Enterprise AI in 2026: How OpenAI, Google, Microsoft and Anthropic Compare

Within the same few weeks of late September 2026, the four largest providers of corporate AI tools each announced major updates to their business products. OpenAI held its DevDay conference on September 29, Anthropic opened a marketplace for Claude add-ons on September 23, Microsoft reported its latest Copilot numbers at its fiscal fourth-quarter earnings call on July 31, and Google had already replaced its Vertex AI platform with a new Gemini Enterprise system in April. The overlap reflects how quickly enterprise AI vendors are moving to lock in corporate customers before competitors do.

OpenAI pushes always-on agents and faster decision-making

At DevDay, OpenAI launched "Dots," background AI agents built on its GPT-6 Astra model that run on their own cloud computers rather than inside a single chat window. Each Dot connects to more than 4,000 third-party apps and to Slack and Microsoft Teams, keeps working on assigned tasks without further prompting, and lets users set rules for which actions it can take on its own versus which require approval. Dots launched the same day for ChatGPT Pro and Business Premium subscribers, with one included per plan, and enterprise customers can enable them with administrator approval.

OpenAI also introduced a Decisions API, a stripped-down version of its GPT-6 Luna model built purely for fast, single-choice tasks such as classifying content or deciding an agent's next move. The company says it responds in roughly 150 milliseconds — about ten times faster than calling GPT-6 Luna through the standard API — and launched it as a limited preview. A new premium "Ultrafast" tier for GPT-6 Astra costs six times standard API rates ($60 per million input tokens and $300 per million output tokens) and is bundled into a new $500-per-month "Pro 500" subscription for ChatGPT Work and Codex, alongside Enterprise plans. OpenAI separately partnered with Amazon Web Services to let developers run OpenAI-built agents on AWS Bedrock.

Korean tech outlet ZDNet Korea reported that OpenAI frames the Decisions API partly as a safety and governance tool: before an agent modifies files or visits external sites, a separate AI check can review and block the action. The same report cited a cybersecurity startup, Qualitech, which found that running a specialised monitoring model called Zev cost about $2.94 per task, versus roughly $372 per task for a full frontier model doing the same oversight — illustrating why companies are interested in cheaper, narrower models for continuous agent supervision.

Google folds Vertex AI into Gemini Enterprise

Google took a different route in April 2026, announcing at Google Cloud Next that its Vertex AI platform would be replaced by the Gemini Enterprise Agent Platform, a single system for building, running, governing and improving AI agents. The platform is organised around four areas: a low-code Agent Studio and Agent Development Kit for building agents; an Agent Memory Bank and fast cold-starts for running them at scale; a cryptographic identity and audit trail for every agent plus a central "Agent Gateway" for governance; and simulation and live-traffic tools for improving them over time. Google committed $750 million to its roughly 120,000-member partner network — including Accenture, Deloitte, Capgemini and TCS — to help deploy the platform, and said six trillion tokens a month were already flowing through the Agent Development Kit in production use. Google cited early customers including expense-management firm Payhawk, Japanese restaurant-guide company Gurunavi, and Comcast's Xfinity.

Microsoft scales Copilot through its existing software

Rather than launching new agent products, Microsoft has focused on expanding Copilot inside Office, Teams and Windows, where it already has hundreds of millions of commercial users. The company told investors it had 30 million paid Microsoft 365 Copilot seats by the end of its fiscal fourth quarter on June 30, 2026 — up 50% from 20 million seats three months earlier. CEO Satya Nadella said the number of customers buying more than 50,000 seats had grown more than sevenfold year over year, and that enterprise customers rolling Copilot out to most of their staff grew nearly 75% in a single quarter. In the prior quarter (fiscal Q3 2026, reported in April 2026), Microsoft had said its broader AI business reached an annualised revenue run rate above $37 billion, up 123% from a year earlier. Named deployments include NHS England, where Microsoft said 505,000 clinicians and staff use Copilot and save an average 43 minutes a day each; KPMG, with 276,000 employees; and HSBC, with 200,000 committed seats.

Anthropic, Meta and Cursor compete on different fronts

Anthropic's enterprise push has centred on both a marketplace and rapid revenue growth. On September 23, 2026, it opened the Claude Marketplace with more than 2,000 connectors, plugins and partner products available at launch. The catalogue is split into three parts: connectors and plugins (including ones from Atlassian, Google, Microsoft, Notion and Salesforce); Claude-powered products from partners such as CrowdStrike, Cursor, Harvey, Hebbia, Legora, Lovable and Snowflake; and consulting partners including Accenture, Boston Consulting Group and Deloitte. Businesses can pay for marketplace purchases out of money they have already committed to spend with Anthropic.

That comment referred to Anthropic's revenue, which Amodei said grew 80-fold on an annualised basis in the first quarter of 2026, pushing its run rate from about $9 billion at the end of 2025 to $30 billion by April 2026. Anthropic says it now has more than 1,000 enterprise customers each spending over $1 million a year on Claude, double the figure from February 2026, with Uber and Netflix among its corporate clients.

Smaller and more specialised competitors moved the same week. On September 29, Meta expanded its Muse AI agent with "Muse for Small Business," adding connectors to tools including Shopify, Slack, Asana, Dropbox, Canva, Stripe and QuickBooks, plus access to a business's Facebook, Instagram and ad accounts; it is free for most uses, with paid plans for higher capacity. The launch followed Meta's unveiling of a new "Meta Enterprise Platform" led by Chirantan Desai, the former MongoDB chief executive who joined the company. Separately, AI coding firm Cursor launched two bots for its Teams and Enterprise customers on September 23: "Rollouts," which tracks a code change from pull request to production and can pause a rollout or request a revert if it detects problems, and "Security Reviewer," which checks every pull request for vulnerabilities. Cursor said its own testing showed Security Reviewer cut average review time from 4.8 to 3.8 minutes.

Why corporate AI spending keeps accelerating

Research firm Gartner forecasts worldwide AI spending will grow 49.5% in 2026 to $2.67 trillion, up from about $1.79 trillion in 2025, with growth continuing at a projected 36.2% in 2027. Within that total, Gartner expects AI infrastructure spending to rise 51.2%, AI software 60.2%, AI agents and assistants specifically 77.3%, and spending on generative-AI models themselves 117% — though models still make up a small share of the overall figure. Gartner analyst John-David Lovelock has described the buildout as one of the largest infrastructure undertakings on record, noting that much of the growth reflects existing IT budgets being redirected toward AI rather than entirely new money.

No single vendor leads on every measure in this comparison. OpenAI and Google are both betting on autonomous agents running on their own cloud infrastructure; Microsoft is leaning on the distribution advantage of Copilot already being built into software most large companies use; and Anthropic is growing fastest by revenue while opening its platform to outside partners through a marketplace. Meta and Cursor show that competition for corporate AI budgets now extends well beyond the three or four biggest names, into small-business tools and developer-focused coding assistants.