OpenAI Frontier and the Agents SDK: Enterprise AI Crosses 40% of Revenue

A Revenue Mix That Looked Different a Year Ago

When OpenAI crossed $25 billion in annualized revenue in February 2026 — up from $20 billion at the end of 2025 — the striking datapoint was not the top-line growth. It was the mix. Enterprise had crossed 40% of total revenue, up from roughly 30% a year earlier, and was on track to reach parity with the consumer business by the end of 2026. Paying business users hit 9 million in February, up from 5 million in August 2025.

For a company that built its brand on a free chat product, this is a structural shift. It means the next phase of OpenAI’s growth depends less on consumer subscription expansion and more on enterprise platform adoption — the same ground Microsoft, Oracle, Salesforce, and ServiceNow have defended for decades.

The Frontier Platform

Frontier, officially launched in February 2026 and expanded through Q1 and Q2, is OpenAI’s enterprise agent platform. The core claim is simple: rather than embedding AI agents inside a single product or workflow, Frontier provides a shared intelligence layer that governs all of a company’s agents across systems of record, SaaS tools, and internal applications.

Structurally, Frontier gives enterprises three things they previously had to cobble together:

• Shared business context. Connectors into CRMs, data warehouses, ticketing tools, and internal apps so every agent starts with the same operational memory.
• Agent execution. A controlled runtime where agents can reason over data, execute code, call tools, and handle long-horizon tasks without silently breaking.
• Identity and governance. Permission boundaries, auditability, and access controls that match enterprise security expectations.

Named early customers on the OpenAI Frontier launch included Uber, State Farm, Intuit, and Thermo Fisher Scientific. HP and Oracle have been referenced in adjacent announcements, and Capgemini publicly joined the Frontier alliance as a systems-integrator partner.

Critically, Frontier is open to agents from competitors. OpenAI has stated that the platform is compatible with agents built on Google’s, Microsoft’s, and Anthropic’s stacks — a move that mirrors the multi-vendor reality enterprises actually operate in, and one that preempts the vendor-lock-in critique that would otherwise block adoption.

The April 16 Agents SDK Expansion

On April 16, 2026, OpenAI released a significant expansion of its Agents SDK. The update added:

• Sandboxing: controlled execution environments where agents can safely call tools, run code, and read files without risking the broader system.
• Advanced tooling for enterprise automation, including an in-distribution harness for deployment and testing of production agent workloads.
• Long-horizon task support — the SDK-level analogue of Anthropic’s positioning for Claude Opus 4.7, reflecting the industry-wide shift toward multi-hour, multi-step agent workflows.

The SDK ships first in Python, with TypeScript support on the roadmap — a language ordering that acknowledges the data-science community as the primary author of first-generation agents but signals that front-end and full-stack developers are next.

In parallel, OpenAI’s numbers on the consumer-agent side reinforce the platform thesis. Codex — OpenAI’s coding agent — reached 3 million weekly active users in Q1 2026, up from effectively zero at the start of the quarter. The API now processes more than 15 billion tokens per minute. These are ecosystem numbers, not product numbers.

The Unified AI Superapp Thesis

Beyond Frontier and the SDK, OpenAI’s April messaging introduced a broader thesis: the unified AI superapp. The claim is that enterprise employees should have “one place” to work with AI agents throughout the day, bringing together ChatGPT, Codex, agentic browsing, and verticalized agents into a single surface.

For enterprise buyers, this raises a question that matters for 2027 procurement: is the future of productivity a collection of vertical SaaS tools with embedded AI (the Microsoft Copilot, Salesforce Agentforce, Google Workspace Studio model), or a single AI-native workspace that agents orbit (the OpenAI thesis)? No vendor has answered this conclusively yet, and the answer will likely be heterogeneous — different enterprises will pick different patterns depending on existing stack and risk tolerance.

What Competitors Are Doing

The context matters. In the same April 2026 window:

• Anthropic shipped Claude Opus 4.7 with a long-horizon agent pitch and leading SWE-bench scores.
• Google announced Workspace Studio, the A2A protocol, and a cluster of new agent products at Google Cloud Next 2026.
• Microsoft continued expanding Copilot Studio and embedding agents deeper into the Microsoft 365 suite.

Every major provider is now shipping an enterprise-agent platform story. OpenAI’s differentiator is the density of its developer community, the scale of Codex usage, and the fact that its enterprise platform is explicitly designed to host competitor agents rather than lock them out. Its vulnerability is the thinner integration with existing productivity suites compared to Microsoft or Google, which is why the unified-superapp thesis matters: it is an attempt to reframe the productivity contest on OpenAI’s turf.

What Enterprise Buyers Should Do

For a 2026-2027 AI platform decision, three practical steps apply.

First, run a real agent workload — not a demo — on Frontier alongside the incumbent SaaS-embedded alternatives. Pick a workflow that has real cost and real failure modes (customer service triage, sales ops, engineering incident response) and measure completion rate, time-to-completion, and per-workflow cost.

Second, plan for multi-vendor agent orchestration. If Frontier can host Anthropic and Google agents, and Google’s A2A protocol can bridge across providers, the single-vendor decision is weaker than it was a year ago. Architect for portability.

Third, watch the commercial terms of the enterprise platform contracts. At 40% of revenue and growing, OpenAI has strong pricing leverage — but the market has at least three credible challengers, which creates genuine competitive pricing for buyers with scale.

What Enterprise Buyers Should Build Into Their 2026-2027 Agent Platform Decision

Enterprise AI platform decisions made in 2026 will shape the productivity architecture for the next three to five years. The proliferation of competing platforms — Frontier, Azure AI Foundry, Google Workspace Studio, AWS Bedrock with Strands — makes the evaluation process more complex than it was when Microsoft Copilot was the only credible enterprise AI platform. Gartner’s April 2026 AI Platform Evaluation Framework identifies three dimensions where enterprise buyers consistently underinvest during initial platform selection, only to pay remediation costs later.

1. Run a Production-Grade Evaluation on One Real Workflow Before Any Contract

The most common buyer mistake in enterprise AI platform selection is treating a vendor proof-of-concept as equivalent to a production evaluation. Vendor POCs are curated for happy-path scenarios; production workloads involve incomplete data, error cascades, rate limits, and security boundary violations that only appear under real conditions. The minimum viable evaluation for a 2026-2027 enterprise agent platform decision should run one workflow with genuine business stakes — customer-support escalation triage, engineering incident summarization, or sales-ops lead qualification — in a controlled production environment for 30 days, with instrumented logging of completion rate, exception rate, cost per step, and time-to-resolution. OpenAI’s Frontier early customers (Uber, State Farm, Intuit, Thermo Fisher) have all publicly referenced structured evaluation processes before commitment. Enterprises that skip this step and proceed directly from demo to deployment are implicitly accepting the vendor’s word for performance claims that cost five to ten times more to remediate post-contract than they would have cost to verify pre-contract.

2. Define Governance Requirements Before Evaluating Platform Capabilities

Enterprise AI governance is not a post-selection checklist — it is a pre-selection filter. The three governance requirements that most frequently create platform incompatibilities after selection are: data-residency constraints (regulated industries in healthcare, banking, and defense require specific geographic bounds on where agent memory and tool-call data are processed and stored), auditability requirements (legal and compliance teams need session-level logs of what agents did, in a format that satisfies internal review cycles and external regulator examination), and identity and access integration (agents must operate within existing IAM policies, not create a parallel permissions model). OpenAI Frontier’s governance documentation and the April 16 Agents SDK sandboxing update address some of these requirements, but the specifics of data residency and audit log retention vary by deployment configuration. Enterprises that define their governance non-negotiables in writing — in a one-page requirements document, not a full RFP — before beginning platform evaluation surface incompatibilities in week one rather than in month six of a deployment.

3. Negotiate Commercial Terms Against at Least Three Credible Alternatives

OpenAI’s enterprise revenue growing to 40% of total, combined with 9 million business users, represents genuine pricing power. But the enterprise AI platform market in mid-2026 has three credible challengers to Frontier: Microsoft Azure AI Foundry with deep M365 integration, Google Workspace Studio with Gemini and A2A protocol, and AWS Bedrock with the Strands agent framework and existing infrastructure agreements. Each of these platforms has negotiated enterprise contracts with reference customers at scale. Buyers who enter Frontier contract negotiations without a documented competing offer from at least one alternative are accepting list pricing in a market where 15-25% discounts are available for multi-year commitments with minimum usage guarantees. The competitive pricing dynamic is temporary — as one platform achieves dominant adoption, pricing leverage will consolidate — which makes 2026 the optimal negotiating window for enterprise buyers with genuine scale and the credibility to walk to an alternative.

Where This Fits in 2026’s Enterprise AI Ecosystem

OpenAI’s April 2026 enterprise moves — the Frontier platform, the Agents SDK expansion, and the 40 percent revenue milestone — are best understood as the opening of a competitive period that will define enterprise AI infrastructure for the next five years. The platform market Frontier is entering already has three credible incumbents: Microsoft Azure AI Foundry with deep M365 integration, Google Workspace Studio backed by the A2A interoperability protocol, and AWS Bedrock with the Strands agent framework and existing infrastructure relationships across most of the Fortune 500. Each of these platforms has been in market for longer and has denser integration with the systems enterprises already run. OpenAI’s differentiation is developer density — Codex at 3 million weekly active users — and the explicit design choice to host competitor agents rather than lock them out. That multi-vendor openness is strategically significant: it is an attempt to win on ecosystem logic rather than exclusivity.

What will determine whether Frontier becomes the productivity architecture of the 2030s or a feature absorbed by a larger platform provider is whether enterprise buyers in 2026 and 2027 treat their platform decisions as multi-year commitments or as modular infrastructure choices. The Gartner AI Platform Evaluation Framework’s finding that buyers consistently underinvest in governance and auditability before selection — and pay remediation costs afterward — suggests that Frontier’s April 16 sandboxing and governance updates are solving a real enterprise problem, not adding marketing features. The companies that win the enterprise AI platform market will be the ones whose governance controls satisfy the risk officers, not just the engineering teams. For OpenAI, crossing 40 percent of revenue from enterprise is the milestone that makes Frontier a strategic priority; the next milestone worth watching is whether Uber, State Farm, and Intuit become the reference for a deployment architecture that others replicate.

Frequently Asked Questions

Sources & Further Reading

• The next phase of enterprise AI — OpenAI
• Introducing OpenAI Frontier — OpenAI
• OpenAI launches a way for enterprises to build and manage AI agents — TechCrunch
• OpenAI Expands Agents SDK with Sandbox and Advanced Tooling for Enterprise AI Automation — The AI Insider
• OpenAI Says Enterprise AI Is Already 40% of Its Revenue — Yahoo Finance