The ideal database for AI agents doesn't exist yet, says Percona CEO

Percona CEO Peter Farkas said current databases are not ideally suited to emerging AI and agentic workloads, which can involve roughly 150 iterative attempts before selecting an output. He expects organizations to rely on existing database technologies until AI use cases and infrastructure requirements mature enough to support a purpose-built solution. Percona is broadening beyond its MySQL heritage into a database-agnostic open-source support business, while Farkas also highlighted unresolved litigation against his co-founded FerretDB by MongoDB.

Analysis

The near-term equity implication is limited: AI experimentation expands database read/write intensity, but it does not yet establish a durable winner at the database layer. MDB's valuation depends on Atlas consumption re-acceleration and sustained enterprise standardization; agentic workloads could raise usage, but iterative inference architectures also incentivize customers to separate transient state, retrieval, and system-of-record data across cheaper open-source and cloud-native components. That fragmentation is more favorable to AMZN, MSFT, GOOGL and ORCL, which can bundle storage, compute, model access and managed databases into a single cloud commit.

The more important 6-18 month risk to MDB is not a newly listed "AI database" competitor, but Postgres-led commoditization. If enterprises standardize on Postgres-compatible operational stores plus vector/search extensions, proprietary document-database differentiation narrows and sales cycles become price-sensitive; that would constrain MDB's long-run gross-margin and multiple premium even if absolute workloads grow. Conversely, evidence that agent applications require flexible-schema, globally distributed operational data with low-latency change streams would support Atlas net revenue retention and reverse this concern.

The litigation backdrop is a modest sentiment overhang rather than a material MDB fundamental catalyst unless it produces enforceable restrictions on MongoDB-wire-compatible alternatives or meaningful damages. The contrarian point is that the market may be too quick to equate higher AI data volumes with MDB upside: the highest-value AI data plane economics are likely to accrue initially to hyperscaler compute/storage bundles, while database vendors bear the integration and support burden. No directional MDB trade is warranted solely from this signal; the next earnings print's Atlas consumption, large-deal commentary and net revenue retention are the relevant validation points.

Key Decisions for Investors

• Maintain neutral MDB over the next 1-3 months; do not chase an AI-database narrative absent evidence of accelerating Atlas consumption and stable/improving net revenue retention. A quarterly consumption-guide reduction or renewed compression in large-enterprise deal activity would support a tactical short.
• Express the infrastructure-bundling thesis as a 6-12 month pair: long AMZN or MSFT versus short MDB in equal dollar beta-adjusted size. The thesis fails if MDB demonstrates that AI-related workloads are lifting Atlas growth faster than hyperscaler database growth, or if Atlas retention re-expands materially.
• Set an earnings watch item for MDB: management disclosure of AI workloads contributing measurable Atlas consumption, expanding workloads per customer, or improved operating leverage would invalidate the commoditization concern. In the absence of those disclosures, treat AI references as marketing rather than a revenue catalyst.
• Monitor legal milestones involving MongoDB-compatible translation layers for optional upside to MDB sentiment, but avoid underwriting damages or injunction value before court rulings. A decisive adverse ruling against MDB's claims would remove even this small protective catalyst.