OpenAI takes direct aim at Anthropic in leaked enterprise memo

OpenAI chief revenue officer Denise Dresser just sent a blunt four-page internal memo to staff, and she didn't mince words: the AI market is in the middle of a brutal dogfight, and OpenAI needs to lock down enterprise customers before they wander off to the next top-ranked model.

Dresser, who recently took over core duties from former COO Brad Lightcap as he shifts to special projects, warned that customer churn is the biggest risk in a world where users switch models at the drop of a hat. Her solution: kill off the side quests and turn OpenAI into an integrated platform company that enterprise buyers cannot afford to rip out.

With both OpenAI and Anthropic reportedly preparing for initial public offerings this year, the gloves are officially off.

The enterprise lock-in playbook

Dresser told employees to stop thinking like a team managing isolated products. Instead, OpenAI wants multiple front doors leading to one enterprise ecosystem.

Her memo outlines five priorities for the second quarter. The first is dominating work tasks with a newly disclosed model code-named Spud, described as OpenAI's smartest model yet, tuned specifically for multi-step reasoning, intent mapping, and production reliability.

The company is also pushing Frontier, an enterprise agent platform designed to make OpenAI the operating system for corporate automation. To bridge the gap between software and real-world rollouts, Dresser announced DeployCo, an internal deployment engine aimed at solving enterprise adoption bottlenecks alongside partner alliances.

Distribution is also shifting fast. While Microsoft remains a key foundation, Dresser admitted that exclusivity kept OpenAI away from massive AWS-native workloads. Following a partnership announced in late February, inbound demand from Amazon Bedrock has been staggering. OpenAI is now rolling out a stateful runtime environment inside AWS to give regulated buyers long-running agent workflows with memory and strict governance.

Firing directly at Anthropic

The spicy part of the memo is a direct takedown of longtime rival Anthropic.

Dresser argued that while Anthropic found an early foothold with software engineers, you do not want to be a single-product company in a platform war. She directly accused Anthropic of inflating its stated run rate and called its decision not to secure massive compute reserves a major strategic misstep.

She also took aim at Anthropic's safety-first brand, writing that their pitch is built on fear, restriction, and the idea that a small group of elites should control AI. The salvo echoes CEO Sam Altman, who posted in February that Anthropic basically sells an expensive product to rich people.

According to Dresser, OpenAI's enterprise pipeline is already yielding multi-year, multi-product deals in the nine-figure range. She claims the company's biggest hurdle right now is not demand, but delivery capacity and engineering talent.

Why it matters

We think this memo marks the end of the honeymoon phase where sheer model benchmarks alone could win the market. When raw intelligence becomes a shifting commodity, whoever controls enterprise workflows and storage runtimes wins the business.

For enterprise buyers and IT leaders, Dresser's strategy is a double-edged sword. OpenAI's integrated stack across ChatGPT for Work, Codex, and AWS runtimes will certainly make deployment smoother. But make no mistake: multi-product adoption is designed specifically to raise your switching costs and make it painful to swap in competing models later.

Strategically, teaming up with Amazon Bedrock and launching hands-on deployment services with DeployCo looks like a masterclass in classic enterprise software playbook execution. It turns a fragile API subscription into sticky corporate infrastructure.

Still, there is an underlying vulnerability here. Leaning heavily on defensive moats and attacking a rival's compute capacity suggests OpenAI feels the heat. If model switching is genuinely as effortless as Dresser fears, holding enterprises captive via bundled contracts might only work until a leaner, cheaper alternative delivers comparable results.