A startup offering products that rival GitHub Copilot and other AI-driven coding assistants has recently reached unicorn status.
On Thursday, Codeium announced the successful closure of a $150 million Series C funding round, led by General Catalyst, which values the company at $1.25 billion post-money. This round also included participation from existing investors, Kleiner Perkins and Greenoaks, bringing Codeium’s total funding to nearly $243 million within just three years of its inception.
Varun Mohan, co-founder and CEO of Codeium, shared that the company still has not utilized any of the $65 million from its Series B funding acquired in January, when it was valued at $500 million. “Our focus remains on leveraging this new capital to enhance our R&D efforts and drive strategic growth,” Mohan stated.
Established in 2021 by Mohan and his MIT classmate, Douglas Chen, Codeium was initially named Exafunction and specialized in GPU optimization and virtualization for AI workloads. However, after recognizing a significant opportunity in generative coding, they rebranded and pivoted in 2022. “Despite the surge of generative AI tools, developers still face challenges with time-consuming coding tasks,” Mohan explained. “Many AI solutions provide generic code snippets that require extensive manual integration into existing codebases. Our AI coding assistant simplifies this process.”
Codeium’s platform leverages generative AI trained on publicly available code, delivering context-sensitive suggestions across an app’s entire codebase. It supports approximately 70 programming languages and seamlessly integrates with popular development environments like Microsoft Visual Studio and JetBrains.
To draw developers away from competitors like Copilot, Codeium has introduced a generous free tier, which has proven effective. Today, the platform boasts over 700,000 users and more than 1,000 enterprise clients, including notable names like Anduril, Zillow, and Dell.
Quentin Clark, Managing Director at General Catalyst, noted that Codeium's client-centric approach to product development has played a key role in securing major contracts.
“The team consistently prioritizes customer needs, enabling them to create tailored solutions deployable across various environments and supporting more languages than competitors,” Clark stated. “Codeium is not merely an idea; it operates as a fully scaling business with large enterprises adopting its offerings throughout their organizations.”
Many businesses are cautious about sharing proprietary code with third-party services—Apple, for instance, reportedly prohibited staff from using Copilot last year due to concerns about data privacy. To mitigate these worries, Codeium offers a self-hosted installation option on top of its standard software-as-a-service plan.
Companies can deploy Codeium’s services on their own servers or opt for a hybrid approach, maintaining control over their data while leveraging Codeium's computational resources. While data transfers to the cloud carry inherent risks, Mohan assures that Codeium uses robust encryption. “We never train our generative autocomplete model on user data, and we ensure data transmission is encrypted,” he emphasized.
Moreover, Codeium has eliminated “non-permissively” licensed code (such as copyright-protected code) from its datasets. Some AI-powered coding tools have been criticized for generating copyrighted code when prompted in specific ways, creating potential legal liabilities. Mohan assures that this concern does not apply to Codeium, thanks to meticulous training data preparation and filtering processes.
“We also remove any residual data resembling non-permissively licensed code to prevent the possibility of uncredited code usage,” he noted. “Additionally, we employ advanced post-generation filtering to track and log any code that closely resembles public coding, irrespective of licensing.”
However, concerns about AI “hallucinations”—the propensity for AI tools to generate incorrect information—remain prevalent. A recent analysis by GitClear found a spike in erroneous code contributions attributed to generative AI tools, while a Purdue University study indicated that over half the programming responses produced by OpenAI’s ChatGPT are inaccurate. Security researchers have raised alarms about the potential for these inaccuracies to compound existing software bugs.
Snyk’s recent survey revealed that 90% of developers are apprehensive about the security risks associated with AI coding platforms. Nonetheless, Mohan contends that Codeium's advanced, context-aware technology delivers more reliable results. “Our context awareness engine grounds suggestions in a user’s existing codebase, resulting in fewer hallucinations and greater adherence to established syntax, semantics, and standards,” he explained.
Whether backed by concrete benchmarks or not, Codeium's message resonates with the right executives, as evidenced by their eight-figure revenue this year. Mohan revealed plans to expand the 80-person, Mountain View-based team to 120 by 2025, aiming for a more significant presence in a competitive market alongside rivals like Tabnine, Anysphere, and Poolside.
Although catching up to Copilot—boasting over 1.8 million paying users—may not be on the immediate horizon for Codeium, the company is well-positioned for growth. Mohan pointed out the increasing adoption of AI coding tools among developers, suggesting that even a modest market share could prove financially rewarding. Polaris Research estimates that the AI coding tools market will reach $27.17 billion by 2032.
Mohan acknowledged the challenge of differentiating amidst the widespread hype in the industry, stating, “Every company must strive to substantiate its claims to resonate with end users. We believe that Codeium, grounded in truth-seeking and realistic AI, will ultimately navigate through the noise.”
Ranking
