OpenAI is moving decisively to strengthen its position in the high-stakes enterprise artificial intelligence market. Newly appointed chief revenue officer Denise Dresser told staff that the company’s alliance with Amazon has become a major catalyst for growth, even as its longstanding relationship with Microsoft increasingly constrains how it reaches customers.

In an internal memo sent Sunday and reviewed by CNBC, Dresser laid out what amounts to a blueprint for OpenAI’s next phase of commercial expansion: diversify distribution, deepen enterprise penetration, and blunt the growing momentum of rivals such as Anthropic and Google. The message arrives at a pivotal moment for the company, which is rapidly shifting from consumer-led growth driven by ChatGPT to a more durable revenue model anchored in corporate clients and large-scale infrastructure partnerships.

The shift is centered on Amazon Web Services and its Bedrock platform, which allows enterprises to access a range of major AI models, including OpenAI’s, within existing cloud environments. For large companies, this matters because AI procurement increasingly follows existing cloud relationships rather than standalone software decisions.

“Our Microsoft partnership has been foundational to our success. But it has also limited our ability to meet enterprises where they are — for many that’s Bedrock,” Dresser wrote.

“Since we announced the partnership at the end of February, inbound demand from our customers for this offering has been frankly staggering.”

Those remarks underscore a subtle but important realignment in the AI industry’s power structure. Microsoft’s more than $13 billion backing of OpenAI since 2019 was instrumental in transforming the company from a research lab into the commercial force that ignited the generative AI boom. Yet as OpenAI scales, exclusivity is giving way to strategic flexibility.

The Amazon partnership is not merely about additional capital or compute. It is about distribution at scale. AWS remains the world’s largest cloud provider, and Bedrock gives OpenAI direct access to enterprises that have already standardized mission-critical workloads on Amazon’s infrastructure. That dramatically lowers friction around deployment, governance, procurement, and billing.

This means OpenAI is no longer content to be primarily associated with Azure. It is now positioning itself as a multi-cloud enterprise platform.

This comes as competition in the corporate AI market intensifies. Anthropic’s Claude model has emerged as a formidable rival, particularly in enterprise environments where coding, reasoning, and workflow automation are central use cases. Dresser’s memo appears aimed in part at countering that narrative internally and externally.

The momentum behind Claude has become a defining conversation in the industry. At the HumanX conference in San Francisco last week, Glean CEO Arvind Jain captured the market mood in unusually vivid terms.

“It has become a religion, that’s the level of that mania,” Jain said.

That description is more than colorful language. In enterprise software, perception often compounds adoption. Once chief information officers and engineering teams converge around a preferred model, usage can accelerate rapidly through internal pilots, department rollouts, and enterprise-wide licensing agreements.

Dresser directly challenged Anthropic’s positioning. She said the rival’s strategy is built on “fear, restriction, and the idea that a small group of elites should control AI,” while arguing that OpenAI’s “positive message” will prevail over time.

She also sharpened the competitive case on infrastructure, saying Anthropic has made a “strategic misstep to not acquire enough compute.”

This point cuts to the heart of the AI arms race because in today’s market, model leadership is inseparable from access to compute. The companies with the deepest access to GPUs, custom accelerators, and hyperscale data-center capacity hold a decisive advantage in training frontier models and serving enterprise inference workloads reliably.

OpenAI has clearly recognized this reality. Beyond Microsoft, it has increasingly tapped providers such as Oracle, Google, and CoreWeave to expand capacity, signaling that its infrastructure strategy is now explicitly multi-provider.

The evolving Microsoft relationship remains central to this story. While both companies continue to describe the alliance as strategic, the relationship has shown clear signs of strain as they expand into overlapping territory. Microsoft’s consumer and enterprise push with Copilot increasingly competes with OpenAI’s own software ambitions, while OpenAI’s growing cloud diversification reduces its dependence on Azure.

This competitive overlap has been building for some time. Microsoft formally added OpenAI to its list of competitors in its annual report in 2024, placing it alongside hyperscale peers such as Amazon, Apple, Google, and Meta.

Talking about the commercial stakes, Dresser recently disclosed that enterprise now accounts for 40% of OpenAI’s total revenue and is on track to reach parity with its consumer business by year-end.

That metric is especially significant in light of the company’s latest valuation, which exceeded $850 billion in its most recent fundraising round. Investors are increasingly focused not just on user growth but on recurring enterprise revenue, contract durability, and long-term monetization.

The memo also lands as OpenAI appears to be laying groundwork for a potential public offering. Reports in recent days suggest preparations are accelerating, with the company considering an IPO as soon as this year.

Against that backdrop, Dresser’s internal note reads as both an operational directive and a market signal. She urged staff to remain disciplined amid volatility and stay close to customers.

“The market is ours to win, let’s execute accordingly,” she wrote.

Duolingo has quietly executed a significant course correction in its workplace AI strategy, stepping back from a policy that formally linked employee use of artificial intelligence tools to performance reviews.

The move speaks to a broader reckoning underway across white-collar workplaces over how AI adoption should be measured.

The reversal, disclosed by Chief Executive Luis von Ahn during an April 10 podcast appearance, marks a shift from measuring tool usage to evaluating business outcomes, a distinction that is increasingly becoming central to how companies govern AI in the workplace.

Speaking on the Silicon Valley Girl podcast, von Ahn said the company reconsidered its earlier approach after employees began questioning whether management was rewarding the mere use of AI rather than the quality and effectiveness of the work itself.

“Do you just want us to use AI for AI’s sake?” he recalled employees asking.

That question appears to have cut to the heart of a growing management dilemma in the AI era. For much of the past year, companies across the technology sector have been racing to institutionalize AI use, often folding it into hiring, workflow expectations, and performance assessments. But Duolingo’s retreat suggests that a more mature phase is emerging, one in which executives are beginning to distinguish between AI adoption as optics and AI adoption as measurable productivity.

Von Ahn acknowledged that the original policy had begun to feel misaligned with the company’s broader performance philosophy.

“At the end, we backtracked,” he said, adding that the company’s primary concern is whether employees are doing their jobs as effectively as possible, with AI serving as a tool rather than a requirement.

Rather than compelling AI usage as a compliance metric, Duolingo is now emphasizing output, judgment, and the suitability of the method. In newsroom terms, this is less a retreat from AI and more a recalibration of governance.

The company had initially taken a far more assertive position. In an internal 2025 memo, Duolingo laid out an “AI-first” framework that included tracking AI use in reviews, prioritizing AI fluency in recruitment, and gradually phasing out contractor work deemed automatable.

That memo sparked a backlash online and raised concerns among users and workers that the language-learning platform was moving toward a technology-led restructuring of its workforce.

The latest reversal suggests the company may have recognized the limitations of turning AI into a key performance indicator. The problem with such metrics is that they often incentivize performative usage rather than productive usage.

An employee may use AI frequently without improving outcomes, while another may deploy it selectively to achieve materially better results. Counting prompts, sessions, or tool engagement does not necessarily capture value creation.

This is where Duolingo’s shift becomes more broadly relevant as it reflects a growing realization among employers that AI is best evaluated as a means, not an end. The real question is not whether staff members use AI, but whether they are producing faster, better, more scalable work because of it.

That framing is especially important for Duolingo, whose core product relies heavily on pedagogy, user psychology, linguistic nuance, and content design. A spokesperson reinforced that point, saying the company’s work still depends on “human judgment, expertise, and creativity,” while AI tools function as support systems rather than decision-makers.

This also helps explain why the company is being careful in its messaging. Duolingo has been using AI for years, particularly in personalization, lesson scaling, and expanding into adjacent subjects such as math, music, and chess. More recently, AI has been central to its content acceleration strategy.

What has changed is not the company’s commitment to AI, but the way it wants employees to engage with it.

The broader corporate context makes this move even more notable. Other major technology firms are moving in the opposite direction. Reports indicate that some teams at Meta and Google have introduced explicit expectations around AI usage, with such activity in some cases feeding into evaluations.

Against that backdrop, Duolingo’s adjustment stands out as an early acknowledgment that AI governance cannot be reduced to usage quotas.

But performance reviews shape behavior. Once AI becomes a scored category, employees may feel compelled to use tools even where they add little value, potentially undermining craftsmanship, judgment, and originality. By reversing course, Duolingo is effectively saying that the quality of the lesson, product feature, or user experience matters more than whether an AI tool was involved in creating it.

That may prove to be one of the more instructive lessons for corporate America as the AI transition deepens. The first phase of the AI workplace revolution was about adoption. The second phase, now underway, is about intelligent accountability.

However, Duolingo’s latest move suggests that the companies most likely to benefit from AI may be those that stop measuring the tool itself and start measuring what it genuinely improves.