Fierce Competition in AI Programming: Devin Raises 48 Billion in Financing, Cursor Acquired by SpaceX, Copilot Overtaken by Claude Code

September 9, AI programming company Cognition announced the completion of a $2 billion Series E financing, with a post-money valuation of $48 billion (approximately RMB 3.4 trillion). According to foreign media reports, institutions including a16z participated in this round of financing.

On the same day, two other pieces of news are more noteworthy. First, according to multiple media reports, Elon Musk's SpaceX acquired the AI programming tool Cursor for $60 billion, and the transaction was completed "without spending a single cent in cash". Second, the latest developer survey from JetBrains shows that Anthropic's Claude Code has surpassed GitHub Copilot in terms of work usage rate.

The landscape of the AI programming track is completely different from what it was half a year ago.

I. The Leading Player Has Changed

GitHub Copilot was the first product to bring AI into the code writing scenario. When GitHub and OpenAI launched it in 2021, "AI helping you write code" was still a novelty. In the following two or three years, Copilot was almost synonymous with this category, holding the largest developer entry. However, the first-mover advantage failed to turn into a competitive moat. In JetBrains' survey, Claude Code has overtaken it.

Devin was launched in March 2024, and Cognition positioned it as "the world's first AI software engineer". Different from the "completion-based" assistance of Copilot, Devin aims to take over a complete task, write code independently, debug by itself, and deliver the final result. This "autonomous agent" route was regarded as a gimmick by many people back then, but now according to Cognition's disclosure, Devin's monthly active users have exceeded one million, and the number of doubts has decreased significantly.

Cursor took another path. It does not pursue independent work by AI, but embeds AI into the editor to perform completion, refactoring and explanation, keeping humans in the loop all the time. This path is lighter, easier to get started, and brings faster results — Cursor achieved an annual recurring revenue of $2 billion in three years. This performance is probably the reason why SpaceX is willing to acquire it at a $60 billion valuation.

As a result, there are several players at the table of the AI programming track: Devin focuses on autonomy, Cursor focuses on collaboration, Claude Code relies on its own model capabilities, and Copilot holds the entry point. Each player bets on a different route, and no one is guaranteed a sure win.

II. Why Is Code Writing the First Scenario to Be Transformed

Among all AI application scenarios, programming is the first one that enterprises are willing to pay for in real terms. The reason is simple: code can be directly converted into measurable economic benefits.

When enterprises spend money on AI, they focus on ROI, that is, whether it can save costs or reduce manpower. No matter how smart a chatbot is, it is just a thing that "has value but cannot generate measurable returns" in enterprises; but an AI that can write code and fix bugs can save programmers' working hours, and working hours are priced. Saving one engineer can reduce the cost by hundreds of thousands of dollars a year, which is very clear to the finance department. Therefore, while other AI applications are still figuring out how to get users to pay, Cursor has reached $2 billion in ARR, and Devin's ARR disclosed in its last round of financing also reached $492 million.

This also explains why Cognition was founded by three Chinese people. Scott Wu, Steven Wu and Walden Yan all came from the programming competition circle, and Scott Wu once topped the Codeforces leaderboard. They are not entering the AI programming track just because they "are optimistic about this track" from an external perspective, but because they are code writers themselves, who know best how expensive and painful the programming work is. Products born from real pain points are often the most practical.

There is a detail that can well illustrate this point: according to Cognition, Devin has already written 90% of its own code within the company. A company that makes AI programmers uses its own AI programmer to develop its own products. If even the people who built it are not willing to use it this way, the $48 billion valuation will be nothing but empty talk.

Scott Wu has repeatedly emphasized that Devin is not meant to replace developers, but to improve their efficiency. However, the market is obviously far more radical than this statement. The valuations of $48 billion and $60 billion are betting on the same thing: AI programming will become an indispensable infrastructure for enterprises, just like cloud computing and computing power. Marc Andreessen's famous saying "Software is eating the world" — now even the act of "writing software" itself is beginning to be taken over by software.

III. Can the Valuation Be Justified

Despite all the excitement, the accounts still need to be settled clearly.

Devin's ARR disclosed in the last round was $492 million, corresponding to a $48 billion valuation, the multiple is close to 100 times. Traditional SaaS companies are usually valued at 10 to 20 times ARR. What does a 100x multiple mean? It means the product is priced as infrastructure, not as common software. On the other hand, SpaceX spent $60 billion to acquire Cursor, which has an ARR of $2 billion, with an implied multiple of about 30 times, which is more like a deal of "acquiring a company that is already profitable". It needs to be noted that the two companies disclosed their ARR at different time points, so the above comparison is only a rough estimate.

High valuation and high elimination coexist in this track. Copilot, from the category pioneer to being overtaken by Claude Code, is a ready example. Cursor, which was acquired for $60 billion today, may not follow the same path tomorrow.

There is deeper controversy at the technical level. Some people believe that general-purpose Coding Agents are difficult to be really practical, because Code Review is the most difficult part to be replaced by AI. AI can write code, but "understanding the code written by others, judging whether it is correct, and deciding whether it can be deployed online" relies on experience, judgment and sense of responsibility, which are exactly the areas where AI is the weakest. From another perspective, this means that senior programmers will be more valuable — people who can master AI and review AI-generated code are more scarce than people who only know how to write code. Some people call this "the spring for 35-year-old programmers".

The competition between different routes has not yet reached a conclusion. Devin bets on autonomy, which has a high ceiling but uncertain reliability; Cursor bets on collaboration, which has a good user experience but limited imagination space. Which route can succeed determines the final outcome of this industry more than whose valuation is higher. And SpaceX's entry adds another variable: Elon Musk's acquisition of Cursor is not only for code writing, but more likely to connect AI programming tools with physical scenarios such as aerospace and manufacturing.

For Cognition and its three Chinese founders, the $48 billion valuation is a high point, but it is still far from the finish line. The battle in the AI programming track is still ongoing, and no one can hold a stable leading position. The fact that Devin has written 90% of its own code itself illustrates one thing: in this track, even who will be "replaced" may happen in a way no one has expected.

This article is from the WeChat Official Account "Tao Finance", authored by Tao Finance, and authorized for release by 36Kr.