Graphon AI came out of stealth on 14 May 2026 with $8.3m in seed funding. Novera Ventures led the round, which The Next Web reported was the first investment from Novera's flagship fund. Arvind Gupta, Novera's founder and managing director, led it.
Other participants were Perplexity Fund, Samsung Next, Hitachi Ventures, GS Futures, Gaia Ventures, B37 Ventures and Aurum Partners, which the company's announcement described as the investment fund affiliated with the owners of the San Francisco 49ers, along with other strategic limited partners. No valuation was disclosed.
“But the world isn’t made of tokens, it’s made of relationships.”Arbaaz Khan, Founder and CEO, Graphon AI
What Graphon does
Graphon describes its product as a pre-model intelligence layer. Large language models can read only a limited amount of text at once, roughly one million tokens for the most advanced ones, according to the company, while organisations hold far more across documents, video, audio, images, logs and databases.
Retrieval-augmented generation, the usual workaround, pulls relevant pieces of content for a model but struggles to find connections between records that were never stored together. Graphon says its system sits before the model, ingests multimodal data and works out how the pieces relate, then saves the result as what it calls persistent relational memory. Any foundation model or agent framework can then query that memory.
SiliconANGLE reported that the platform reportedly uses small models of about 200 million parameters to find patterns, working with graphs. It also uses graphon functions, the mathematical objects the company is named after. A graphon is the limit of a sequence of dense graphs. The Next Web noted that the company has offered no independent benchmarks and few customer case studies, which makes it hard to judge how mature the technology is.
Team and customers
Arbaaz Khan is chief executive, Deepak Mishra is chief operating officer and Clark Zhang is chief technology officer. The company says its team includes former researchers and engineers from Amazon, Meta, MIT, Rivian, Google, Apple, NVIDIA, Samsung AI Center and NASA.
Its technical advisors are Jennifer Chayes, dean of the College of Computing, Data Science, and Society at UC Berkeley, and Christian Borgs, a Berkeley computer science professor who helped formalise the graphon concept.
The one early customer named is GS Group, a South Korean conglomerate that is also an investor through GS Futures. Ally Kim, a vice president at GS, said the company has used Graphon to analyse customer movement in convenience stores and to improve safety through CCTV analysis at construction sites. Graphon says its platform has also been used for enterprise content management, industrial intelligence, agentic workflows and on-device applications on phones, cameras and wearables.
Why the investors backed it
Gupta said Graphon changes where the intelligence happens. In his view, most companies are trying to build ever-larger models, while Graphon improves the layer between raw enterprise data and the model, which makes existing models more capable without making them bigger.
The company's chief executive framed the pitch differently. Khan said an LLM with Graphon is better than an LLM alone, and that the company is not replacing models but amplifying them.
A busy corner of AI infrastructure
SiliconANGLE placed Graphon among several venture-backed startups trying to increase how much data language models can take in. It cited Subquadratic, which raised $29m the week before with a transformer variant that can process up to 14 million tokens per prompt, and Standard Intelligence, which uses a masked compression objective to remove unimportant data from prompts.
Graphon's funding total beyond this seed was not reported, and the sources did not say how the seed money will be spent.
Company website: graphon.ai