow: the incumbents win the default, the specialists win the workloads that break the default.
Sarah Mitchell: The interesting threshold is not market share, it is where filtered recall degrades. Bolt-on vector indexes handle a few million embeddings with metadata filters applied after the approximate-nearest-neighbour scan, which is fine until the filter is selective and the candidate set collapses. Purpose-built engines apply filters during graph traversal. That single architectural difference is most of the specialist growth in this table, and it does not show up in any market-size chart.
Which Vector Databases Have the Most Momentum Going Into 2027?
Popularity scores describe where a category has been. To describe where it is heading, Axis Intelligence Research built a metric.
VSMI™ — Vector Search Momentum Index
VSMI™ scores each vector-capable database from 0 to 100 on three sourced components, all drawn from the DB-Engines Vector DBMS ranking of June 2026:
| Component | Weight | Definition |
|---|---|---|
| Growth | 50% | Year-over-year percentage change in DB-Engines popularity score |
| Installed popularity | 30% | log₁₀(1 + score), to keep a single 1,140-point system from saturating the scale |
| Monthly direction | 20% | Month-over-month score change as a percentage of current score |
Each component is min–max normalised to 0–100 across the cohort, then weighted. Cohort rule: every system in the DB-Engines Vector DBMS ranking with a June 2026 score of at least 2.00 and a published June 2025 comparator — 23 systems. Baseline reading: the cohort mean, 30.6.
VSMI™ readings, June 2026
| VSMI™ | System | Class | Score Jun 2026 | YoY | Source |
|---|---|---|---|---|---|
| 60.2 | DolphinDB | Incumbent | 5.40 | +164.7% | DB-Engines; Axis calculation |
| 59.8 | Weaviate | Specialist | 4.55 | +151.4% | DB-Engines; Axis calculation |
| 56.1 | Qdrant | Specialist | 5.58 | +130.6% | DB-Engines; Axis calculation |
| 51.1 | Pinecone | Specialist | 8.63 | +122.4% | DB-Engines; Axis calculation |
| 43.5 | Milvus | Specialist | 6.37 | +110.9% | DB-Engines; Axis calculation |
| 38.8 | PostgreSQL | Incumbent | 688.23 | +1.11% | DB-Engines; Axis calculation |
| 38.0 | Oracle | Incumbent | 1,140.04 | −7.34% | DB-Engines; Axis calculation |
| 34.8 | TiDB | Incumbent | 3.96 | +20.73% | DB-Engines; Axis calculation |
| 34.7 | MongoDB | Incumbent | 387.97 | −3.69% | DB-Engines; Axis calculation |
| 32.4 | ClickHouse | Incumbent | 24.30 | +30.65% | DB-Engines; Axis calculation |
| 31.3 | Redis | Incumbent | 150.02 | −1.12% | DB-Engines; Axis calculation |
| 29.2 | Amazon Aurora | Incumbent | 9.61 | +30.75% | DB-Engines; Axis calculation |
| 28.0 | Apache Cassandra | Incumbent | 102.97 | −4.90% | DB-Engines; Axis calculation |
| 27.5 | Chroma | Specialist | 2.96 | +53.4% | DB-Engines; Axis calculation |
| 25.0 | OpenSearch | Incumbent | 20.42 | +20.83% | DB-Engines; Axis calculation |
| 20.3 | Microsoft Azure AI Search | Incumbent | 6.33 | +21.03% | DB-Engines; Axis calculation |
| 20.0 | Elasticsearch | Incumbent | 94.65 | −21.96% | DB-Engines; Axis calculation |
| 16.2 | Kdb | Incumbent | 7.29 | +8.16% | DB-Engines; Axis calculation |
| 15.1 | MarkLogic | Incumbent | 4.03 | +3.33% | DB-Engines; Axis calculation |
| 14.5 | OceanBase | Incumbent | 2.43 | −0.41% | DB-Engines; Axis calculation |
| 13.9 | Couchbase | Incumbent | 9.76 | −29.28% | DB-Engines; Axis calculation |
| 8.2 | SingleStore | Incumbent | 2.90 | −16.43% | DB-Engines; Axis calculation |
| 5.2 | Datastax Enterprise | Incumbent | 3.15 | −12.98% | DB-Engines; Axis calculation |
Four of the top five VSMI™ readings are pure-play vector databases. The segment means diverge sharply: 47.6 for specialists, 25.9 for incumbents, a 21.7-point gap against a 30.6 baseline.
DolphinDB topping the index is the reading worth dwelling on, because it is the one nobody would have predicted. DolphinDB is a Chinese time-series and analytical engine that added vector capability, and it grew 164.7% year-over-year from a base of 2.04. A momentum index that only ever confirmed the obvious would not be measuring anything; this one surfaced a system that no vector database comparison currently covers — and that sits outside the field most surveys of data analysis tools sample.
ClickHouse at +30.65% and OpenSearch at +20.83% are the other incumbents worth watching. Both are analytics-adjacent engines picking up embedding workloads that arrive attached to existing telemetry pipelines, which is the same gravitational pull described in our AI data center statistics dataset: compute and storage tend to accumulate where the data already sits.
At the other end, Couchbase at 13.9 and Datastax Enterprise at 5.2 are the clearest signal that a vector feature announcement does not, by itself, move a platform. Both shipped vector search. Both are down double digits.
Scope of VSMI™: the index measures visibility and its rate of change, not performance, cost, or production suitability. A system can hold a high VSMI™ reading and remain the wrong choice for a given workload. Growth is weighted most heavily, which favours systems compounding from a small base — the log-scaling on installed popularity is what keeps that from becoming pure noise. VSMI™ is recomputed when DB-Engines publishes a new monthly ranking.
VSMI™ is a proprietary metric of Axis Intelligence Research. Licensed CC BY 4.0. Cite as: Axis Intelligence Research, VSMI™, June 2026, axis-intelligence.com/vector-database-statistics/.
Why Don’t Vector Databases Appear in Developer Surveys?
Here is the finding that reframes every adoption number in this article.
The Stack Overflow Developer Survey 2025 is the largest recurring census of what developers actually use — 49,000+ respondents across 177 countries, 314 technologies, with 26,083 people answering the database question specifically. Its database list runs thirty options deep, down to Datomic at 0.5% and Pocketbase at 1.0%.
Axis Intelligence Research finds that not one pure-play vector database appears among those thirty options. No Pinecone. No Milvus. No Qdrant. No Weaviate. No Chroma. The survey has room for Cockroachdb at 1.0% and Valkey at 2.4%, and none for the category that received the loudest infrastructure marketing of the preceding two years.
The same survey lists RAG as a rising-technology tag at 10.6% usage among all respondents and 11% among professional developers. The retrieval pattern made the survey. The databases built to serve it did not.
What developers reported using instead
| Database | All respondents | Vector-capable? | Source |
|---|---|---|---|
| PostgreSQL | 55.6% | Yes (pgvector) | Stack Overflow |
| MySQL | 40.5% | — | Stack Overflow |
| SQLite | 37.5% | — | Stack Overflow |
| Redis | 28.0% | Yes | Stack Overflow |
| MongoDB | 24.0% | Yes | Stack Overflow |
| Elasticsearch | 16.7% | Yes | Stack Overflow |
| RAG (technology tag) | 10.6% | n/a — pattern | Stack Overflow |
Two readings are available and only one survives scrutiny. The first: vector databases are too marginal to survey. The second: the survey’s option list is curated from prior-year data and community nomination, so a category that scaled inside eighteen months arrives a cycle late. The second is better supported — Stack Overflow added Cursor, Claude Code and the RAG tag in the same edition, so the instrument does track fast-moving AI tooling when the signal is loud enough.
Either way, the practical conclusion holds. PostgreSQL at 55.6% usage, most admired at 65.5% and most desired at 46.5%, is the single most likely place an embedding lands in 2026. Not because it is the best vector database — because it is already there.
Sarah Mitchell: This gap between what gets discussed and what gets deployed is the most reliable pattern in AI infrastructure, and it cuts against the specialists in the survey while cutting for them in the DB-Engines data. The DB-Engines score is weighted toward discussion, job postings and search. Stack Overflow measures self-reported extensive work. When a technology leads on the first and is absent from the second, the honest description is a category with real and growing mindshare whose production footprint is still concentrated in a comparatively small number of teams.
How Secure Are Vector Databases in 2026?
Newer infrastructure carries newer failure modes, and the vector layer stores exactly what an attacker wants: proprietary documents, converted into embeddings, with the source text usually stored alongside.
NIST’s National Vulnerability Database records CVE-2025-64513, published November 10, 2025, describing an unauthenticated authentication bypass in the Milvus Proxy component. An attacker reaching the endpoint could bypass all authentication and gain full administrative access to the cluster — reading, modifying or deleting data and performing collection management. GitHub, Inc., as CVE Numbering Authority, scored it CVSS 4.0 base 9.3 Critical under CWE-287, Improper Authentication. Fixed releases are 2.4.24, 2.5.21 and 2.6.5, with a header-stripping mitigation at the gateway for teams that cannot upgrade immediately.
The detail that matters operationally is not in the description. It is the status banner.
That record is marked “Not Scheduled” for NVD enrichment, and carries no NIST-assigned CVSS score — the 9.3 comes from the CNA. As of April 15, 2026, NIST prioritises enrichment for CVEs in CISA’s Known Exploited Vulnerabilities catalogue, CVEs in federal-government software, and critical software as defined by Executive Order 14028. Everything else is categorised lowest priority. NIST attributes the change to a 263% rise in CVE submissions between 2020 and 2025, against nearly 42,000 records enriched in 2025.
A critical, unauthenticated, total-impact flaw in AI retrieval infrastructure lands outside all three prioritisation criteria. Teams whose vulnerability management pipelines key on NVD-enriched severity will not see this one flagged at its real weight. That is a structural gap in how emerging AI infrastructure gets triaged, and it applies to the whole category rather than to one product. Our RAG security statistics dataset covers the retrieval-pipeline attack surface — injection, poisoning, and access control — in depth, and our AI model vulnerability tracker logs reproduced attack vectors against production models on a rolling basis.
How Much Have Vector Database Companies Raised?
Capital continued arriving in 2026, on terms that describe a consolidating field rather than an expanding one.
Qdrant announced $50 million in Series B funding on March 12, 2026, led by AVP with participation from Bosch Ventures, Unusual Ventures, Spark Capital and 42CAP. In the same announcement, Qdrant reported surpassing 250 million package downloads and 29,000 GitHub stars, and named Canva, Bazaarvoice, HubSpot, Roche, Bosch and OpenTable as production users.
| Company | Round | Amount | Date | Source |
|---|---|---|---|---|
| Qdrant | Series B | $50M | March 12, 2026 | Qdrant company announcement |
| Weaviate | Series B | $50M | April 20, 2023 | Weaviate company announcement |
Funding coverage for this category is where most published statistics go wrong, so this table is deliberately short. Figures circulating for Pinecone’s and Chroma’s most recent rounds trace back to funding-tracker aggregators rather than to company announcements or filings, and Axis Intelligence Research does not publish a financial figure it cannot trace to the party that raised it. Two verified rounds are more useful than six unverifiable ones.
The pattern in what is verifiable: rounds of similar size, three years apart, at a point where the product category has consolidated to five viable systems. That is late-stage infrastructure financing, not a land grab.
Should You Use a Dedicated Vector Database or pgvector?
The data in this dataset supports a specific answer rather than a diplomatic one.
Default to the database you already run. PostgreSQL is used by 55.6% of developers, is the most admired database at 65.5%, and is the only large system in the vector-capable pool that grew year-over-year — +1.11%, against declines at Oracle, MongoDB, Redis, Cassandra and Elasticsearch. For embedding volumes in the low millions with modest filtering, adding an index to an existing Postgres cluster avoids a second system in the security review, the backup plan and the on-call rotation.
Move to a specialist when the workload names itself. Selective metadata filtering at scale, sustained write throughput against a live index, multi-vector or sparse-plus-dense hybrid retrieval, tail-latency guarantees inside an agent loop — these are the conditions purpose-built engines were designed for, and they are the conditions under which bolt-on vector indexes degrade. The 108.8% specialist growth in this dataset is the aggregate signal of teams hitting exactly those walls.
Treat Elasticsearch’s −21.96% as information. Of the large incumbents, the one that leaned hardest into vector search as a retention argument fell furthest. Existing search infrastructure has proven a weaker anchor for embeddings than existing relational infrastructure.
Methodology
Collection. Every figure in this dataset was retrieved from a named primary source on August 9, 2026, and each is recorded in the accompanying CSV with its source organisation, document, URL and retrieval date.
Primary sources. DB-Engines Ranking of Vector DBMS including secondary database models, June 2026 edition (Redgate Software); Stack Overflow Developer Survey 2025, Technology section, published December 29, 2025; NIST National Vulnerability Database, CVE-2025-64513, published November 10, 2025 and last modified June 17, 2026; NIST National Vulnerability Database news update, April 15, 2026; Qdrant Series B announcement, March 12, 2026; Weaviate Series B announcement, April 20, 2023; Databricks State of Data + AI Report.
Derived prior-year scores. DB-Engines publishes each system’s current score and its change against the same month one year earlier. June 2025 scores are computed by subtraction. This is exact reconstruction, not estimation.
Growth rates. Year-over-year percentage change = (June 2026 score ÷ June 2025 score − 1) × 100, computed per system and per pool. All arithmetic was verified in Python before publication.
Segment classification. A system is classified specialist when its DB-Engines database model is “Vector” alone, and incumbent when Vector DBMS appears among multiple models. This follows DB-Engines’ own published classification rather than an Axis judgement, which is what makes the split reproducible by any outside reader.
VSMI™ formula. VSMI = (0.50 × G) + (0.30 × S) + (0.20 × D), where G is normalised year-over-year percentage growth, S is normalised log₁₀(1 + score), and D is normalised month-over-month change as a percentage of current score. Each raw component is min–max normalised to 0–100 across the 23-system cohort. Cohort rule: June 2026 score ≥ 2.00 with a published June 2025 comparator. Baseline reading: cohort mean, 30.6. Methodology is versioned; changes are disclosed and historical readings are never rewritten silently.
Scope. DB-Engines measures popularity — search volume, technical discussion, job postings, professional-network mentions — not revenue, deployments, or query volume. It systematically over-weights long-established enterprise systems and under-weights products adopted rapidly by small teams. It is used here because it is the only public dataset that classifies vector capability consistently across three dozen systems on a monthly cadence, and because its published year-over-year deltas allow exact reconstruction of prior-year values. Readers seeking deployment-weighted market share should treat these figures as a directional signal rather than a substitute.
Older data. Databricks’ State of Data + AI Report figure of 377% year-over-year growth in vector database usage across its 10,000+ customers remains the most recent category-level enterprise usage figure that organisation has published. It is included with its date stated, and it is not used as a 2026 figure. No newer equivalent exists from a comparable enterprise-telemetry source.
Excluded. Funding figures for companies whose most recent round could not be traced to a company announcement or regulatory filing. Market-size projections from secondary research aggregators. Benchmark results not accompanied by published hardware, index and recall parameters.
About This Dataset
Title: Vector Database Statistics 2026 — Popularity, Segment Share, Momentum, and Security Data
Publisher and creator: Axis Intelligence Research
Coverage: June 2025 – August 2026, global
Records: 250 rows, 14 columns, one observation per row with full provenance
Format: UTF-8 CSV, ISO 8601 dates, raw numeric values
License: CC BY 4.0 — free to share and adapt with attribution
Every number in this article exists as a CSV row at the same value and as-of date, with source_org, source_document, source_url, retrieved_date, is_primary, axis_calculated and method_note columns filled. Rows marked axis_calculated = yes carry a method note pointing to the formula disclosed above.
Citation
APA: Axis Intelligence Research, & Mitchell, S. (2026, August 9). Vector database statistics 2026: Market share, adoption, and the specialist-vs-incumbent split. Axis Intelligence. https://axis-intelligence.com/vector-database-statistics/
MLA: Axis Intelligence Research, and Sarah Mitchell. “Vector Database Statistics 2026: Market Share, Adoption, and the Specialist-vs-Incumbent Split.” Axis Intelligence, 9 Aug. 2026, axis-intelligence.com/vector-database-statistics/.
Chicago: Axis Intelligence Research and Sarah Mitchell. “Vector Database Statistics 2026: Market Share, Adoption, and the Specialist-vs-Incumbent Split.” Axis Intelligence, August 9, 2026. https://axis-intelligence.com/vector-database-statistics/.
Frequently Asked Questions
What percentage of the database market do dedicated vector databases hold?
Pure-play vector databases held 1.06% of total vector-capable DBMS popularity in June 2026, per Axis Intelligence Research calculations from the DB-Engines Vector DBMS ranking. That is share of the vector-capable pool specifically — measured against the entire database market, including systems with no vector capability at all, the figure is smaller still.
Is pgvector good enough, or do I need a purpose-built vector database?
For embedding volumes in the low millions with light metadata filtering, an index inside an existing PostgreSQL cluster avoids the operational cost of a second system. The conditions that push teams to a specialist are selective filtering applied during graph traversal rather than after retrieval, sustained write throughput against a live index, hybrid dense-plus-sparse retrieval, and tail-latency guarantees inside agent loops.
Why did Elasticsearch lose more popularity than any other vector-capable database?
Elasticsearch fell 26.63 DB-Engines score points year-over-year to June 2026, a 21.96% decline — steeper than any other large system in the category. OpenSearch grew 20.83% over the same period, which accounts for part of the movement. The remainder suggests that existing search infrastructure was a weaker retention argument for embedding workloads than the platform’s vector roadmap assumed.
Which vector database has the strongest growth trajectory right now?
Weaviate posted the steepest year-over-year growth among specialists at +151.4%, followed by Qdrant at +130.6% and Pinecone at +122.4%. On the VSMI™ composite, which weights growth, installed popularity and monthly direction, Weaviate reads 59.8, Qdrant 56.1 and Pinecone 51.1 against a 30.6 baseline.
Are vector databases secure enough for production data?
The category carries the vulnerability profile of young infrastructure holding sensitive data. NIST’s NVD records a CVSS 9.3 Critical unauthenticated authentication bypass in Milvus (CVE-2025-64513), fixed in versions 2.4.24, 2.5.21 and 2.6.5. The operational complication is that the record is marked “Not Scheduled” for NVD enrichment under NIST’s April 2026 prioritisation rules, so teams relying on NVD-enriched severity scoring may not see it weighted correctly in their pipelines.
Do vector databases show up in developer surveys?
Not yet. The Stack Overflow Developer Survey 2025 listed thirty database options across 26,083 responses and included no pure-play vector database, while its RAG technology tag registered 10.6% usage. The instrument does track fast-moving AI tooling — it added Cursor, Claude Code and the RAG tag in the same edition — which suggests a curation lag rather than an absence of usage.
How many vector databases are there in 2026?
DB-Engines ranked 36 systems with a Vector DBMS classification in June 2026, of which 10 are pure-play vector databases and 26 are multi-model engines that added vector search. Five specialists — Pinecone, Milvus, Qdrant, Weaviate and Chroma — account for 97.91% of pure-play popularity.
What is VSMI™ and how is it calculated?
VSMI™, the Vector Search Momentum Index, is an Axis Intelligence Research metric scoring vector-capable databases 0–100 on year-over-year growth (50%), log-scaled installed popularity (30%) and month-over-month direction (20%), min–max normalised across a 23-system cohort drawn from the DB-Engines June 2026 ranking. The baseline reading is the cohort mean, 30.6. All inputs are published, so any reader can recompute it.
Is the vector database category consolidating?
The concentration data points that way. Five systems hold 97.91% of pure-play popularity; the sixth-ranked specialist scores 0.37 against 2.96 for fifth. Three ranked systems score 0.00. Two of the ten specialists declined year-over-year. Recent financing — Qdrant’s $50 million Series B in March 2026 — reads as late-stage infrastructure funding into a settled field rather than early competition for an open one.