Turbopuffer
Vector database built on object storage, powering AI search for Cursor, Notion, Anthropic.
On the record
Checkpoint: AWS S3 Vectors moving from preview to general availability, and outcome of Turbopuffer's 2026 price cuts (94% query cut, minimum plan cut to $16)
The call
Turbopuffer grew fast on a real cost edge, but MongoDB, Elastic, and Postgres are folding vector search into tools companies already pay for.
Turbopuffer stores and searches large amounts of data for AI applications (like Cursor and Notion use it) and does this much cheaper than older competitors by storing data differently. The problem is that bigger database companies like MongoDB and Elastic have already added the same search feature to their own products, and Amazon's cloud service could do the same since Turbopuffer actually runs on top of Amazon's storage. The company keeps cutting its prices sharply (down 94% on some features), which suggests it is fighting to keep customers but also losing money faster. One thing that could save Turbopuffer is if it builds such strong relationships with customers that switching becomes too annoying, or if the price cuts actually work and stop once competitors give up, but right now it looks like a price war with no winner in sight.
Why
01
AI pressure: None apparent. Turbopuffer is infrastructure that AI agents and apps consume, not a product AI models replace. Could AI do the core job? Unlikely.
02
Bigger players: MongoDB and Elastic, both shipped native vector search inside their existing databases. Elastic's CEO calls vector search 'a feature, never a business': both rivals now bundle it free into products companies already run. Could a giant copy it? Yes, easily.
03
What protects it. Object-storage architecture cuts costs 70-90% versus RAM-based rivals, plus deep integration with Cursor, Notion, Anthropic workflows.
What would change this call
Whether Turbopuffer's per-query and minimum-plan pricing cuts continue past 2026. Repeated cuts would show margin pressure and category commoditization.
Whether MongoDB or Elastic report vector-search adoption gains in accounts Turbopuffer targets. Would show incumbents absorbing the feature and shrinking Turbopuffer's addressable market.
Whether Cursor, Notion, or Anthropic publicly change or reduce their Turbopuffer usage. Would signal customer concentration risk materializing.
Whether MongoDB or Elastic report vector-search adoption gains in accounts Turbopuffer targets. Would show incumbents absorbing the feature and shrinking Turbopuffer's addressable market.
Whether Cursor, Notion, or Anthropic publicly change or reduce their Turbopuffer usage. Would signal customer concentration risk materializing.
Key numbers
- ~$100M annualized revenue, up from ~$75M
- Sales grew 10x, headcount grew 5x in 2025
- Less than $1M raised before December 2025 seed
Named risks
- Category commoditization as MongoDB, Elastic, and Postgres absorb vector search as a feature.
- Repeated steep price cuts (94% query pricing Feb 2026, minimum plan halved June 2026) signal margin pressure.
- Customer concentration in a small number of large AI-native accounts (Cursor, Notion, Anthropic).
- Very small team (about 22-38 employees) scaling infrastructure at trillions-of-vectors scale.
- Reported user complaints of bill spikes and recall degradation with metadata filters.
Rivals, as read in July 2026
| Rival | Direction | Evidence |
|---|---|---|
| MongoDB | contesting | Shipped native vector search on ~$2.5B revenue base, per Sacra comps, 2026 |
| Elastic | contesting | Added native vector search; CEO called vector databases 'a feature, never a business' in 2026 |
| Pinecone | gaining on | Notion reportedly left Pinecone for Turbopuffer, per 2026-05-20 report |
| pgvector/Postgres | losing to | Turbopuffer's own docs concede pgvector wins under ~10M vectors, 2026 |
| AWS S3 Vectors | contesting | AWS launched first-party native vector service on S3, generally available 2025-12-02 |
Questions people ask
- Is Turbopuffer profitable in 2026?
- No profitability figures are disclosed. Turbopuffer reportedly hit roughly $100M in annualized revenue by March 2026, up from about $75M at the end of 2025, per a Sacra estimate published 2026-05-21, but no net income or margin data has been made public.
- Could AI replace Turbopuffer?
- No. Turbopuffer is infrastructure that AI systems consume, not a model AI could substitute. It stores and indexes vectors for RAG and agent retrieval, and even Anthropic uses it as a customer under a bring-your-own-compute deployment as of 2026.
- Who is Turbopuffer's biggest competitor?
- Pinecone is the closest direct rival on architecture, while MongoDB (about $2.5B revenue) and Elastic (about $1.7B revenue) have both shipped native vector search that threatens to commoditize the category, and pgvector already beats Turbopuffer under 10 million vectors.
- Who owns Turbopuffer?
- Turbopuffer is privately held, founded in 2023 by ex-Shopify engineers Simon Hørup Eskildsen and Justin Li. It ran on less than $1M in angel funding from Lachy Groom until closing a seed round with Thrive Capital in December 2025.
- How much does Turbopuffer charge and is pricing under pressure?
- Turbopuffer cut query pricing by up to 94% in February 2026 and lowered its minimum monthly plan from $64 to $16 by June 2026, per its own roadmap changelog, a pattern consistent with active price competition in the vector database market.
- Does Turbopuffer have big-name customers?
- Yes. Cursor was its first major customer, signed November 2023, and reportedly cut infrastructure costs about 95%. Notion migrated workloads in mid-2024 and Notion's own security docs list Turbopuffer as a sub-processor as of May 2026.
Timeline
- 2026-06Launch plan minimum lowered from $64/month to $16/month
- 2026-03Sacra estimates annualized revenue reached about $100M, up from $75M
- 2026-02Query pricing cut by up to 94%
- 2026-01MCP Server beta and FTS v2 (20x faster full-text search) launched
- 2025-12Closed seed round with Thrive Capital; Lachy Groom follows on; ~22 employees reported
- 2025-07Cursor pricing controversy highlighted usage-based revenue exposure for infra vendors
- 2024-06Notion migrated search workloads to Turbopuffer
- 2024Sub-$1M angel investment from Lachy Groom with prove-PMF-or-shut-down terms
- 2023-11Cursor signed as first major customer
- 2023-10Turbopuffer launched as a vector-only search database
Revision history
31 July 2026
First assessed. Losing its protection. The text above is fixed as of this date and is not edited afterwards.
In the same band
Sources
- Ex-Shopify engineers raise fresh financing to scale Turbopuffer's AI search (2025-12-19)
- Turbopuffer revenue, funding & news
- $100M/year PostHog of vector databases (2026-05-21)
- Ex-Shopify engineers raise fresh financing to scale Turbopuffer's AI search (2025-12-19)
- Medium (2026-04-12)
- Enterprise Search Security & Privacy Practices (2026-05-28)
- fast search engine built on object storage
- Management and Employees List
- Roadmap & Changelog (2026-07-16)
- What Is Pinecone? (2026-05-20)
- Do You Still Need a Vector Database in 2026? (2026-04-27)
- Cursor apologizes for unclear pricing changes that upset users (2025-07-07)