The Database Decoupled: How MongoDB's Atlas Infinite Tames AI Spikes

NEW YORK, NY – September 29, 2026 — Beneath the frictionless surface of modern digital life—every seamless login, instantaneous financial transaction, and rapid-fire AI response—lies an invisible, laboring infrastructure. For years, the systems governing our data have been forced to over-provision, bracing for the worst-case scenario of traffic spikes by hoarding expensive, idle computing power. Today, at its Investor Day at the Nasdaq MarketSite, database giant MongoDB announced a fundamental dismantling of that legacy model.

With the general availability of MongoDB 9.0 and the public preview of Atlas Infinite, the company is executing its most significant architectural overhaul since the inception of its cloud platform in 2016. By officially decoupling compute from storage and shifting to a pure consumption-based pricing model, the provider is positioning itself to directly manage the erratic, high-throughput demands of the AI era.

But beyond the technical specifications of throughput and query speeds, this shift represents a broader evolution in corporate responsibility and operational efficiency. As artificial intelligence agents increasingly act autonomously on our behalf, the infrastructure that supports them must become as adaptable and resilient as the society it serves.

The Decoupling Shift: Dismantling Legacy Architecture

Historically, operational databases have operated on a coupled architecture: to get more computing power (the engine), you also had to buy more storage (the trunk space), regardless of whether you needed both. This monolithic approach has long been a bottleneck for enterprise scalability, forcing companies to pay for massive, underutilized servers just to ensure they had enough headroom for sudden bursts in user activity.

Atlas Infinite, now available in public preview on AWS, severs this dependency. By separating compute from storage so each can scale independently, the new deployment option cuts the time required to scale by more than 96 percent.

This architectural pivot aligns the company with a growing industry consensus. Cloud-native competitors like Snowflake have long championed decoupled architectures for analytical data warehouses, while Amazon Web Services offers similar elasticity through DynamoDB and Aurora Serverless. However, applying this level of decoupled elasticity to a document database designed for real-time, mission-critical operational workloads presents a unique engineering triumph. It fundamentally changes how organizations build and scale the applications that communities rely on daily.

Taming the AI Agent Traffic Spike

The urgency behind this redesign is not merely theoretical; it is driven by the rapid proliferation of agentic AI. Unlike human users, who sleep, pause, and navigate interfaces linearly, AI agents can flood an enterprise's system with thousands of concurrent reads and writes in a fraction of a second. These unpredictable, spiky traffic patterns are uniquely hostile to rigid database architectures.

"The growing use of AI agents is causing applications to use and store more data," said Ben Cefalo, Chief Product Officer of Core Products at MongoDB. "MongoDB 9.0 hardens the foundation that more than 70,000 customers already trust with extreme performance, making it significantly faster at scale for mission-critical transactional workloads and AI-powered applications. Meanwhile, Atlas Infinite delivers the elasticity, responsiveness, and accuracy that agentic workloads demand, so enterprises never have to stretch legacy architecture to keep up."

Early adopters are already testing the limits of this elasticity. PicPay, a Brazilian fintech company serving millions of users, recently subjected Atlas Infinite to sustained stress tests.

"Atlas Infinite enables us to grow our data footprint with greater operational simplicity with its order-of-magnitude faster backups and decoupled storage architecture," noted Erick Schroder, Database Reliability Engineer at PicPay. "For a platform that has to be ready for extreme traffic spikes, that's exactly the kind of scalability we need to handle more transactions, at greater speed."

According to the company, PicPay sustained four times its normal peak traffic for two hours on the new architecture with zero failures, illustrating the system's capacity to absorb digital shocks without degrading the end-user experience.

FinOps and the Database: The Promise and Peril of Consumption Pricing

The technological decoupling of Atlas Infinite is accompanied by an equally significant financial restructuring. The company is renaming its legacy managed offering to Atlas Core and introducing a new pricing model for Atlas Infinite tied strictly to actual compute and storage consumption.

For enterprise leaders and FinOps practitioners, this is a double-edged sword. On one hand, it promises the end of over-provisioning waste. Internal tests indicate that Atlas Infinite provides 189 percent more throughput per dollar than Atlas Core for I/O-heavy workloads.

"Balancing MongoDB's data model flexibility with true elasticity has helped Okta avoid over-provisioning capacity," said Andrew Yu, VP of Engineering at Okta. "Atlas Infinite will help empower our teams to automatically scale with unpredictable authentication traffic in real time without paying for idle resources, improving efficiency across our identity platforms."

On the other hand, pure consumption models carry the inherent risk of billing unpredictability. While users no longer pay for idle resources, a sustained, unexpected surge in AI agent traffic—perhaps triggered by a viral product launch or a market anomaly—could result in a massive, unbudgeted invoice at the end of the month. As cloud infrastructure becomes more fluid, the burden shifts from upfront capacity planning to real-time financial monitoring, requiring organizations to implement strict guardrails to prevent runaway costs.

Security in the Age of Queryable Encryption

While Atlas Infinite addresses the physical constraints of scaling, the release of MongoDB 9.0 tackles an equally pressing societal concern: data privacy. As applications consume vast oceans of personal information to feed AI models, regulated industries like healthcare and finance face a dual mandate. They must secure sensitive data against relentless cyber threats while maintaining the ability to quickly search and utilize that data for patient care and customer service.

The 9.0 release expands the platform's industry-first Queryable Encryption. Previously limited to exact matches, the system now supports prefix, suffix, and substring searches on encrypted data. This means an organization can query a database for a specific string of sensitive information—such as the last four digits of a social security number or a partial medical record—without ever decrypting the underlying data on the server side.

Cryptographic experts note that performing complex search operations on ciphertext inherently introduces performance overhead compared to querying plaintext. However, the trade-off is often necessary for organizations that must comply with strict data sovereignty and privacy regulations. By keeping data encrypted throughout its lifecycle, the attack surface is drastically reduced, protecting communities from catastrophic data breaches.

To mitigate the performance impact of these heavy workloads, the 9.0 engine includes new Intelligent Workload Management features designed to automatically protect database clusters when they are overloaded, ensuring that short-running, critical operations continue even during unpredictable demand.

"At Coinbase, trading volume can spike in minutes, and our systems have to scale in real time, or customers feel it," explained Krishnan Seshadrinathan, VP of Engineering at Coinbase. "Intelligent Workload Management gives us confidence that our clusters will stay responsive during peak trading windows."

Ultimately, the announcements out of New York today signal a maturation in how the tech industry views its foundational tools. As we integrate autonomous AI deeper into the fabric of our economy, the databases powering these systems can no longer be static vaults. They must become living, breathing entities capable of expanding and contracting in real-time, balancing the relentless demand for speed with the uncompromising need for security and fiscal responsibility.