PearX has carved out a niche as one of the most closely watched accelerator programs in the venture capital world, despite the crowded landscape of startup incubators. Run by Pear VC, a firm that concentrates on pre‑seed and seed investments, the program accepts only twenty founders per cohort and runs for twelve intensive weeks.

What sets PearX apart from the better‑known Y Combinator is its deliberately small size, the absence of standard term sheets, and the flexibility to write checks as large as two million dollars. Unlike YC, where some of the hottest companies announce funding before the program ends, PearX keeps its participants under wraps until demo day, creating a moment of heightened anticipation for investors.

The most recent demo day, held in San Francisco last week, drew a roster of top‑tier venture capitalists eager to see which of the sixteen companies in the batch would capture their interest. After the presentations, TechCrunch lingered to ask several VCs which startups had generated the most buzz.

Speridlabs – Spatial foundational models for robotics, gaming, and special effects. The team argues that existing world‑model players such as Runway, Odyssey and Google’s Genie cannot be queried or modified once a scene is built. Their answer is Mundus, a tool they describe as a “3D Midjourney” because it preserves geometric persistence when a part of the model is altered, enabling iterative editing without losing the underlying structure.

Saia – A fast, cost‑efficient chip that runs AI inference directly on device. Saia claims its architecture bypasses the traditional memory bottleneck by executing models out of flash storage, delivering speeds that vastly exceed those of Nvidia’s Jetson platform while offering eight times the memory capacity and consuming only a quarter of the power. The startup says it is already in talks with Samsung about memory integration, plans to fabricate test chips next year, and targets mass production by 2028. Founder Ayaan Govil, who is twenty years old, convinced Pear VC co‑founder Mar Hershenson—a semiconductor engineer with a PhD in circuit design—to back his vision.

Ren – A more secure AI personal assistant. Ren aims to match the capabilities of Muse and Instinct but places privacy at the forefront. It processes data on‑device whenever possible, falls back to a secure private cloud when needed, and checks every action against user‑defined guardrails. Importantly, Ren replaces the human‑operator call‑flow used by some competitors with a fully automated voice system, eliminating any human involvement in the loop to protect user confidentiality.

Veros – AI native trust and estate planner. Veros seeks to replace the costly, time‑intensive work traditionally performed by attorneys, wealth managers and trust administrators. Its AI recommends optimal trust structures and manages assets over their lifetime. The company already oversees roughly 250 million dollars in assets under management and is pursuing a trust charter that would allow it to operate as a regulated trust entity.

Datum – AI engineer for industrial design. Datum accelerates physical product development by helping engineers rediscover and reuse past designs. It indexes a company’s library of 3D models and applies a proprietary Geometric Fingerprint technique that matches parts based on shape alone. This approach streamlines engineering workflows across large‑scale manufacturing and product design markets.