Welcome to the podcast.
I'm your host, Jaden Schaefer.
Today on the podcast, we are talking about Cursor, which is a coding assistant, and it has just completed an absolutely astronomical amount of money raised.
$2.3 billion.
What's crazy is this is coming five months after its previous round.
It just continues to...
I mean, just absolutely absorb venture capital.
Its valuation continues to rise.
They've actually just hit a 293 billion valuation with this new round of funding that they've done, which was reported by the Wall Street Journal.
And what's insane is this actually more than doubles the company's previous valuation, which was 99 billion.
It did this when it raised $900 million in a Series C in June.
When that happened, everyone was pretty shocked and, you know, blown away that they were able to raise almost a billion dollars.
And now, just five months later, they're at $2.3 billion raised.
This most recent fundraise was led by, well, it was actually co-led.
There's two leads on it.
Asell was one of them, who is an existing investor.
They put money in.
And then there was Coat, which is new on the cap table.
They have some strategic investors, including Nvidia, of course, and Google, who also joined in the round.
No shocker here.
Nvidia is basically handing out money to the top AI use cases, as they know the money is going to come straight back to them.
As more compute will be needed.
So Joshua Kushner, who's from Thrive Capital, led the company's two prior rounds that they did.
And he was also participating in this round as well.
The co-founder and CEO is Michael Truel.
He was talking to the Wall Street Journal recently.
And he told them that all of the money they've just raised from this round is going to be put towards developing Composer, which is basically their new AI model.
This was released back in October.
And so now they are trying to improve this.
What's impressive here is so often, with these AI companies, They get into this tricky situation where they're reliant on either OpenAI or Anthropic and they use that for their foundational code model.
They're kind of wrapped around it.
And if there's any sort of or issues with those foundational code models, it can be really tricky.
In addition, they have to pay a ton of their revenue to those code models to help.
So the fact that they've actually developed their own AI model is obviously why they're able to raise so much money.
It's the number one coding tool Cursor is.
And so, because there's just so many users, they're paying so much money to these third-party AI vendors by making their own tool.
This makes a lot of sense.
Next year could be a really interesting one for Cursor.
The company is growing very quickly.
OpenAI and Anthropic are both really getting into coding products.
OpenAI has their own tool and Cloud Code from Anthropic is obviously, you know, one of the preferred tools by all developers.
So this is getting really competitive.
But that's not to say Cursor won't be able to keep up up until this point.
They have one of the largest user bases of uh, coding developers of any other company.
In the most recent numbers I could find, cursor had over 1 million daily users and they have tens of thousands of enterprises.
They have opening eye Instacart, Salesforce all of them are using um, which is interesting because OpenAI has their own coding tool, but a lot of their employees are still using Cursor.
The user base includes a whole bunch of different individuals and teams.
They have subscription tiers for pro and business plans.
So they got a bunch of different options.
One thing that I do find very interesting with Cursor is the fact that their coding tool is actually built on top of Microsoft's open source VS Code editor.
So they have essentially kind of grabbed that open source project and built their own tool.
One of the LLMs embedded in their editor, like I mentioned, is called Composer.
This just came out in October.
And according to them, it's a mixture of experts algorithm, right?
So that basically means that when there is a question, it sends it to a number of quote unquote experts inside of the AI model that each determine which can answer the question best.
They also oftentimes will send the question to multiple sort of like fine-tuned models inside and then they'll collaborate and see who came up with what answer and they'll like come up with an answer.
That's kind of a consolidation of all of those.
This all happens in the background very quickly, so you're not going to notice it.
You just ask a question.
It will give you a response.
But that's how it's working.
It's an algorithm.
It runs four times faster than llms with comparable output quality.
According to them,
So it can complete a lot of coding tasks in under 30 seconds.
So the reason why I think a lot of people are getting excited and they're able to continue to raise so much money is because yes, you could go use something like you know Cloud Code.
Yes, you could go use OpenAI's product.
But by using them if you actually can get stuff done faster.
A lot of people love that, because I know for us.
We use cloud code at AI box which, by the way, if you don't have it, go check out AI box.
Dot AI, you get access to all of the top models in one place for 20 bucks a month.
But we use cloud code a lot to help us.
It does take a very long time when we launch it on a task or a new project.
It can sit there for 10 or 15 minutes working on the code base.
So if you could bump that up four times faster, it does make a really big difference.
We know, of course, like as a refresher, an LLM is comprised of a bunch of different kernels.
So these are parallelized snippets of code.
They are all running across a whole bunch of graphics card cores at once.
And developers.
They're usually going to be writing kernels with the help of what's called a CUDA library.
It's, you know...
It basically abstracts away some of the complexity that is associated with this.
They have these kind of prepackaged code that removes the need to write everything from scratch.
What's interesting here is Cursor says that they did not use any CUDA libraries while they were building Composer.
And they said that they implemented the model's kernels using PXT.
This is kind of what's called a low-level machine language.
NVIDIA chips are using this a lot and that approach apparently has helped Cursor achieve more than 3x performance increase across a bunch of their components.
So they've kind of built this own thing in-house.
You can see, because they're using Nvidia chips to run this.
You can see why Nvidia would get in on this most recent round and give them a lot of funding for this.
Nvidia is definitely licking their chops and sees this as an amazing opportunity to continue their lead and dominance in the chip space.
So it's going to be, or I guess in the GPU space.
So it's going to be interesting to see how that plays out.
In any case, thank you so much for tuning into the podcast today.
If you learned anything new, it would mean the world to me.
If you could leave a rating or review on the podcast.
A ton is happening and I will try my best to keep you up to date on everything happening in the AI news.
Hope you guys all have a great rest of your day and I'll catch you in the next episode.
