Last week, Cursor (via its parent company Anysphere) closed a massive $2.3 billion funding round at a $29.3 billion valuation.
The round focused on Composer, their own coding-specialized model, stating they have the data and user feedback to build the best model for coding.

That caught my attention, given that most people currently use Anthropic models within Cursor. Indeed, if you look at coding agents’ rankings, like Gosu Evals, Cursor is only #10 in quality, while using Claude Sonnet 4.5. What’s more interesting is that seven out of the top 10 coding agents use Claude Sonnet 4.5, so Anthropic objectively has access to far more data than Cursor when talking about training the best model for coding. It not only has access to most of Cursor's users inference, but also to most of the rest of the coding agents.

Cursor’s team must know that well, and while the rhetoric of building the best model for coding is powerful to fundraise, the movement seems more focused on reducing business risks. Until now, Cursor has had a strong dependency on the big model providers. And a few months ago, Anthropic already limited Claude's access to Windsurf, the biggest Cursor’s competitor.

Composer looks less like actually building the best possible coding model, rather than a strategy to own the whole stack and reduce risks, which is still a smart move, and Composer likely won’t also be a bad model.

Cursor publicly stated that they are losing money with every user. If you ever wondered why your $20 Cursor subscription buys more Claude tokens than if you directly use Anthropic APIs, it’s because Cursor pays for it. This was the main reason behind the pricing changes we saw in July.
Having their own model not only gives them autonomy, but it also allows them to increase their margins, as they won’t have to spend all the revenue on model providers like Anthropic.

Something that personally worries me about Cursor is the fact that, unlike previous born big players, they didn’t have something unique. Google was born out of figuring out page rank, Facebook won by network effects, but Cursor, without its own model, is easily replicable. They have built great distribution thanks to the first-mover advantage; however, they sell to developers, who often change tools and test new ones frequently.
That is probably the reason why they are focusing now on enterprises. These will sign (multi-)year contracts and won’t change the tool for something just slightly better. Procurement processes are painful to go through, and Cursor is positioning well to avoid being replaced by better coding agents in those enterprises.

Owning their own model means they are shifting from just being a tool to an infrastructure provider. This is important because they are building the basis to become an AI-powered dev shop.

Some steps already point in that direction. With the background agents and some new integrations with Linear, Slack, and other project management tools, we can start to see Cursor as a third-party dev agency that we can throw tasks to when needed. Still, the small tasks, but as models advance, we will have them completing more complex ones. We will shift to design and review, while “outsourcing” the development to AI coding agents.

That is the logical next step. Once we can trust agents to produce the right code with a certain consistency, we won’t need them embedded in the editor anymore, since we won’t have to edit the code they produce very frequently.

However, there is a final step where becoming the code factory could have zero value. It’s usually called “Liquid apps”, the AI generating pixels, acoustics and actuator commands, as described by Elon Musk.
I believe there will be an intermediate step before that one, where being the code factory will be extremely valuable. We could call it “Liquid code”, the AI generating and running the code just in time.

The day after Cursor’s announcement, Elon Musk announced a round for xAI focused on building new infrastructure for Grok.

It’s interesting to see that announcement just the day after Cursor, given Elon’s perspective on liquid apps.

But what is more interesting is the fact that both companies - Cursor and xAI - raised massive rounds to build infrastructure. Infrastructure in the AI world is translated into data centers and GPUs. Will Nvidia, after all, be the big winner of both rounds?