Participated · Seed · Jul 2026

Nums AI develops tabular foundation models that analyze and predict missing values in structured data (rows and columns) in a single inference step. Its TFM approach is designed to handle multiple tasks—such as demand forecasting, anomaly detection, and credit evaluation—without building separate task-specific models. The company plans to commercialize via API offerings and on-premise deployments for customers with data-export constraints, targeting sectors like manufacturing, finance, retail and commerce, and healthcare. Nums AI intends to use its recent seed funding to expand GPU infrastructure and grow its research team to accelerate model development. The company was founded by CEO Jae-min Yoo, an academic with a Google PhD Fellowship and assistant professorship background, alongside co-founders Doo-ho Lee and Min-yong Cho, who bring research and engineering experience in structured-data AI.

Led · Series B · Jun 2026

INNOVAID’s Dentlink platform digitizes the end-to-end dental prosthetics workflow, integrating data acquisition, dentist-technician communication, fabrication, quality control, and logistics in a single system. The company says digitization has reduced prosthetic remake rates from around 10% to roughly 1%. Dentlink is live in six countries: the United States, Australia, Canada, New Zealand, Singapore, and the United Kingdom. INNOVAID reports that both clinic count and revenue have grown more than 2.5 times year over year and that its paid customer retention rate in the U.S. is 95%. The company has a strategic partnership with implant manufacturer Straumann to support global expansion. Using Series B proceeds, INNOVAID plans to scale operations in the U.S. and Australia and to build an automated, robot-integrated manufacturing center in Korea.

Participated · Seed · Jun 2026

a2sys is developing a software platform aimed at optimising memory utilisation and data movement for AI infrastructure. The company was co-founded by former Samsung and Naver executive Lee Dong-soo and KAIST professor Yu Min-soo and was founded shortly before its seed round. It positions itself as a connector between memory manufacturers (including Samsung and SK Hynix) and AI service providers rather than as a chip maker. The founders argue that AI's evolution toward agentic systems is shifting bottlenecks from GPU computation to memory and data transfer efficiency. a2sys announced a KRW 16 billion (about $11M) seed raise to support development of its platform.

Participated · Seed · Mar 2026

Advanced Machine Intelligence (AMI) is an artificial intelligence research company founded by Yann LeCun that focuses on building new self-supervised learning systems known as Joint Embedding Predictive Architectures (JEPA). Unlike traditional large language models that predict the next word, JEPA is designed to understand and reason about the physical world through multimodal sensor data, enabling planning and decision-making. The company plans to translate this core technology into products for industrial process control, automation, wearable devices, robotics, and healthcare, where reliability and safety are paramount. AMI maintains its headquarters in Paris with additional offices in Singapore, New York City, and Montreal. Although still pre-revenue, the firm’s substantial seed financing positions it to invest heavily in research talent and compute infrastructure. The $3.50 billion pre-money valuation reflects investor confidence in the commercial potential of its foundational AI approach.

Participated · Series A · Mar 2026

Founded in 2024 by serial networking entrepreneur Drew Perkins and chip designer Omar Hassen, Eridu is redesigning data-center networking from the silicon up to meet the bandwidth demands of modern AI workloads. The company is building a switch that integrates far more networking functionality directly onto the chip, reducing optical links, lowering latency, and cutting power and cost compared with traditional multi-tiered architectures. Its eventual product line will replace conventional gear from vendors like Arista Networks in AI data centers, positioning Eridu at the heart of the ongoing build-out of GPU clusters that now span millions of processors. Perkins argues that while GPU compute and memory bandwidth are improving roughly 10× per year, incumbent switch technology advances only 2–3× every few years, creating a performance gap Eridu aims to close. The startup has assembled roughly 100 employees and leverages the founders’ deep experience, including Perkins’ past exits such as Lightera, Infinera, and Gainspeed. Although the company has not disclosed revenue metrics or valuation, it has raised substantial capital to fund product development and scale.