From 0 to $29.3 billion in two and a half years: Cursor sets a new global AI programming valuation ceiling

the (whole) worldTop valuations for AI programming companies, has just been refreshed again.

AI Programming's Hot Chicken of the Moment Cursor(the company behind Anysphere), has once again stretched the industry's imagination - the latest valuation soared to $29.3 billionApprox. 207.9 billion.

On November 13, Anysphere announced the completion of the $2.3 billion Series D financing(about 16.3 billion yuan). After this round of funding landed, the company's valuation went from $9.9 billion at the end of the previous round (in June this year) to $29.3 billion - almost tripling.

Cursor's official blog reveals that this round of funding has a stellar lineup of investors, with new additions including NVIDIA, Google and top fund CoatueCursor's existing investors, Accel, Thrive, a16z, and DST, have also raised additional capital. The round was led by Accel and Coatue.

However, according to CNBC, Anysphere's CEO Michael Truell said that their startup has no plans to go public in the near future:

“Our immediate focus is on growing the company and the team, and we have a lot of work to do before we think about things like going public.”

  Two and a half years after its release, CursorBelt company valuations explode

In 2023, a new product called the Cursor Native AI IDE It went live quietly. Since then, it's quickly become a hit in the global developer community and one of the fastest-growing SaaS products today, with a minimal marketing budget.

The rise of Cursor is a step towards the emerging trend of “Vibe coding” in software engineering: developers use natural language to talk to AI, which automatically generates, modifies, and refactors the code. The core of this paradigm is not just modeling power, but rather Cursor ×Composer's dual-engine architecture - Composer is a model trained to write code, the equivalent of Cursor's “brain”, and the Cursor team has rewritten the IDE's underlying architecture to carry that intelligence, giving it a lead in contextual memory, multilingual support, and reasoning speed. The Cursor team has rewritten the underlying IDE architecture to carry this intelligence, making it better than similar tools in terms of contextual memory, multi-language support, and reasoning speed, and allowing Composer to truly understand and manipulate real code bases rather than generate text in isolation.

With this architecture, Cursor has achieved phenomenal business growth within two years of launch: annualized recurring revenue (ARR) exceeded $500 million, more than 1 million daily active users, 360,000 paid users, 50,000 enterprise reach, and more than half of the Fortune 500.

The outbreak of this round also stems from the release of industry pain points: shortage of engineers, high costs, heavy enterprise legacy systems that are difficult to maintain, and repeated debugging that takes up a lot of time; Cursor's product form is a natural “cure” for these problems, and with the free version driving trials and the community ecosystem driving proliferation, the product flywheel accelerates.

principalThere is little hesitation about this growth curve.

Starting with a $60 million funding round in July 2024, through a $105 million Series B that closed in December of that year, a $900 million Series C in May 2025, and a $2.3 billion Series D in November 2025, Anysphere's valuation jumped from $2.6 billion to $29.3 billion in less than a year.

This pace is unheard of on the global AI programming circuit. Its investors include top tier organizations such as Accel, Thrive, a16z, DST, NVIDIA, Google and Coatue.

Another episode had been rumored in the industry:OpenAI has led Anysphere in seed rounds and made two attempts to acquire it, with the highest offer reportedly reaching the tens of billions of dollars levelThe company was unable to do so due to Anysphere's refusal to sign an exclusive modeling agreement.

However, there are also concerns behind the rapid growth. It has been discussed that Cursor is deeply dependent on the underlying models (OpenAI, Anthropic) and the VS Code ecosystem, and that theoretically “a single API change” could pose structural risks.

At the same time, the accelerated catch-up of big open source models is leading to a convergence of underlying capabilities and a continuous compression of technical barriers.

In the face of these challenges, Anysphere is trying to embed Cursor deeper into the enterprise R&D chain - from testing, DevOps, to observability and data return; trying to build a next-generation moat with “full-link integration” that is harder to replace. Moat.

a16z's latest podcast reveals the exclusive story behind Cursor

Cursor is at the center of industry attention today, but it didn't start out that way.

A few days ago, a16z interviewed Michael Truell, the founder of Anysphere, in a podcast. The founder, who drove the company to a $200 billion valuation in two and a half years, gave the first systematic account of Cursor's real growth trajectory - how it made a decisive turnaround from the almost-unworkable direction of “3D Robotics CAD”, and eventually Cursor's founder is the first to systematically tell the real growth trajectory of Cursor - from the direction of "3D robot CAD", which was almost impossible, how to make a decisive U-turn at a key turning point, and finally enter the core of the global AI programming track.

In this interview, Michael shares Cursor's product positioning, key design tradeoffs, why he insisted on starting with VS Code, how the team made a usable first version of the IDE in two weeks with minimal resources, and their real-world decision-making process for dealing with infrastructure bottlenecks, multi-cloud, multi-vendor architectures, and modeling supply chain risks as scale skyrocketed.

At the same time, as one of the most highly regarded AI programming companies, Cursor's judgment on the future of the industry is also quite forward-looking: software engineering is moving from “writing code” to “determining whether the code is correct,” and the role of engineers is being redefined as The role of the engineer is being redefined as a technical decision maker rather than a keyboard operator.

Photo credit: Anysphere, from left to right, Aman Sanger, Arvid Lunnemark, Sualeh Asif, Michael Truell

On the show, Michael retraces Cursor's shift from the wrong direction to a growth path. A few years ago, the team was experimenting with a project that later proved difficult to justify - automating 3D robotic CAD design using large models. The idea sounded advanced in the early days, but in practice, they ran into structural problems such as data scarcity, incompatibility between model reasoning and 3D scenarios, and difficulty in filling in industry knowledge, and progress stalled for a long time.

As Truell recalls, this experience made the team realize that they did not have enough mechanical engineering background and could not bridge the professional gap with large models alone. the CAD project reached a bottleneck, and the team's pressure built up. The reorientation was both accidental and inevitable. The four founders, all deep engineers familiar with the command line and having tried GitHub Copilot at a very early stage, observed that for the first time, AI could intervene in software engineering at a higher quality, but the supply of tools was still very limited. After discussion, the team abandoned CAD and refocused on more familiar territory.

Since then, the company's path has changed, and Cursor chose to start with VS Code rather than build a new IDE from scratch, believing that the core issue was not “reinventing the editor” but “making the model work in a real development process”. The team believes that the core issue is not "reinventing the editor" but "making the model work in the real development process. Based on this judgment, the first version of Cursor was completed in a few weeks and quickly spread to the developer community after its release.

With no marketing budget and early growth coming primarily from the community, Cursor quickly spread through the developer community, became a frequently discussed AI programming tool, and the company entered a high growth cycle.

Rapid growth also brings more complex challenges.

At its peak, Cursor's API calls once accounted for a double-digit percentage of a cloud provider's revenue; the team had to decentralize model vendors, build its own inference and training system, and rewrite some of its infrastructure in a short period of time.

These adjustments were made more in response to the pressure for stability brought about by growth than in the pursuit of “technological leadership”. The team realized that only by mastering more underlying capabilities could they differentiate themselves in the engineering tools market in the future.

The Cursor is also organized in an atypical way. It'sThe hiring process involves candidates writing code in the office for two days, rather than traditional whiteboard interviews. This approach gives candidates direct exposure to the real work environment and helps the team judge the match.

The M&A strategy is also centered around “talent density”. For example, they acquired a team of five people, but the key reason was that one of the members was responsible for the core development of the predecessor features of GitHub Copilot, not to fill out a product module.

For Cursor, this type of organizational structure is the underlying strategy for dealing with the uncertainty of high growth.

The future of the industry is often described in terms of “automated programming,” but Truell is more cautious. He believes that full automation is still a long way off. The industry is in the early stages of tool upgrades, and differences in modeling capabilities, engineering practices, and corporate environments make it difficult to fully implement “unmanned engineering” in the short term.

He likened the current phase to “between the iPod and the iPhone” - new forms are emerging, but technology paths and usage patterns are still evolving.

Cursor's goal is to maintain speed during this transition period and explore the possibilities of the next generation of development tools.

Looking at Cursor's trajectory, the competition for AI programming tools has shifted from auto-completion and single-file generation to deeper integration of the engineering chain, including testing, deployment, observability, and data return.

Cursor's rise is not a typical “one successful product iteration” story, but rather a case of pulling back from the wrong direction to find a position where the demand was clear and the team had an advantage. Its growth has come from a combination of product-scenario fit, team density, and infrastructure capabilities, while the industry backdrop of a tight supply of engineers and increasing complexity of enterprise systems has been a long-term driver on the demand side.

With its valuation rising to $29.3 billion, Cursor is one of the most talked about companies in this space. But it still has to deal with underlying model competition, upstream and downstream dependencies, and the degree of consolidation in the enterprise market in the future.

The following is a transcript of this a16z interview with Michael Truell.

a16z: I've had people say “you're the 3D version of Cursor” to which you responded “actually Cursor started out as a 3D company”. Can you talk about that origin story?

Michael Truell: Absolutely. We really got started with Cursor, driven by some of the experiences my co-founders and I had at school working with other places. Two key moments that got us excited were the first time we used an early practical AI product like GitHub Copilot, which made us realize that AI wasn't just research anymore, but was really a tool for building real-world systems, and seeing Scaling Laws at the beginning of the 2021-2022 period, which showed that performance naturally improves as models get bigger. The idea for Cursor came from a whiteboarding exercise, where we imagined “Cursor for X” as an AI tool for vertical knowledge work, thinking that Microsoft would focus on coding, while we wanted to do something more niche and less directly competitive. We had a lot of mechanical engineering friends around us, so we started with CAD, but quickly realized that we had absolutely no intuition about the day-to-day life of a mechanic, and the founder's market fit was extremely low. Looking back, I wish we had gone straight to an internship for those 6 or 7 months. In the end, we abandoned the mechanical engineering direction and went back to what we really love: programming.

a16z: I have a theory - Cursor's early success was due to your “focus”. You chose VS Code instead of rewriting the editor or being an engineer's agent. Do you agree?

Michael TruellI think it makes a lot of sense. A lot of companies were going in a very “sci-fi” direction, trying to rewrite editors, make engineer agents or base models, whereas we were extremely focused. We really wanted to “own the interface layer”, the surface that engineers interact with every day. Even though everyone was saying “no one is going to change editors”, we knew that wasn't the case because we ourselves had migrated from command-line Vim to VS Code - if the mousetrap was better, engineers would migrate. We didn't touch the model at first, we just focused on getting the product out there and making it really usable.

a16z: How did you get Cursor in front of users when you were moving forward in an extremely chaotic competitive environment?

Michael TruellWe had limited funding and were still figuring out how to expand the team, but one of the fastest things we did was to make an IDE that we could use every day in two weeks, and we didn't even fork from VS Code at first, we wrote one from scratch, hacking every day and treating ourselves as the primary users. We hacked every day and made ourselves the primary users. A few weeks later we handed it off to someone else, a few months later we released a beta, and the momentum started to build. The “commitment mechanism” that really kept us going was the monthly updates to our investors - no one may have read them, but we had to write them.

a16z: Is it intentional that when the momentum gets bigger and users want you to support CLI, IntelliJ, etc., you still insist on not doing it?

Michael Truell: It's deliberate. The four co-founders spend three meals a day discussing core strategy: Should we scale? Do we model? Do we make a new IDE? From day one we've been determined to own the interface layer, not to be a plug-in. In the future we will definitely go into the modeling side, but we won't start there.

a16z: You had serious scale issues early on, and one day you told me you were “falling off a big cloud” because the other side couldn't handle the traffic. How did you solve that?

Michael Truell: Yes, we went through two phases of scale challenges. The first was a very small team with a very fast-growing service, where we ran our own internal file synchronization system (like a mini Dropbox), our own search engine, and a huge Kubernetes cluster (with only 5 people). Many of these were “normal services on a normal cloud”, but we pushed them to their limits. This was solved by architectural improvements and team expansion. The second phase was when we pushed API providers to double-digit percentages of their revenue, and they even did capacity planning and even raised money as a result. So we learned to decentralize the providers, collect “all the tokens on the market” across platforms, and gradually started to train and reason on our own before the system stabilized.

a16z: Are you going to go the “multi-cloud + multi-vendor” route in the future, or are you going to build your own infrastructure?

Michael Truell: We've been the multi-cloud route from day one. We use Databricks, Snowflake, AWS, GCP, Azure, and PlanetScale. We've had problems with RDS scaling to capacity, sharding issues, and even AWS's “never shard” service ended up sharding. We believe that different vendors are good at different things, and that combinations are the strongest.

a16z: You've expanded from one product to multiple products such as BugBot, CLI, Team Collaboration, etc. How do you decide to prioritize?

Michael Truell: We try to say “no” to a lot of things, but we have to be a multi-product company. We believe that there will be a whole “AI programming toolkit” and that Cursor is the “glass surface” that engineers stare at every day. Editors are not only changing the way engineering is done, but also the way teams work together, so BugBot and CLI are natural strategic additions. Multi-product GTM is complex and we're still learning, but the early results are exciting.

a16z: Your hiring process is unique, especially the “come to the office for two days to work on a real project”. Tell us about your philosophy?

Michael TruellWe tried to do away with this process several times, but have kept it in place until now. Anyone joining the editor or design team has to come into the office for two days, and instead of a whiteboard interview, they're just given a workstation, a computer, three real projects to work on, and a frozen version of an old codebase - and then “get to it”. This process shows whether the candidate can work from start to finish on a real code base, has a sense of product, is self-driven, and is a partner we'd like to work with. Candidates also get to really see what Cursor is like. Even with a team of over 200 people, this process remains in place. Early on in sales it was even more direct - giving real customers, quotas, and data directly and letting them tell us how to sell.

a16z: You've done multiple acquisitions as a two-year company. What's the logic?

Michael Truell: We follow one principle: get the most talented people at all costs. We've done crazy things: if the person rejects us, we fly to the other side of the world to meet them; if they continue to reject us, we maintain the relationship for six months, rekindle the conversation, and ultimately bring the person on board. Many of the best people are entrepreneurs themselves, and M&A is the natural way to go. We also believe that there are a lot of complementary products in this market and that they should be “packaged”. That's why we used M&A as a strategic tool earlier than most companies. The first real M&A was Supermaven - the creators of TabNine and OpenAI researchers. They were doing complementary modeling and we were doing complementary modeling, very complementary. We spent a couple months approaching them and were very aggressive.

a16z: Someone asked: Cursor is disrupting software, and Cursor itself is software - will you be automated out of yourselves?

Michael Truell: I would like to answer this in two ways. First, we are still far from fully automating software development. Despite the huge demand, development in large organizations is still very inefficient, the difficulty of automation has been grossly underestimated, and there is a long and confusing transition period ahead. Second, our market is moving from the “iPod moment” to the “iPhone moment”. There will be many “iPhone moments” in the future, and companies will have to keep creating them or they will be obsolete. That's both the challenge and the physics of the space, and that's why Microsoft is having such a hard time competing in this space.