frontier-models
(8)
Vijay Krishnan Has Spent 20 Years Teaching Machines What Matters
Before AI became the loudest trade in technology, Vijay Krishnan was quietly following the same question from search to recommendations to remote work: how do you turn human judgment into something a machine can use?
Maingen
Maingen is a Y Combinator (S26) startup building reinforcement learning environments for industrial operations - the roughly $5 trillion slice of the US economy that runs factories, power plants, and solar farms. Its first product, SolarBench, tests whether frontier AI agents can act as the on-call engineer for a portfolio of solar farms, handling alerts, telemetry, work orders, and priced field actions. The company sells the environments and datasets that let AI labs train and evaluate models on messy, long-horizon physical-world work.
Foundation Models: The Industry Report
A deeply researched report on the foundation model industry as of mid-2026: a sector where two labs, OpenAI and Anthropic, capture roughly 89% of startup AI revenue; where hyperscaler capital spending has tripled to an estimated $725 billion a year; where inference has overtaken training as the dominant cost; and where open-weight Chinese models now trail the closed frontier by only months. The report maps market structure, competitive strategy, the economics of scale, regulation, and where value is migrating up the stack.
Misha Laskin
Misha Laskin is the co-founder and CEO of Reflection AI, a New York lab building open-weight frontier models and autonomous coding agents. A theoretical physicist by training (Yale, University of Chicago), he led reward modeling for Google DeepMind's Gemini and worked in reinforcement learning at Berkeley and DeepMind before launching Reflection in 2024 with AlphaGo co-creator Ioannis Antonoglou. In 2025 the company raised roughly $2 billion at an $8 billion valuation, positioning itself as America's open-source answer to DeepSeek.
Deccan AI
Deccan AI is a Mountain View-based GenAI data company that runs the post-training and production layer for frontier labs and enterprises. It builds super-accurate SFT and RLHF datasets, reinforcement learning environments and agentic evaluation pipelines using an elite expert network and a purpose-built quality platform.
MatX
MatX is a Mountain View semiconductor company building chips designed exclusively for large language models. Founded in 2022 by ex-Google TPU engineers Reiner Pope and Mike Gunter, it aims to deliver an order-of-magnitude more performance-per-dollar for frontier model training and inference than current GPUs.
Mercor
Mercor is an AI-powered talent marketplace that recruits domain experts - doctors, lawyers, engineers, scientists - to train and evaluate frontier AI models for labs like OpenAI, Anthropic, Google, Meta, and Amazon. Founded in 2023 by three Thiel Fellows, it has grown from a hiring tool into the human data backbone of the AI economy.
Sarah Wang
Sarah Wang is a General Partner at Andreessen Horowitz (a16z), where she leads growth-stage investments across AI, enterprise applications, and infrastructure. A Harvard and Stanford GSB alumna with rare dual-mode investor instincts - equally comfortable asking 'Is it working?' and 'What if it works?' - she has backed defining companies of the AI era including Cursor, Wiz, Character.ai, Gamma, Decagon, and Thinking Machines Lab, while working closely with a16z's positions in OpenAI, Databricks, SpaceX, SSI, and Figma.