Graphon AI, a startup building the layer of software that sits between an organization's raw data and its AI systems, has raised $8.3 million in seed funding. The company came out of stealth this week with the goal of solving a problem that affects nearly every business trying to use AI at scale: models can only process a small amount of data at a time, which limits how well they can reason across large, complex information environments. Arbaaz Khan founded the company and serves as CEO, with Deepak Mishra as COO and Clark Zhang as CTO. The broader team includes researchers and engineers who previously worked at Amazon, Meta, MIT, Google, Apple, NVIDIA, Samsung AI Center, Rivian, and NASA.
The seed round was led by Arvind Gupta of Novera Ventures, for whom Graphon is the first investment from his flagship fund. Other participants include Perplexity Fund, Samsung Next, GS Futures, Hitachi Ventures, Gaia Ventures, B37 Ventures, and Aurum Partners, an investment vehicle connected to the ownership group of the San Francisco 49ers.
To understand what Graphon is trying to do, it helps to understand the problem it is addressing. AI language models work with what is called a context window, which is essentially the amount of information a model can read and process at one time. Even as context windows have expanded to around one million tokens, that capacity falls far short of what a large organization actually holds. Most enterprises store documents, video footage, logs, and databases that together span trillions of tokens.
The standard workaround is a technique called Retrieval-Augmented Generation, or RAG. It works by pulling only the most relevant pieces of information into the model's context window when a query is made. Useful, but limited. RAG can surface individual snippets of data, but it cannot show how different pieces of information connect to each other. Research published in 2025 found that the effective performance of AI models can fall far below their advertised limits, by up to 99 percent on complex tasks, due to technical constraints around memory and attention. For a business trying to run AI across surveillance footage, compliance records, and customer databases at the same time, that gap is a real operational problem.
The deeper issue is structural. As one analysis found, current RAG systems tend to treat knowledge infrastructure as separate from security, governance, and observability, which makes them difficult to scale reliably in enterprise environments.
"Graphon is improving the layer between raw enterprise data and the model itself. That gives today's foundation models a much better understanding of complex data — and makes them far more capable without needing to be bigger." — Arvind Gupta, Founder and Managing Director, Novera Ventures
Graphon's approach is to process data before any AI model ever reads it. Using what the company calls graphon functions, its system automatically maps how information connects across different types of data, including video, audio, documents, images, and structured databases. The idea is to give AI models a structured picture of relationships rather than a pile of isolated content to sift through.
The name and the method come from the same mathematical idea. Graphons are a mathematical framework for representing relationships at scale, co-developed by Christian Borgs, a computer science professor at UC Berkeley, together with colleagues including Jennifer Chayes. Both Borgs and Chayes serve as technical advisors to the company. Chayes is also the Dean of the College of Computing, Data Science, and Society at UC Berkeley.
"AI has spent the last decade learning to mimic language. But the world isn't made of tokens, it's made of relationships. By preserving that structure, we make foundation models more accurate and more useful at enterprise scale. An LLM with Graphon is better than an LLM alone. We're not replacing models — we're amplifying them." — Arbaaz Khan, Founder and CEO, Graphon AI
Graphon works with any foundation model or agent framework, so businesses are not required to switch providers to use it. Stated use cases include enterprise knowledge management, industrial process monitoring, automated decision workflows, and on-device reasoning across phones, cameras, and wearables.
Graphon has not disclosed a detailed breakdown of how it plans to allocate the $8.3 million. The company is expected to direct the capital toward product development, hiring, and growing its customer base, with open roles already listed on its careers page.
One early enterprise customer is GS Group, one of South Korea's largest conglomerates. GS has been using Graphon to analyze customer movement in its convenience stores and to improve safety monitoring at construction sites through CCTV systems. Both applications involve processing multiple data types at the same time, which is what Graphon is specifically built to do.
"Graphon has been an invaluable partner in GS Group's AI transformation journey, bringing exceptional passion and AI expertise to the table. Their multimodal AI solutions have been instrumental in solving real-world challenges, such as analyzing customer movement in convenience stores and enhancing safety through CCTV analysis at construction sites." — Ally Kim, Vice President, GS Group
The investor group reflects a mix of corporate venture capital and independent funds. Samsung Next and GS Futures each represent major South Korean industrial and technology groups, which gives the round a practical dimension beyond the money itself. Hitachi Ventures brings a similar industrial angle. The Perplexity Fund, affiliated with the AI search company, adds a connection to the AI infrastructure market Graphon is entering.
Novera Ventures led the round. Its founder, Arvind Gupta, chose Graphon as the first company backed by his flagship fund. Aurum Partners, tied to the San Francisco 49ers ownership group, is also among the participants.
