A startup with a product that competes with GitHub Copilot and other AI-powered coding assistants has achieved unicorn status.
Codeium announced Thursday that it had closed a $150 million Series C funding round led by General Catalyst, valuing the company at $1.25 billion. Existing investors Kleiner Perkins and Greenoaks Capital also participated in the round, bringing the company's total funding to nearly $250 million just three years after founding.
Codeium co-founder and CEO Varun Mohan told TechCrunch that the company has yet to tap the $65 million Series B tranche it raised in January, when the company was valued at $500 million just eight months ago.
“While we have barely leveraged our existing funds, we believe this capital injection will enable us to significantly accelerate research and development and growth and take even bigger strategic bets,” he said.
Codeium was founded in 2021 by Mohan and childhood friend and MIT alumnus Douglas Chen. Prior to Codeium, Chen was at Meta, where he worked on building software tools for VR headsets such as Oculus Quest, and Mohan was a technical lead at autonomous delivery startup Nuro, where he managed the autonomous infrastructure team.
The startup began as an entirely different company called Exafunction, focused on GPU optimization and virtualization for AI workloads, but in 2022, Mohan and Chen saw a huge opportunity in generative coding and decided to rebrand and pivot.
“Despite the proliferation of generative AI tools, developers still struggle with time-consuming coding tasks,” says Mohan. “Many AI-driven solutions offer generic code snippets that require significant manual effort to integrate and secure into the existing codebase. That's where our AI coding assistance comes in.”
Codeium's platform is powered by generative AI models trained on public code to provide suggestions in the context of an app's entire codebase, supports around 70 programming languages, and is integrated with many popular development environments, including Microsoft Visual Studio and JetBrains.
Codeium's AI-generated coding suggestions. Image credit: Codeium
To lure developers away from Copilot and other rivals, Codeium initially released a generous free plan, a strategy that appears to have worked: The startup now has more than 700,000 users and over 1,000 enterprise customers, including Anduril, Zillow, Dell, and AthenaHealth.
General Catalyst managing director Quentin Clark suggested that Codeium has won some of the biggest contracts by taking a consistently customer-centric approach to product research.
“The team's approach has always been to follow the customer, guiding the company to build solutions that fit their needs – deployable in any environment and supporting more languages than anyone else,” Clark said in a statement. “What Codeium has built is not just a demo, announcement, or idea – this is a fully scalable business, with large enterprises adopting the product across their entire organizations.”
Companies are often wary of exposing proprietary code to third parties — Apple, for example, reportedly banned its employees from using Copilot last year over concerns about sensitive data being leaked — and to ease those concerns, Codeium has begun offering a self-hosted installation option in addition to its standard SaaS plan.
Image credit: Codeium
Businesses can now deploy Codeium's services on their own hardware if they wish, or they can adopt a hybrid setup, keeping their data on their own devices but using Codeium's servers for their computing needs.
While data transfer to the cloud always carries some risk, Mohan asserted that Codeium leverages strong encryption. “We never train our own generative autocomplete models with user data, we never sell data, and we ensure that all data transfers are encrypted,” he added.
Codeium has also taken steps to remove code with “non-permissive” licenses (such as copyrighted code) from the datasets it uses to train its AI models. Some code generators trained with restrictively licensed or copyrighted code have been shown to spit out that code when prompted in certain ways, creating liability risks (developers who incorporate that code could be sued). Mohan said that this isn't the case with Codeium, thanks to its approach to training data preparation and filtering.
“We also remove any remaining data that resembles code under explicit non-permissive licenses, in case others have copied the code without providing proper attribution and licensing,” he added. “Furthermore, we have state-of-the-art post-generation attribution filtering and logging in case these large-scale probabilistic models generate code that resembles public code, whether under a permissive or non-permissive license.”
But what about hallucinations? Most AI coding tools are notorious for making things up, which can be extremely disruptive in an enterprise environment.
An analysis by developer tools startup GitClear found that generative AI tools have pushed more erroneous code into codebases over the past few years, and a Purdue study found that more than half of the answers OpenAI's ChatGPT returns to programming questions are incorrect. Security researchers have warned that such tools can amplify existing bugs in software.
A recent survey by cybersecurity firm Synk found that nine in 10 developers are concerned about the security implications of using AI coding platforms, but Mohan argued that Codeium's superior, deep, context-rich technology produces more reliable results than most others.
“Our context-aware engine is able to build results based on what's already in the user's codebase, resulting in fewer misunderstandings and suggestions that better adhere to existing syntax, semantics and standards,” he said.
Whether or not the benchmarks bear that out, Codeium's pitch seems to be resonating with the right executives. Revenue hit eight figures this year, and Mohan said the San Jose-based, 80-person startup plans to grow to 120 employees by 2025 as it seeks to make a bigger impact in a market crowded with strong competitors like Tabnine, Anysphere and Poolside.
Codeium probably can't catch up with Copilot, which had more than 1.3 million paying users as of April, at least not anytime soon. But it doesn't have to be: As Mohan correctly points out, given the widespread adoption of AI coding tools among developers (despite concerns), it's safe to say they could make a sliver of money in this emerging field.
Polaris Research predicts that the AI code tools market will reach a value of $27.17 billion by 2032.
“Hype is a challenge facing the industry,” Mohan says, “which makes it difficult for all companies to truly convince end users that they're at the cutting edge of what's possible. But we believe that truth-seeking, pragmatic AI companies like Codeium will ultimately cut through the noise.”