The AI Cloud Nobody’s Talking About Is Quietly Beating AWS at Its Own Game

A 1,500-person company is outmaneuvering cloud giants with quarterly platform updates — and the numbers back it up.

Let’s say you want to manage your AI infrastructure by talking to it. Type “create a shared file system with 50GB capacity,” hit enter, review the plan, approve it. Done. No digging through nested menus. No YAML. No wading through six layers of AWS documentation to figure out which service you actually need.

That’s what Nebius shipped this week.

Everyone assumes the future of cloud belongs to Amazon, Google, and Microsoft. They have the talent, the capital, the brand recognition. Why would a company that didn’t exist in its current form three years ago threaten that?

Here’s the thing. Speed is now the moat. And the three biggest clouds are cruise ships.

The Release That Matters (And Why Timing Is Everything)

On June 24, Nebius unveiled AI Cloud 3.6 — codenamed Aether — its fourth major platform update in roughly a year. The headline feature is Nebius Echo, an AI agent embedded directly into the cloud console that lets developers control infrastructure through natural language. You type. It plans. You approve. It executes.

That approval step isn’t a footnote — it’s the design philosophy. Echo runs with guardrails that prevent unintended actions, and it surfaces exactly what it’s about to do before doing it. The agent runs on open-source models hosted on Nebius’s own Token Factory inference platform, which means Nebius eats its own cooking: the same production inference stack customers use for their own workloads is the one powering Echo.

Beyond Echo, 3.6 includes a new Key Management Service with customer-managed encryption keys, cryptographic erasure, Workload Identity Federation, local SSDs on GPU servers to eliminate I/O bottlenecks during training, and intelligent object storage tiering that automatically archives cold data. The update also introduced a 100x improvement in IOPS for metadata-heavy workloads and validated shared filesystem clusters that can scale to 100 petabytes.

One more thing buried in the release: Nebius Certifications. Starting with an AI Cloud Ops (Associate) track, they’re building a credentialing ecosystem for AI infrastructure engineers — the same playbook AWS used to build lock-in loyalty in the mid-2010s.

This isn’t a minor patch. This is a company shipping enterprise-grade infrastructure updates at startup speed.

The Speedboat Problem (And Why It Isn’t Really a Technology Problem)

AWS has around 200,000 employees. Google Cloud has roughly 50,000. Azure, similar. Nebius has 1,500.

That gap is usually framed as a weakness for Nebius. It isn’t.

Think about what it takes to ship a new feature at AWS. An engineer has an idea. It goes to a team lead. The team lead escalates to a VP. The VP needs to loop in product and legal. Someone asks, “Does this cannibalize our existing service?” Someone else asks, “What does this do to our existing contracts?” Three weeks later, it’s in a roadmap. Nine months later, it ships.

Nebius doesn’t face that problem. When the company has 1,500 people total, decisions reach the people who make them in hours. There’s no internal politics about protecting existing revenue streams because the entire company is pointed at one direction: capture market share before the window closes.

The result is a cadence AWS can’t match in this specific domain. Nebius went from 3.5 to 3.6 in roughly four months. Each release adds material new capabilities. The quarterly update rhythm is intentional — and it signals something important about how the company operates internally.

This is the structural advantage that gets underestimated. Technical talent matters. Capital matters. But the rate at which an organization can identify a gap, make a decision, and ship the fix is a function of structure, not headcount.

The Numbers Are No Longer Small

Two years ago, Nebius reported roughly $90 million in annual recurring revenue. For 2026, the company is guiding for a run rate between $7 billion and $9 billion. That’s not a rounding error. That’s a fundamental change in what kind of company this is.

The deals driving that growth are real: a five-year agreement with Microsoft worth $17.4 billion for dedicated GPU infrastructure capacity. A $3 billion deal with Meta that was subsequently expanded to $12 billion. A $2 billion strategic investment from Nvidia, which also gives Nebius preferred positioning on next-generation GPU architectures including Rubin and Blackwell Ultra. And in June, Nebius closed its acquisition of Eigen AI — an inference optimization company — for around $643 million.

The company also recently launched a Physical AI Living Lab in the UK, partnered with Komodor for autonomous site reliability engineering, and is reportedly scoping data center expansions into Oklahoma and Canada.

A 355% year-over-year revenue jump. A data center infrastructure business up 400%. These are the numbers of a company that has found genuine product-market fit, not a speculative bet on a trend.

What Echo Actually Signals

Here’s the question worth sitting with: Why does it matter that an AI agent can provision cloud resources through natural language?

Because the cloud interface is still, in 2026, mostly a console full of forms. You navigate to the right service. You fill in the right fields. You configure networking, permissions, storage in separate flows. You make a mistake, you go back. ML engineers and data scientists — the people Nebius is specifically trying to serve — lose enormous amounts of time to this before the actual work begins.

Echo is a bet that the interface layer of cloud infrastructure is about to change the same way the software layer changed when developers stopped writing assembly code. You describe what you want. The system figures out the how.

What’s notable is where Nebius is positioning this. Echo isn’t marketed as a cost-cutting feature or an automation tool. It’s described as “a step toward a cloud built for agentic AI, where the platform itself helps evaluate options, and provision and configure resources as workloads demand them.” The roadmap includes infrastructure debugging and multi-step Infrastructure-as-Code deployments.

That’s not a feature. That’s a platform thesis: the next-generation cloud will be one where the cloud itself participates in building on top of itself.

AWS has thousands of engineers who could build something similar. The question isn’t capability. It’s whether the org can prioritize it, align around it, and ship it fast enough to matter. AWS launched its own AI infrastructure assistant, Amazon Q, but it operates at a completely different scope — it’s a developer productivity tool, not an agent for infrastructure control embedded in the core console.

The gap isn’t in talent. It’s in velocity.

The Real Bet

Shopify, Revolut, Cursor, Black Forest Labs, Photoroom — these are the kinds of companies trusting Nebius with production AI workloads right now. The customer base looks like the early AWS roster: fast-moving companies that needed something purpose-built for what they were actually doing, not adapted from general-purpose infrastructure.

The Nebius thesis isn’t complicated. AI compute demand is going up. The companies that win the most of that demand will be the ones that make AI development faster, simpler, and cheaper for the teams doing the work. The three hyperscalers are optimizing for billions of existing users across thousands of use cases. Nebius is optimizing for one: the AI developer who needs to train, deploy, and scale models without fighting the platform.

That focus, combined with an org structure that can ship every quarter, is what Echo represents. Not a chat interface bolted onto a cloud console. A signal that the company is building toward something specific — and building it faster than the people with 100 times as many employees.

The infrastructure wars are getting interesting.

If you follow the AI infrastructure space, the Nebius Certifications program is worth watching. First-mover credentialing ecosystems tend to create durable platform loyalty — it’s how you turn power users into advocates. The AI Cloud Ops (Associate) track is live now.