OpenAI acquires Ona to put Codex agents inside customer-controlled clouds

OpenAI agreed on 2026-06-11 to acquire Ona, the cloud-sandbox company that operates as Gitpod GmbH, in a deal that hands Codex agents a path to run inside an organization’s own cloud rather than on OpenAI infrastructure. Terms were not disclosed and closing is gated on regulatory review. The acquisition is the clearest signal so far that OpenAI is treating long-running agent execution as an enterprise procurement problem first, and a model-capability problem second.

What shipped

Ona’s platform runs developer workloads inside persistent cloud sandboxes that stay live after the developer’s local workstation goes offline, so agents working over hours or days do not lose state when a laptop closes. The company reports it has helped roughly two million developers move work into reproducible cloud environments and has a shared customer list with OpenAI. Post-close, Ona’s team is set to join OpenAI and embed inside the Codex group, and the two companies will operate independently until the transaction completes.

The strategic framing OpenAI offered is execution model, not headcount. According to the announcement, Ona’s customer-controlled architecture is meant to let Codex agents run inside a buyer’s existing cloud account while OpenAI continues to supply the intelligence and orchestration layer above. That preserves the enterprise security boundary that has been the friction point for buyers running pilots that need data-processing addenda for every model. The backdrop is Codex usage: OpenAI says weekly active users now exceed five million, up roughly 400% from earlier this year, and that the most valuable agent work has shifted from minutes-long sessions to runs that unfold over hours or days.

Where this lands in the market

The competitive frame is Anthropic’s Claude Code, which has been picking up enterprise share through a combination of long-horizon coding agents and Anthropic Managed Agents on Modal Sandboxes and on Cloudflare. OpenAI was the last of the frontier labs without a first-party answer to that sandboxing layer, and the Ona deal closes the gap by bringing the runtime in-house rather than partnering for it. The contrast with Anthropic is the architectural choice: Anthropic ships agents into partner sandboxes, while OpenAI now appears to bet that owning the sandbox provider matters when the agent needs persistent state, IAM-scoped access to internal systems, and audit trails buyers can review themselves.

It also fits a 2026 M&A pattern. Promptfoo went to OpenAI in March, the healthcare startup Torch closed in January for around $60 million, and the Sky-for-Mac team at Software Applications was acquired in October 2025. The through-line is OpenAI buying the surfaces it expects to ship Codex through: evaluation tooling, vertical workflow teams, desktop interfaces, and now execution infrastructure. The two-million-developer footprint Ona brings is the largest distribution wedge of the bunch and overlaps directly with the audience Codex needs to reach to outpace Claude Code on adoption.

The deal mechanics are the part to watch. OpenAI did not disclose price, and the regulatory-review gate is not a formality: Gitpod has European headquarters, European user data, and a developer-tools market presence that puts the deal within the EU’s interest in cloud-developer concentration. Closing pacing will indicate how OpenAI plans to handle that scrutiny while Anthropic’s confidential S-1 advances.

What’s worth watching

1. Pricing for customer-cloud Codex. OpenAI has not said how a Codex deployment running inside a buyer’s cloud will be billed against the same per-token model that runs on OpenAI infrastructure. If customer-cloud carries an enterprise uplift, it confirms OpenAI is using Ona to widen the enterprise SKU, not just to remove a friction point.
2. Modal and Cloudflare’s response. Modal closed a $355M Series C in May on the strength of its sandbox-for-agents narrative, and Cloudflare shipped Anthropic Managed Agents on Workers in the same window. With OpenAI buying a competing runtime, both incumbents have a reason to tighten exclusivity with Anthropic or push toward neutral multi-model sandbox positioning. Either path reshapes the inference-and-runtime procurement story.
3. Ona under OpenAI branding. The Gitpod brand has carried two decades of developer goodwill in the remote-dev category. Whether OpenAI keeps Ona as a sub-product, folds it entirely into Codex, or runs the pre-acquisition product line in maintenance mode is the signal for how seriously OpenAI plans to court the non-Codex sandbox audience.

The next checkpoint is the closing announcement and the first Codex release that ships the customer-cloud option as a first-class deployment target. The acquisition itself is straightforward; the interesting question is whether OpenAI ships the new runtime as a Codex feature or as a separately priced enterprise tier.

Sources cited

1. OpenAI: OpenAI to acquire Ona openai.com
2. Bloomberg: OpenAI to Acquire Cloud Platform Ona to Support AI Agents www.bloomberg.com
3. CNBC: OpenAI to acquire Ona to support its AI coding assistant, Codex www.cnbc.com
4. SiliconANGLE: OpenAI acquires AI agent orchestration startup Ona siliconangle.com