Cursor Bets on Product, Not Models, to Beat OpenAI and Anthropic

Anysphere (company behind Cursor) CEO Michael Truell has a bold theory about why OpenAI and Anthropic won’t crush his $29.3 billion AI coding startup: they’re building engines when developers need cars.

Speaking at Fortune’s AI Brainstorm conference this week, Truell drew a sharp distinction between foundation model providers and application builders. “It would be like taking an engine and a concept car around it instead of a whole end-to-end car that was manufactured,” he said, describing how he views competitors’ coding products compared to Cursor.

The analogy captures a strategic gamble that has propelled Cursor from a research project to one of the most valuable AI startups in history. Rather than competing on model development, Truell’s company aggregates intelligence from multiple providers—including the very companies positioned as threats—while focusing relentlessly on the user experience developers actually need.

The Integrator’s Advantage

Cursor’s approach inverts the typical AI startup playbook. Instead of racing to train frontier models, the company sources the best available intelligence from OpenAI, Anthropic, and others, supplementing with in-house models where product-specific optimization matters most.

“What we do is we take the best intelligence that the market has to offer from many different providers,” Truell explained. “We also do our own product-specific models in places. We take that, we build it together and integrate it then also build the best tool and end UX for working with AI.”

The results suggest the strategy is working. Cursor reached $1 billion in annualized revenue in 2025, having crossed $500 million ARR just months earlier. The company now counts over half the Fortune 500 as customers, including NVIDIA, Uber, and Adobe. Its Series D round in November brought in $2.3 billion from Accel, Thrive, Andreessen Horowitz, and notably, both NVIDIA and Google as new investors.

From Individual Coders to Team Infrastructure

Truell signaled a significant strategic pivot at the conference: Cursor is moving from serving individual developers to “thinking about teams as the atomic unit that we serve.”

This shift acknowledges how AI coding tools are maturing. When Cursor launched, developers used it for quick JavaScript questions. Now, Truell says, users turn to it for “hours of work.” That evolution demanded new pricing—Cursor has moved toward consumption-based models—and new product thinking focused on collaborative workflows like code review.

The team focus also provides competitive moat. While AI coding assistants proliferate, few have cracked enterprise deployment at scale. Cursor’s code review product, which Truell says some customers use to analyze every pull request whether written by humans or AI, represents exactly the kind of workflow integration that’s difficult for model providers to replicate without building full applications.

The Competition Question

OpenAI approached Anysphere earlier this year about a potential acquisition, but talks went nowhere. OpenAI then pursued Windsurf, another fast-growing AI coding assistant, entering a $3 billion acquisition agreement in May—but that deal collapsed in July when the exclusivity period expired. Microsoft’s IP rights over OpenAI acquisitions proved a dealbreaker; Windsurf’s leadership refused to let their technology fall under Microsoft’s umbrella given GitHub Copilot’s competing position. Google subsequently hired Windsurf’s CEO and key engineers via a $2.4 billion licensing deal, while Cognition acquired the remaining assets.

Anthropic’s Claude Code has grown aggressively, reaching $1 billion in annualized revenue and integrating directly into Slack. GitHub Copilot, backed by Microsoft and OpenAI, remains the incumbent to beat. Google has pushed Gemini into development workflows. The market is crowded and getting more so.

Yet Truell’s confidence appears rooted in a specific bet: that the application layer will capture more value than the model layer. If foundation models commoditize—as pricing trends suggest they might—then the companies building the best interfaces on top of them could prove more defensible than the model providers themselves.

Cursor’s internal models reportedly “generate more code than almost any other LLMs in the world,” according to the company. That claim, if accurate, suggests the line between integrator and model developer may be blurring. Cursor is becoming a significant AI research operation in its own right, now with over 300 engineers and researchers.

The Valuation Test

At $29.3 billion, Cursor carries expectations that require continued hypergrowth. The company tripled its valuation in five months between its Series C and Series D rounds. Enterprise revenue grew 100x in 2025 alone.

Truell says an IPO isn’t on the horizon—the focus remains on building features. But the pressure to justify that valuation will eventually demand an answer to whether product excellence alone can defend against well-resourced competitors who could integrate similar features into their own offerings.

If Cursor wins, as Truell framed it, it won’t be by out-modeling OpenAI or Anthropic’s Claude. It will triumph because it out-products them for the job developers really want done—sending better code to the customer faster, with fewer surprises. That’s a bet on execution over scale, on integration over invention.

Whether that bet pays off may determine not just Cursor’s future, but whether the AI application layer can sustain independent companies or inevitably consolidates under the model providers who supply its intelligence.