The Alphaist - English Edition

#6 - The State of Embodied AI - Q2 2026: World Models, Dexterous Hands, and Humanoids at Work

Welcome to this episode of The Alphaist. The Alphaist is a deep-dive conversation series hosted by Peter Chen, Founder of Alphaist Partners, a fund focused on hard tech. Each episode explores the first principles of technology and entrepreneurship, featuring early-stage founders, engineers, and product innovators who are shaping the future. This episode is the second installment of our quarterly series on Embodied AI, produced in collaboration with LateTalk — the tech podcast from LatePost, hosted by Manqi Cheng. Manqi has been tracking the embodied AI landscape closely, with deep access to the founders, engineers, and investors driving the field forward. Each quarter, Peter and Manqi take stock of what actually moved the field — across humanoid hardware, dexterous manipulation, foundation models, and the capital flowing into all of it. In this second installment, Peter and Manqi unpack the five developments that defined Q2 2026 in embodied AI: • The Beijing E-Town humanoid half-marathon — where 100+ humanoids ran the course and a phone maker's robotics division (Honor) swept the podium, signaling that large manufacturers can now build competitive humanoids fast • Figure's 100-hour package-sorting livestream — the first time humanoids' industrial value was demonstrated at scale to a global audience, and why non-standard logistics sorting is such a clean fit for today's embodied models • The breakout in dexterous hands — Wuji's second-generation direct-drive hand at ICRA, and why high-DoF hands became the quarter's competitive battleground • NVIDIA's Cosmos 3 — arguably the first fully open-source "Omni" world model, pulling world models from lab concept to industrial-grade product and igniting the hottest startup wave of the quarter • The next turn for VLA — Physical Intelligence's π0.7 and Generalist's GEN-1, and the first real signs of VLA and world models beginning to merge Along the way, Peter and Manqi get into the harder questions the quarter surfaced: • Tendon-drive or direct-drive: is Tesla's tendon-driven hand a first-principles choice, or a bet the independents are right to fade? • Who ends up owning dexterous-hand data — the hand makers, the humanoid body makers, or third-party collectors? • If the general-purpose model giants go Omni, is there any reason a standalone embodied-AI model startup should exist? Listen in for a ground-level read on where embodied AI actually stands. Contact Alphaist Partners Official Website: ⁠⁠https://alphaist.vc/⁠⁠ [https://alphaist.vc/] Partnership & Media: ⁠⁠peter@alphaist.vc⁠⁠ [peter@alphaist.vc] Disclaimer: This episode was originally recorded in Chinese. The English version was translated and voiced using Fish Audio's AI voice tool, with the consent of all participants. Visit ⁠⁠https://fish.audio/⁠⁠ [https://fish.audio/] for more details.

#5 - D-Robotics: The Picks-and-Shovels Play for Embodied AI

Welcome to this episode of The Alphaist. The Alphaist is a deep-dive conversation series hosted by Peter Chen, Founder of Alphaist Partners, a fund focused on hard tech. Each episode explores the first principles of technology and entrepreneurship, featuring early-stage founders, engineers, and product innovators who are shaping the future. In this episode, we are joined by Cong Wang, CEO of D-Robotics, and Wei Sui, VP of Technology at D-Robotics — for a deep dive into what it takes to build the infrastructure layer for embodied AI. While the industry races to build humanoid robots, D-Robotics is betting on a different path: not building robots, but building the intelligent computing platform that powers them. The company has shipped millions of chips, raised $270M in Series B, and is positioning itself as the picks-and-shovels provider for the robotics era. In this episode, Cong Wang and Wei Sui share the story and strategic logic behind D-Robotics, including: * How the team survived a near-fatal downsizing — from a several-hundred-person division to a few dozen — and rebuilt into a standalone unicorn * Why "feeling torn" is the best compass for strategic decisions — and why the team walked away from profitable but misaligned business lines * The two endgames for chip companies: creating demand vs. serving demand — and which path D-Robotics chose * Why D-Robotics is betting on VA over VLA as the more viable path to deployment — achieving comparable performance at 1/10 the cost * What "Robot Infra" really means — and whether it could be an NVidia-scale opportunity Peter has known Cong Wang for six years. This conversation traces the full arc — from a PhD dropout's first startup, through the darkest chapter of layoffs and strategic pivots, to the conviction that building infrastructure for robots is a game worth playing for decades. Listen in for a ground-level read on what it takes to build the computing backbone of embodied AI. Contact Alphaist Partners Official Website: https://alphaist.vc/ [https://alphaist.vc/] Partnership & Media: peter@alphaist.vc [peter@alphaist.vc] Disclaimer: This episode was originally recorded in Chinese. The English version was translated and voiced using Fish Audio's AI voice tool, with the consent of all participants. Visit https://fish.audio/ [https://fish.audio/] for more details.

#4 - The State of Embodied AI - Q1 2026: Humanoids, Dexterous Hands, and World Models

Welcome to this episode of The Alphaist. The Alphaist is a deep-dive conversation series hosted by Peter Chen, Founder of Alphaist Partners, a fund focused on hard tech. Each episode explores the first principles of technology and entrepreneurship, featuring early-stage founders, engineers, and product innovators who are shaping the future. This episode marks the launch of our new quarterly series on Embodied AI, produced in collaboration with LateTalk — the tech podcast from LatePost,hosted by Manqi Cheng. Manqi has been closely tracking embodied AI landscape, with deep access to the founders, engineers, and investors driving the field forward. Each quarter, Peter and Manqi take stock of what actually moved the field — across humanoid hardware, dexterous manipulation, foundation models, and the capital flowing into all of it. In this first installment, Peter and Manqi unpack the five developments that defined Q1 2026 in embodied AI: • Unitree's Spring Festival Gala performance — and what 20+ G1 robots moving in sync really signaled about the current lead in locomotion and hardware consistency • Sharpa's dexterous hand debut at CES — and why the field is converging on high-DoF hands as the next research default • NVIDIA's DreamZero and DreamDojo — and the shift from VLA to world models as a new paradigm for robot brains • Galbot's tennis-playing humanoid — and what real-time, closed-loop control means for our ceiling of imagination • Boston Dynamics' new electric Atlas — whose modular design reflects a distinctly American answer to the manufacturing question Along the way, Peter and Manqi get into the harder questions the quarter surfaced: • Why has no one successfully challenged Unitree's G1 twenty months after launch? • Is Tesla's tendon-driven hand a first-principles choice — or a bet against physics? • Why does the U.S. have only a handful of humanoid companies while China already has over twenty with billion-dollar valuations? • As Unitree files for IPO, what does the coming listing wave mean for the ecosystem? Listen in for a ground-level read on where embodied AI actually stands. Contact Alphaist Partners Official Website: ⁠https://alphaist.vc/⁠ [https://alphaist.vc/] Partnership & Media: ⁠peter@alphaist.vc⁠ [peter@alphaist.vc] Disclaimer: This episode was originally recorded in Chinese. The English version was translated and voiced using Fish Audio's AI voice tool, with the consent of all participants. Visit ⁠https://fish.audio/⁠ [https://fish.audio/] for more details.

#3 - Fish Audio: Beyond Speech, Into Emotion

Welcome to this episode of The Alphaist. The Alphaist is a deep-dive conversation series hosted by Peter Chen, Founder of Alphaist Partners, a fund focused on hard tech. Each episode explores the first principles of technology and entrepreneurship, featuring early-stage founders, engineers, and product innovators who are shaping the future. In this episode, we are joined by Lengyue, Chief Scientist of Fish Audio, and Rissa,CEO of Fish Audio — for a deep dive into the next era of AI voice. Founded from open-source roots, Fish Audio has grown from open-source roots into a leading platform for text-to-speech and voice cloning, powering over 1.1 million user-generated voice models worldwide. In this episode, Lengyue and Rissa share Fish Audio's thinking on AI Voice 2.0, including: why the team bet on a unified end-to-end architecture over cascaded pipelines — and what it unlocked for expressiveness and latency; how "noisy" data — overlapping speech, arguments, emotionally charged conversations — became their most valuable training asset rather than something to discard; and where the road toward full-duplex, real-time voice interaction is ultimately heading. We believe that only when voice truly carries emotion can AI begin to feel human. Listen in and hear what's coming next in AI voice. Contact Fish Audio Official Website: https://fish.audio/ [https://fish.audio/] Contact Alphaist Partners Official Website: https://alphaist.vc/ [https://alphaist.vc/] Partnership & Media: peter@alphaist.vc [peter@alphaist.vc] Disclaimer: This episode was originally recorded in Chinese. The English version was translated and voiced using Fish Audio's AI voice tool, with the consent of all participants. Visit https://fish.audio/ [https://fish.audio/] for more details.

#2 - ACE Studio: The Open Ecosystem Play in AI Music

Welcome to this episode of The Alphaist. The Alphaist is a deep-dive conversation series hosted by Peter Chen, Founder of Alphaist Partners, a fund focused on hard tech. Each episode explores the first principles of technology and entrepreneurship, featuring early-stage founders, engineers, and product innovators who are shaping the future. Our guest is Joe Guo, Founder and CEO of ACE Studio. Founded in 2019, ACE started with AI virtual vocalists and has since evolved into the world's first open ecosystem AI music platform. In February 2026, ACE launched ACE Music, a consumer-facing platform that unifies end-to-end music generation with the full creative depth of ACE Studio. In this episode, Joe shares ACE's fundamental thinking on where AI music is headed, including: why professional-first was the right entry point and how expert feedback builds better models, how ACE's suite of proprietary models — from vocal synthesis to instrument generation to stem separation — creates moats that no single foundation model can replicate, why ACE is betting on an open ecosystem over a closed platform to ultimately beat Suno, and why music may be the first AI-generated content format that is truly ready for end-user consumption at scale. We believe music is a human instinct — not a privilege. AI should give everyone the tools to express it. Listen in and feel the pulse of ACE. Contact ACE Studio Official Website: ⁠https://acestudio.ai/⁠ [https://www.youtube.com/redirect?event=video_description&redir_token=QUFFLUhqbnNnQkFhSGxsUEk3SFdLUEVqRmxuS0xOMERrUXxBQ3Jtc0tuSjh3LUNnalhMTnlsdGxtTmZvVnlpcWMwTDg4bFpkTFRscHhwNTdNdHl5V25kd0JCSWhLU3hqSDhsRnN3U3g5eUF0cG1HcDdtdjhnSm52Z3Jud19VZThxaTE1a3JlZThpSzQwdDhaY2FtTDkxS1Qwbw&q=https%3A%2F%2Fwuji.tech%2F&v=r86fXB8kEgw] Contact Alphaist Partners Official Website: ⁠https://alphaist.vc/⁠ [https://alphaist.vc/] Partnership & Media: ⁠peter@alphaist.vc⁠ [peter@alphaist.vc] Disclaimer: This episode was originally recorded in Chinese. The English version was translated and voiced using Fish Audio's AI voice tool, with the consent of all participants. Visit ⁠https://fish.audio/⁠ [https://fish.audio/] for more details.