On July 7, 2026, Anthropic expanded Claude Cowork to web, iOS, and Android, enabling AI agents to run background tasks with no device online. Two days later, Fortune confirmed that Anthropic had overtaken OpenAI in annualized enterprise revenue: $47 billion run rate against OpenAI’s most recent $25-33 billion disclosure.
Those two events are not coincidental. They are the same story told from different angles: the enterprise AI race has a leader, and it earned that position by building AI into the operations of companies rather than the habits of consumers.
The Revenue Lead and How It Happened
For most of 2024 and early 2025, OpenAI was the presumed winner of the enterprise AI market. It had more brand recognition, a larger user base, and ChatGPT embedded in the public imagination. Anthropic was the technical competitor, better on safety and benchmarks, but trailing in revenue and adoption.
That changed through 2025 as enterprise AI budgets moved from experimentation to production. Anthropic’s business model, which concentrates roughly 85% of revenue in enterprise and developer API customers, compounded faster than OpenAI’s consumer-subscription base as those budgets scaled.
Claude Code was the single largest accelerator. It reached a $500 million annualized run rate by August 2025 and surpassed $2.5 billion by February 2026, according to internal Anthropic data cited by Sherwood News. What began as a developer tool became the entry point for enterprise adoption: engineering teams that adopted Claude Code for agentic coding then expanded to Claude Enterprise for broader organizational workflows.
The Ramp AI Index published in May 2026 captured the market share shift: 34.4% of businesses were paying for Anthropic services, compared to 32.3% for OpenAI. That 2.1-point difference translated, in revenue terms, to a gap of more than $10 billion in annualized run rate.
Claude Cowork: What the Mobile Expansion Actually Means
Anthropic launched Claude Cowork on web, iOS, and Android on July 7, 2026, starting with Max subscribers at $100 per month. The product had existed since January 2026 as a desktop-only tool, but the July update introduced two capabilities that change its enterprise utility significantly.
First, background execution without a device. Tasks now continue even when the user’s laptop is closed and no device is active at all. A finance analyst who queues a quarterly spend reconciliation before leaving for the day collects finished output the next morning, with the work completed while they slept.
Second, scheduled tasks. Users can specify that an agent should begin work at a set time. Combined with background execution, this enables asynchronous delegation patterns that were previously only possible with dedicated automation infrastructure.
The mobile expansion itself adds a third capability: task handoff across devices. A user can start a research session on their desk, receive a progress notification on their phone during lunch, and review the finished output from the same mobile app. The work follows the user rather than staying tethered to one machine.
Help Net Security reported that Anthropic also added richer admin analytics for Claude Enterprise alongside the Cowork launch: model-level entitlements, spend alerts, and usage visibility by team. Those additions matter as much as the product features for enterprise buyers, who need cost control and audit capability before they can approve broad adoption.
The Usage Data That Changes the Narrative
The most significant data point from the July 7 launch was not the mobile rollout itself. It was what Anthropic released about how organizations are actually using Cowork.
From 1.2 million anonymized sessions across more than 600,000 organizations:
| Usage Category | Share of Cowork Sessions |
|---|---|
| Business process operations | 33.4% |
| Content creation and copywriting | 16.4% |
| Research and analysis | 14.1% |
| Data processing and reporting | 12.8% |
| Coding and software tasks | 9.3% |
| Other | 14.0% |
More than 90% of Cowork usage has nothing to do with coding. The largest category, business process operations, includes work that runs through finance, HR, and administration roles: pulling scattered updates into a single weekly report, building onboarding checklists from employee handbooks, reconciling spreadsheets from multiple systems, converting folders of contracts into renewal trackers with flagged risk dates.
These are not tasks that ChatGPT’s consumer offering targets. They are workflows that live inside enterprise software stacks and require context about the organization, not just general knowledge. The patterns explain why Anthropic’s enterprise revenue has scaled faster: the workflows are stickier, run more frequently, and involve larger organizational commitments than individual consumer subscriptions.
Two Models, Two Markets
The revenue lead reflects a structural divergence that is now measurable.
Anthropic earns roughly 85% of its revenue from enterprise customers and developers who pay for API access. The pricing model is consumption-based: more agent calls, more tasks delegated to Cowork, more Claude Code sessions all translate directly into revenue. As organizations expand usage, revenue scales with them.
OpenAI earns roughly 85% of its revenue from ChatGPT consumer subscriptions: flat monthly fees that do not scale with usage. ChatGPT Work, which OpenAI launched on July 9 to compete directly with Cowork, is a meaningful step toward enterprise consumption revenue, but it launches into a market where Anthropic has been accumulating organizational workflow data and product refinements since January.
For enterprise leaders deciding how to allocate AI vendor budget, this divergence has a practical implication. A vendor whose revenue concentrates in enterprise will invest its product roadmap in enterprise priorities: admin controls, compliance tooling, audit logging, multi-agent orchestration, and workflow integrations. A vendor whose revenue concentrates in consumer subscriptions will optimize for things that individual users notice: chat quality, response speed, and interface design.
The product priorities follow the revenue. Anthropic’s July enterprise additions (admin analytics, model entitlements, spend alerts) confirm the pattern. As the enterprise AI adoption gap has widened, the organizations winning are the ones buying tools from vendors whose financial model depends on enterprise success, not just enterprise interest.
What This Means for Enterprise GTM and Ops Teams
The Cowork usage data points to a shift that enterprise GTM and operations leaders should build into their planning now.
Agentic AI is not spreading through developer workflows and then eventually reaching the rest of the business. It is spreading through business operations directly, in finance, HR, administration, and content, with coding as a minority use case. The teams adopting it most quickly are the ones with high-volume repetitive processes, clear output standards, and enough structured data to give an agent something to act on.
Claude Tag’s integration into enterprise Slack showed the pattern in communication workflows. Cowork’s usage data shows the same pattern at the task level: the AI gets adoption when it handles the work that already exists, not when it requires teams to learn new workflows from scratch.
The practical implication for enterprise teams is sequencing. The highest-value early applications of background agents are not ambitious transformation projects. They are the repetitive, structured, high-frequency tasks that drain attention without requiring judgment: spend reconciliation, contract tracking, report assembly, onboarding documentation. Those tasks have clear success criteria, produce measurable time savings, and build the organizational trust needed before teams delegate higher-stakes work.
For teams ready to map that analysis to their current AI stack and vendor selection, the Enera team works through it directly with enterprise GTM and ops leaders.
Sources: Fortune: Anthropic overtakes OpenAI in revenue (July 2, 2026); TechCrunch: Claude Cowork expands to mobile and web (July 7, 2026); Help Net Security: Claude Cowork turns your phone into a remote control for AI work (July 8, 2026); VentureBeat: Anthropic brings Claude Cowork to mobile and web (July 9, 2026); Sherwood News: Anthropic’s revenue shooting past OpenAI (July 2026); Ramp AI Index via iNews (May 2026).