OpenAI Refocuses on Enterprise Amid Rising Anthropic Competition

OpenAI is experiencing a profound change in its strategic direction. Once celebrated for wide-ranging consumer innovation, the company’s leadership now sees business users as the key to future growth and profitability. CFO Sarah Friar, formerly of Nextdoor, reported that since her hiring in 2024, the percentage of revenue from business clients has doubled from 20% to 40%, and is expected to reach 50% by year’s end. This shift is driven by a growing realization that while over 900 million weekly users interact with ChatGPT, 95% do not pay, placing enormous strain on costly, high-performance compute infrastructure ^[1][2][3].

The recalibration is evident in OpenAI’s product decisions. High-profile consumer projects like the Sora video generator app and advertising on ChatGPT have been shelved, with resources now redirected toward models and tooling for “high-value professional work.” The appointment of Slack’s Denise Dresser as chief revenue officer further emphasizes a laser focus on enterprise sales, as she courts corporate leaders looking to automate complex information work ^[1][3].

CEO Sam Altman acknowledged on the 'Mostly Human' podcast that tech companies must avoid spreading themselves thin with too many ventures, a lesson internalized after flirtations with consumer entertainment, ad-tech, and fringe proposals for paid adult content. The company is instead concentrating R&D and go-to-market efforts on sophisticated, domain-specific models tailored for business and industry ^[1][3].

Anthropic, OpenAI’s most formidable rival, has rapidly capitalized on the growing enterprise appetite for reliable AI assistants. Recent industry data reveals that nearly a third of American businesses now pay for Anthropic’s offerings, up more than six percentage points in a month, while OpenAI’s own enterprise adoption plateaued at around 35% ^[5]. The surge is underpinned by Anthropic’s strategic focus: while ChatGPT first built trust among consumers and expanded into the workplace, Claude was designed with professional precision and technical risk in mind, making it a preferred choice for contract analysis, code review, and complex research out of the gate ^[5][6].

Anthropic’s newly released Claude Mythos exemplifies the company’s audacious trajectory. Mythos is so advanced, particularly in cybersecurity tasks such as vulnerability discovery and patching, that Anthropic has limited its use to an exclusive cohort of vetted enterprise customers. Meanwhile, its more broadly available Opus 4.7 model now leads its commercial portfolio. This staged availability marks a fundamental reconfiguration of how best-in-class AI is commercialized and controlled ^[1][5][6].

The enterprise focus has proven lucrative: Anthropic’s annualized revenues have soared to $30 billion—outpacing much of OpenAI’s reported figures—demonstrating that targeting business-critical workflows from the outset can rapidly close competitive gaps ^[2].

A decisive trend reshaping the field is the restriction of access to the most capable AI models. Both OpenAI and Anthropic are now implementing strict gatekeeping around powerful new releases. For instance, while Anthropic’s Mythos has demonstrated an ability to outperform human cybersecurity experts in identifying zero-day vulnerabilities, it is only available to a select few major defense and tech firms, likely fewer than 50 globally. OpenAI has mirrored this approach, limiting its Spud model and its own advanced cybersecurity tools to a small subset of clients due to both commercial and ethical considerations ^[1][6].

This marks a sharp departure from earlier years, when rapid API growth and open, generalized model access dominated. The reasoning is part technical and part regulatory: as AI models cross benchmarks like Cyber-Bench with capabilities that pose real infrastructure threats, providers are compelled to treat access as a matter of national security and compliance. The result is a tiered AI ecosystem, with the most potent technologies sequestered behind layers of vetting, legal agreements, and enterprise contracts ^[6].

Such measures are seen as necessary to prevent abuse and mitigate catastrophic risk, but they also mark the end of parity in AI access—a development with broad implications for innovation, digital sovereignty, and market competition ^[6].

The current wave of enterprise AI is defined less by core language model capability and more by managed infrastructure, context management, and agentic workflow reliability. Anthropic has repositioned Claude not as merely a chatbot but as a managed agent environment, where the AI acts as an operating system for executing multistep business processes. Internal benchmarks show that out-of-the-box autonomous systems built on open-access APIs fail in 42% of workflows requiring five or more tool calls, usually due to context management problems rather than model intelligence per se ^[6].

By deeply integrating context management (with Claude 3.5’s server-level context window at 200,000 tokens and tool log optimization), Anthropic claims to have reduced failure rates to under 5%. Latency issues, which add up to 15% execution delay in DIY agentic loops, are nearly eliminated in their managed environment, yielding both cost and performance advantages that are difficult for enterprises to match with in-house orchestration ^[6].

For OpenAI, this new paradigm means prioritizing product offerings like GPT-Rosalind, designed for life sciences and pharmaceutical R&D, and Spud, which promises stronger reasoning, intent comprehension, and production reliability. These models are engineered for seamless integration into business processes, not just for generating text or answering ad hoc questions. The move away from general-purpose tooling toward domain-specialized and production-grade agentic systems is what distinguishes the next phase of enterprise AI ^[1][3][6].

The introduction of managed, restricted-access AI agents also has profound economic and operational consequences for providers and customers. Both OpenAI, recently valued at $852 billion, and Anthropic at $380 billion, have yet to turn a profit, as infrastructure costs rise with usage and development pace ^[1][2][3]. To sustain growth and the astronomical compute required for state-of-the-art models, business sales and high-value contracts are essential—not only for revenue but for shaping the future development roadmap ^[1].

For practitioners, this means fewer open releases and more sandboxed environments where AI providers own the full technical stack, from inference to orchestration to compliance. Democratized access is giving way to strategic partnerships and service agreements, especially for functions involving sensitive data, AI automation, or industry-specific compliance ^[5][6].

This shift resets expectations about where and how AI breakthroughs will impact workflows, as technical sophistication is increasingly tied to managed services, reliability, and supply-chain security. For the AI industry, it is an unmistakable sign of maturity, but also a new era of constrained access and stratified innovation ^[1][5][6].