Read OpenAI’s latest internal memo about beating the competition — including Anthropic

At a glance:

• OpenAI’s chief revenue officer outlines a five-point strategy to dominate enterprise AI, focusing on models, agents, Amazon partnership, full stack, and deployment.
• The memo criticizes rival Anthropic, accusing it of inflating its run rate by $8 billion and making a strategic misstep in compute acquisition.
• OpenAI emphasizes building a platform with multiple entry points to make it harder for users to switch to competing models.

What happened

OpenAI’s chief revenue officer, Denise Dresser, sent a four-page memo to employees on Sunday about the company’s strategic direction, emphasizing the need to lock in users and grow its enterprise business. The memo, viewed by The Verge, underscores the importance of building a moat around AI products to combat how easily users can switch between whichever model is topping the charts. Dresser, who recently took over duties from former COO Brad Lightcap, stresses focusing on enterprise clients and avoiding “side quests” to prioritize revenue drivers. CNBC also reported on the memo.

Dresser writes that “multi-product adoption makes us harder to replace” and urges a shift from thinking like a company with separate product lines to a platform with multiple entry points and one integrated enterprise offering. She addresses the intensifying competition with Anthropic, noting the market is as competitive as ever. While acknowledging Anthropic’s early coding focus gave it a wedge, she warns against being a single-product company in a platform war. The memo also accuses Anthropic of inflating its stated run rate and criticizes its failure to acquire enough compute, affecting product reliability. Both companies reportedly plan to go public this year.

The five strategic priorities

Dresser outlines five customer-backed priorities to extend OpenAI’s lead in enterprise AI. These priorities focus on winning at every layer of the AI stack, from models to deployment, and leveraging partnerships to expand market reach. The goal is to transition from a product vendor to an operating infrastructure that enterprises trust to build, deploy, and scale AI systems. The priorities are designed to create a flywheel effect: better models drive usage, which leads to deeper integration and multi-product adoption, making OpenAI harder to replace.

1. Win the model layer for work

Enterprises buy models that improve outcomes like faster writing, better analysis, productive coding, effective customer support, and higher-quality decisions. Spud, OpenAI’s latest model, is highlighted as a key step for the next generation of work. Early feedback shows Spud delivers stronger reasoning, better intent understanding, and more reliable output for high-value professional work. It lifts the entire stack, making all key products better and expanding workflows OpenAI can own. The compute advantage enables continuous leaps in capability, with customers experiencing higher token limits, lower latency, and more reliable execution. Every step forward in compute allows training stronger models, serving more demand, and lowering the cost per unit of intelligence.

2. Win the agent platform layer

The market has shifted from prompts to agents, creating an opportunity for OpenAI. Customers want systems that can reason, use tools, operate across workflows, and perform reliably in business environments. This requires orchestration, control, observability, security, integration, and governance. Frontier is positioned as the default platform for enterprise agents, tying model intelligence directly to agent performance. As models improve, the platform becomes more valuable, and as it embeds, switching costs rise. More workflows running through the system make OpenAI central to how work gets done, moving the company from product vendor to operating infrastructure.

3. Expand the market through Amazon

While the Microsoft partnership has been foundational, it limited OpenAI’s ability to meet enterprises where they are, particularly on AWS Bedrock. Since the partnership announcement in February, inbound demand for the Amazon offering has been staggering. The Amazon Stateful Runtime Environment expands access and upgrades the product surface simultaneously by enabling memory, context, and continuity across interactions. This moves beyond stateless model access to systems that operate reliably over time and across complex processes. It expands the market in three ways: lowering adoption friction for AWS-native customers, strengthening position with regulated buyers by running inside their AWS environment, and integrating the platform from model access to production runtime for long-running agents.

4. Sell the full AI-native stack

Customers want a platform, not point solutions. OpenAI’s full stack includes ChatGPT for Work as the front door for knowledge work, Codex for software and agentic development, the API for embedded intelligence, Frontier as the agent platform, and the Amazon runtime for stateful execution. This breadth is a strategic advantage because customers start in different places—some with employees, some with developers, some with internal or external products. OpenAI meets them at their entry point and expands them across the full stack. The flywheel is: better models drive usage, usage drives integration, integration drives multi-product adoption, and multi-product adoption makes OpenAI harder to replace. Dresser urges thinking like a platform company with multiple entry points and one integrated offering.

5. Own deployment

The biggest bottleneck in enterprise AI is no longer whether the technology works, but whether companies can deploy it successfully at scale. DeployCo is a deployment engine that turns product demand into repeatable enterprise transformation, helping companies prove value faster, reduce risk, and scale adoption. It becomes a force multiplier, helping customers move faster and surfacing repeatable deployment patterns that improve product, sales, and customer success. Alongside Frontier Alliance partners, it provides a path to scale execution across the market. Dresser stresses that winning enterprise AI requires not just the best models but the best ability to deploy them into real workflows with measurable value.

Competitive landscape: Anthropic

Dresser acknowledges the market is as competitive as ever, which she believes is ultimately good for innovation and customers. However, she warns that competition can be noisy and distracting, urging the team to stay focused on customers. She specifically addresses Anthropic, building its story on “fear, restriction, and the idea that a small group of elites should control AI.” OpenAI’s positive message—building powerful systems with safeguards, expanding access, and helping people do more—will win over time. She criticizes Anthropic’s strategic misstep of not acquiring enough compute, which is now affecting product through throttling, weaker availability, and less reliability. OpenAI acted faster on the exponential compute curve, gaining a structural advantage. While Anthropic’s coding focus gave it an early wedge, Dresser argues that being a single-product company in a platform war is a liability as AI spreads beyond developers. She also accuses Anthropic of inflating its run rate by roughly $8 billion through accounting treatment, including grossing up revenue share with Amazon and Google. OpenAI reports Microsoft revshare net, aligning with public company standards.

Conclusion: Let's go build

Dresser expresses pride in the team and the opportunity to be at the epicenter of the future. She calls for staying focused, working as one team, and operating at the highest level of excellence. The market is theirs to win, and she urges execution accordingly.