A confidential internal memo from OpenAI Chief Revenue Officer Denise Dresser was leaked, revealing both a sharp critique of competitor Anthropic and a detailed exposition of OpenAI's strategic direction in enterprise AI. Dresser stated that Anthropic's claimed $30 billion annualized revenue (ARR) was inflated by $8 billion, making its actual figure $22 billion, which is lower than OpenAI's $24 billion.
Dresser highlighted that the enterprise AI market is maturing. Customer demand has shifted from raw capabilities to AI's deep integration into workflows, knowledge bases, control systems, and daily operations, along with effective deployment, trust, and continuous improvement. OpenAI aims to build a trustworthy system that enterprises can confidently build upon, encompassing work-optimized models, an agent platform, deep integration with business contexts, and scalable deployment capabilities.
OpenAI is securing an increasing number of multi-year, multi-product, nine-figure enterprise deals, with existing clients expanding their utilization. Dresser indicated that OpenAI's primary bottleneck is currently capacity, not demand. Consequently, talent acquisition remains a top priority for Q2.
The memo detailed five key strategic priorities for OpenAI in enterprise AI:
1. Winning the Model Layer for Work: Enterprises invest in business outcomes. The upcoming "Spud" model is presented as a significant step for foundational work intelligence. Early customer feedback is positive; Spud is not only OpenAI's smartest model to date but also excels in critical dimensions for high-value professional work, including stronger reasoning, better understanding of intent and dependencies, improved execution follow-through, and more reliable outputs in production. OpenAI's compute advantage ensures continuous capability leaps, leading to higher token limits, lower latency, and more reliable execution of complex workflows.
2. Winning the Agent Platform Layer: The market is transitioning from prompts to agents. OpenAI intends to position "Frontier" as the default platform for enterprise agents—the core intelligence layer for building, deploying, managing, and scaling enterprise systems. Frontier directly links model intelligence with agent performance; as models improve, platform value increases. Deep platform embedding also raises migration costs, making OpenAI an increasingly indispensable core for operational workflows.
3. Expanding Market via Amazon: While the partnership with Microsoft is foundational, it limited OpenAI's ability to meet enterprise clients in the AWS ecosystem. The Amazon collaboration aims to bridge this gap, reaching AWS-native enterprise customers. The Amazon Stateful Runtime Environment is crucial as it expands reach and upgrades the product interface by enabling memory, context, and continuity between interactions, transitioning from stateless model access to systems reliably operating across complex business processes over time. This will reduce adoption friction for AWS-native clients, strengthen OpenAI's position among regulated and security-sensitive buyers, and further integrate the platform from model access into a production-grade runtime for long-running, multi-step agents.
4. Selling the Complete AI-Native Tech Stack: Customers seek a comprehensive platform, not fragmented point solutions. OpenAI offers a full stack including ChatGPT for Work (entry point for knowledge work), Codex (system for software and agent development), API (embedded intelligence engine), Frontier (agent platform), and the Amazon runtime (production-grade, stateful execution layer). This breadth is a significant strategic advantage, allowing OpenAI to meet customers at any entry point and guide them to expand across the entire tech stack.
In its critique of Anthropic, the memo indicated that Anthropic faces compute scarcity, leading to product throttling and availability issues. Furthermore, Anthropic was criticized for being overly focused on coding, with "a single product being a liability" in the platform wars.