Graphon AI Exits Stealth With $8.3M to Fix AI’s Core Memory Problem

SAN FRANCISCO, CA – May 14, 2026 – A new AI infrastructure company, Graphon AI, emerged from stealth today, announcing $8.3 million in seed funding to tackle one of the most fundamental limitations of modern artificial intelligence: its inability to reason over vast, interconnected data. Founded by a team of former researchers and engineers from Amazon, Meta, and Google, the company is introducing a new architectural component called a “pre-model intelligence layer” designed to make foundation models more accurate and capable by understanding the relationships within data before it is ever processed by an AI.

The round was led by Arvind Gupta of Novera Ventures, with significant participation from Perplexity Fund, Samsung Next, GS Futures, Hitachi Ventures, and other strategic partners. The investment signals a growing appetite for solutions that address the core operational bottlenecks of AI, rather than simply building larger, more resource-intensive models.

Today’s most advanced large language models (LLMs) are constrained by their “context window”—the finite amount of data they can consider at any one time. While these windows have expanded, they remain a drop in the ocean compared to the trillions of tokens of information held by large enterprises in documents, videos, logs, and databases. This limitation means AI can answer questions about individual files but struggles to discover hidden patterns or reason about how disparate pieces of information connect. Graphon AI proposes to solve this by creating a structured, relational memory that any AI model can access, effectively providing unlimited and persistent context.

Beyond the Context Window: A New Layer for AI

Graphon AI’s core innovation is its pre-model intelligence layer, a system that operates on raw data before it reaches a foundation model. Instead of feeding an LLM a sequence of isolated tokens, the technology first discovers and maps the relationships between data points across unlimited multimodal sources, including video, audio, images, documents, and structured databases.

This process is powered by graphons, a mathematical concept for which the company is named. The term was coined by UC Berkeley professor Christian Borgs, who, along with fellow graphon pioneer and Berkeley Dean Jennifer Chayes, serves as a technical advisor to the company. Graphons provide a mathematical framework for understanding and analyzing large-scale networks, allowing the system to treat the structure of data as a first-class citizen. “Graphon’s technology automatically discovers relationships across data and treats that structure as a first-class citizen,” Chayes and Borgs stated jointly. “We’ve known that the future of AI would depend on understanding structure beyond tokens.”

This approach differs significantly from existing solutions like Retrieval-Augmented Generation (RAG), which retrieve relevant data chunks to augment a model's knowledge but often fail to capture the complex, cross-document relationships. Graphon AI’s system builds a persistent relational memory, enabling models to reason over a connected system rather than isolated token sequences. The goal is to dramatically improve accuracy and reduce the “hallucinations” that plague LLMs when they lack sufficient context.

Arvind Gupta of Novera Ventures, who made Graphon the first investment from his firm's new flagship fund, emphasized the paradigm shift. “Graphon changes where the intelligence happens,” Gupta said. “Most companies are trying to build ever-larger models. 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.”

The 'AI Dream Team' Betting on Relational Data

Graphon AI’s ambitious vision is backed by a leadership team with deep experience building AI systems at the world’s top tech companies and research institutions. Founder and CEO Arbaaz Khan, who holds a PhD in Robotics, previously worked as a Senior Applied Scientist at Amazon and an Autonomous Driving Software Engineer at Rivian, with stints at Google, Apple, and NVIDIA. His co-founders include Deepak Mishra, former Principal at Hitachi Ventures, and Clark Zhang, who serves as CTO.

The team’s philosophy is not to compete with foundation model developers like OpenAI or Anthropic, but to enhance their products. “AI has spent the last decade learning to mimic language,” Khan explained. “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.”

This vision is further supported by the company’s technical advisors, Drs. Chayes and Borgs. Their decades of foundational research into graph theory and its application to machine learning lend significant academic weight to the company’s commercial strategy. Their work at Microsoft Research, where they co-founded multiple interdisciplinary labs, and now at UC Berkeley, places them at the center of the AI universe, providing Graphon AI with a direct line to the cutting edge of theoretical and applied computer science.

From Factory Floors to Smartphones: Real-World Impact

While the technology is advanced, Graphon AI is already demonstrating its practical value with early customers and developers. The platform is being applied across a range of high-impact use cases that push the boundaries of current AI capabilities.

In industrial settings, the technology is used for process intelligence, analyzing video feeds from factory floors and construction sites alongside enterprise system data to spot compliance issues, identify process gaps, and perform root-cause analysis. For enterprise content management, it enables reasoning across vast, siloed archives of video, audio, images, and documents simultaneously—a task that is nearly impossible with conventional tools.

One of its early customers, South Korean conglomerate GS Group, is already seeing results. “Graphon has been an invaluable partner in GS Group's AI transformation journey,” said Ally Kim, Vice President at GS. Kim noted that Graphon's multimodal AI solutions are solving real-world challenges, such as “analyzing customer movement in convenience stores and enhancing safety through CCTV analysis at construction sites.”

The company is also targeting the rapidly growing field of agentic workflows, where AI agents can use the rich, multimodal context provided by Graphon to automate complex decisions. Furthermore, the technology is designed to run on-device, enabling phones, cameras, and wearables to understand and reason about the data they generate in real time, a key feature that aligns with the strategic interests of investors like Samsung Next, which is heavily focused on embedding powerful AI into its consumer electronics portfolio.

By focusing on the foundational layer of data relationships, Graphon AI is making a strategic bet that the future of artificial intelligence lies not just in building bigger models, but in providing them with a fundamentally better understanding of the world they are meant to analyze.