Today, California-based Codeium, an AI startup leveraging proprietary large language models (LLMs) to aid with software development, announced a $65 million series B funding round at a valuation of $500 million. The investment has been led by VC firm Kleiner Perkins and with participation from Greenoaks as well as General Catalyst.

Codeium said it will use the capital to grow its team across functions and expand its generative AI-powered coding toolkit that already writes over 44 percent of newly committed code for over 300,000 developers.

The backing for the company comes at a time when a large chunk of enterprises and startups are looking at specialized language models to accelerate their software development projects.

Many known players, including GitHub and Amazon, have already ventured into the space, which is expected to become a $106 million opportunity by 2030 — growing at a CAGR of over 25%.

What makes Codeium unique?

Software development has historically been manual, with developers facing the challenge of writing code and being stuck in long development cycles.

However, when generative AI came to the scene with the launch of ChatGPT and Codex, things got fast-tracked and techies started working with smart assistants to automate and enhance various facets of coding, right from writing code snippets to enhancing the actual project code and translating languages.

The move marked a major shift in software development, but even with AI, the code snippets generated are far from what’s needed. They are more like boilerplate samples which can be adopted into a working solution for the targeted use case.

Plus, most AI coding offerings do not integrate with the environments where developers work, focus on just one modality at a time and send code to the model over the internet, which can risk security.

MIT grads Varun Mohan and Douglas Chen saw these challenges first-hand and decided to address them with the launch of Codeium, a security-focused LLM toolkit that provides intelligent code suggestions in the context of the codebase. This gives developers personalized generations to accelerate every task they do in their development workflow.

“We leverage proprietary AI models to maximize developer productivity using state-of-the-art integration with your codebase. Codeium supports over 70 languages and works in over 40 Integrated Developer Environments (IDEs), including Visual Studio Code, the JetBrains suite, Eclipse and Jupyter Notebooks,” Codeium wrote in a blog post announcing the series B funding.

The AI toolkit, which can be self-hosted or deployed as SOC2 Type 2-compliant SaaS, integrates with existing Source Code Management (SCM) systems for full repository context awareness. It includes in-editor capabilities, such as autocomplete and chat, and is writing over 44% of newly committed code for customers, the company said in a press release.

However, developing LLMs and using them to power the toolkit is not the only work here. Codeium also claims to own the hardware layer, with the infrastructure for large-scale AI workloads being in development for years now. This has helped the company keep the costs of taking the models to users at a fraction of the cost of other players in the industry.

“With this differentiated background, we’ve been able to scale from zero to over 300,000 users in less than fifteen months, while simultaneously increasing the size and quality of the proprietary large language models (LLMs) that power our product, building intuitive and industry-leading user experiences and personalizing suggestions to each developers’ tasks,” the company added in the blog post. 

Plan to cover entire software development lifecycle

Over the last year, Codeium has offered its AI toolkit to enhance internal developer productivity for “hundreds of companies,” including Atlassian, Anduril, Clearwater Analytics and several Fortune 500 giants. The company offers a base, individual-centric version of the toolkit for free, while the business-centric offering starts at $12/user per month. 

In all the cases, the company emphasizes that the customer code is neither saved nor used for training the public system.

Now, with the series B round, the company plans to grow its engineering and sales teams as well as expand the platform to support the large and varied needs of developers.

The goal is to go beyond writing and running code and address tasks in every stage of the software development lifecycle, including system design, code maintenance (especially legacy code migrations) and fixing complex issues such as security vulnerabilities. 

“Today, we believe Codeium makes developers 20% more productive. While impressive, we believe there is so much more that can be done. We envision Codeium as not just a coding assistant, but a productivity multiplier that will make developers 20x more productive rather than just 20%,” the company noted.

Other notable players working in this category are Replit, Github Copilot, Sourcegraph, Codiga and Amazon Sagemaker and Google’s AlphaCode.