Horizon Together Summit 2025: Journey to the Future with Ecosystem Partners

The inaugural Horizon Together Summit 2025 kicked off in Shenzhen under the theme "Journey Upwards Together" on December 8.

Driven by the industry-wide consensus around open ecosystems, the summit brought together senior executives from across the global industrial chain. The event drew a full spectrum of Horizon’s ecosystem partners, major OEMs, international Tier 1 suppliers, robotics brands and authoritative industry experts.

Participants discussed growth opportunities for smart vehicles and intelligent robots, collaborating to chart new paths for ecosystem cooperation. The summit provided a platform for discussing market trends and exploring new models of ecosystem collaboration.

During the summit, Dr. Yu Kai, Founder and CEO of Horizon Robotics, delivered a keynote address titled Rising Together to New Heights. Drawing on the company’s decade-long industrial practice and in-depth insights, he laid out a systematic overview of how Horizon’s three core technological pillars — its proprietary BPU®, toolchain compiler, and algorithm stack — propel intelligent driving from standalone technical breakthroughs toward cross-industry collaborative growth, empowering a broad ecosystem spanning smart passenger vehicles and general-purpose robots.

Across the product announcements, industry insights, and technology demonstrations, one message remained consistent: to partner with the widest range of ecosystem players, democratize cutting-edge technology once accessible only to a limited number of companies, and make high-end intelligent solutions accessible to enterprises and end users alike, ultimately delivering universal benefits in the intelligent era.

Three Technology Pillars Advancing AI Computing Beyond Moore’s Law

Building on Horizon’s "New Moore’s Law for Intelligent Computing," first proposed in 2016, Dr. Yu compared AI computing to a symphony:

"BPU® serves as a first-class symphony orchestra, the compiler acts as a master conductor, and algorithms function as exceptional composers. Working in perfect synergy, the three pursue performance-power efficiency and performance-cost efficiency, delivering optimal balance in algorithmic efficiency, hardware efficiency, and system flexibility. These three core technological pillars form the fundamental infrastructure for physical-world AI and steer the evolution of intelligent computing."

Over the past decade, Horizon’s BPU architecture has delivered a more than 1,000-fold increase in computing performance, with each generation designed around mathematical principles that improve AI computing efficiency. Dr. Yu Kai offers an incisive analogy: while the Internet aims to capture human worldly behaviors and insights, the essence of artificial intelligence is to continuously approach the true laws of the physical world.

Mathematicians have proven that beneath the complex appearances of high-dimensional spaces lie concise low-dimensional manifold structures. The iterative evolution of Horizon’s BPU architecture alongside autonomous driving technology essentially forms a chronological history of intelligent computing interpreted through mathematical logic.

The iteration journey spans multiple generations: the BPU Bernoulli Architecture, which pioneered fixed-point computing to deliver high-performance, low-power perception for early ADAS scenarios; the BPU Bayes Architecture, which enables efficient end-to-end perception and prediction computing for highway NOA; and the latest BPU Nash Architecture, which efficiently supports interactive game-theoretic decision-making for full-scene NOA applications.

Today, Horizon officially unveils its fourth‑generation BPU architecture: Riemann, named in tribute to the mathematical genius Bernhard Riemann, whose theory of Riemannian manifolds revealed the fundamental nature of complex spaces. As Horizon’s ultimate architecture for general‑purpose robotic computing, Riemann delivers a 10× leap in core operator performance, a 10×+ increase in supported high‑precision operators, full floating‑point computation capability, and a 5× improvement in energy efficiency for large language models. From Bernoulli to Riemann, Horizon’s BPU has scaled compute performance by over 1,000-fold in a decade—far outpacing the pace of traditional Moore’s Law.

The compiler has evolved in lockstep, entering a new era of AI‑driven compilation. As the conductor of the silicon symphony, Horizon’s compiler technology has undergone three generations of iteration. Today, we launch OpenExplorer® 4.0, the fourth‑generation compiler, achieving breakthroughs in precision, speed, and performance. Compilation time has been reduced from hours to minutes, and model performance is boosted by 20%. A key advancement is the AI‑driven optimization strategy, marking the compiler’s transition from tool automation to a new era of machine‑learning‑powered self‑evolution.

With a powerful computing foundation now in place, foundation models serve as the universal musical score for the AI world. Dr. Yu Kai pointed out that foundation models have revolutionized the AI industry in recent years, driving a paradigm shift from task-specific intelligence to general-purpose intelligence. They enable emergent intelligent capabilities and cross-modal fusion that breaks down isolated information silos, laying a critical groundwork for true artificial general intelligence.

For years, Horizon Robotics has focused its efforts on the computing architecture, compiler technology and foundation models that power physical AI. Its innovations have empowered a wide range of physical AI scenarios, including intelligent driving and intelligent robotics.

HSD Together Enables Large-Scale Win-Win

Dr. Yu Kai believes that large models for intelligent driving mark the starting point of foundation models for physical AI. A robust intelligent driving large model delivers three core ecosystem values: it acts as a demonstration of technical capabilities, a shared technical foundation for scalable product development, and a common ground for partners to pursue differentiated innovation freely.

He noted that as the industry fully shifts to a data-driven paradigm, all market players face three major industry-wide challenges: exorbitant computing costs for model training, an acute shortage of top-tier AI algorithm engineers, and ever-faster product iteration cycles measured on a monthly basis.

To address these pain points, Horizon Robotics has continuously evolved its partnership framework, adopting an increasingly open business model to help the industry navigate these hurdles. During the Journey® 2 to Journey® 5 era, Horizon followed the classic cooperation model of "SoC + algorithm IP licensing + hardware reference design", enabling ecosystem partners to launch a wide array of mature, reliable highway NOA products. Entering the Journey® 6P and HSD era, the company advances its technology roadmap, independently developing the full-stack, one-stop HSD solution as a production-ready reference solution. This breakthrough shattered previous performance and experience limits and validated the viability of its end-to-end technical roadmap.

Today, Horizon Robotics announced that it will open its fully validated full-stack capabilities to the entire industry via a brand-new algorithm-as-a-service model: HSD Together. This framework enables deep co-creation with partners under a clear division of labor. Partners can focus on their core strengths, including system integration testing, vehicle-specific adaptation, customized feature development, and hardware & software delivery. In turn, Horizon provides end-to-end algorithm services spanning the entire product development lifecycle, consisting of data services and the AIDI SaaS platform, professional algorithm adaptation engineering & consulting, and its core foundation model licensing.

Dr. Yu Kai explained: "The fundamental goal of the HSD Together algorithm service model is to empower partners to reduce engineering effort, computing resource requirements, and time-to-market by up to 90%, built on Horizon’s mature, field-proven intelligent infrastructure. We aim to help every enterprise efficiently translate top-tier intelligent driving capabilities into unique brand competitiveness with lighter, faster development workflows."

Open Foundations for the Next Era of Vehicles and Robotics

Horizon Robotics’ technological influence and ecosystem layout have long extended far beyond the automotive sector. Dr. Yu Kai stated: "Horizon has built China’s largest computing platform for consumer robots. Our D-Robotics lineup boasts over 100 commercialized products, connecting more than 100 upstream and downstream partners alongside over 100,000 developers."

An ever-growing array of robotic products and platforms embodies the thriving ecosystem. These include Narwal’s premium smart home vacuum robots, Vita Power’s autonomous robotic dogs delivering emotional companionship, and Insta360’s world-first panoramic drone that revolutionizes visual creation. Horizon’s computing power is enabling intelligence across ground-based, aquatic, and aerial devices, weaving smart capabilities into every facet of daily life.

As a leading universal hardware and software infrastructure provider for the robotics industry, Horizon also actively advances full-stack core technologies for embodied intelligence. During the summit, Dr. Yu officially unveiled two open-source foundation models for embodied intelligence: HoloMotion, a foundation model for whole-body motion control, and HoloBrain, a foundation model for spatial perception, reasoning, and manipulation.

HoloMotion, the cerebellum model, specializes in motion intelligence, enabling robots to move with human-like agility. A single model delivers sophisticated motion control across diverse terrains. HoloBrain, the cerebrum model, endows robots with human-like spatial perception and precise manipulation capabilities. It can interpret general user commands and generate sequential motion plans to complete assigned tasks.

In addition, Horizon Robotics is committed to uniting global developers to build a thriving robotics ecosystem and offering optimal growth pathways for developers of all ages. Its D-Robotics division has partnered with more than 500 universities, primary and secondary schools to cultivate talent for the future intelligent industry.

Advancing Together to Make Technology More Accessible

By advancing technology and strengthening ecosystem collaboration, Horizon Robotics aims to make intelligent technology accessible at scale. As the symphony of innovative technologies resonates atop the collaborative ecosystem foundation, a future where intelligent technology is truly democratized for all and empowers a better life is rapidly evolving from blueprint into reality.

Dr. Yu Kai’s keynote laid bare a clear value cycle rooted in technical conviction, forged by a thriving ecosystem, and ultimately serving public well-being. From Horizon Robotics’ perspective, the ultimate destination of the greatest technological revolution lies not in academic breakthroughs confined to research halls, but in meaningful improvements in everyday life.

"By empowering smart vehicles and robots, we make human life safer and better," Dr. Yu Kai concluded his keynote. "Guided by this corporate mission, Horizon Robotics will keep developing and mass-producing cutting-edge technology, turning advanced innovations once available only to a limited market into everyday accessible solutions for everyone."

Horizon Robotics firmly believes that only by journeying together and advancing side by side with the broadest ecosystem of partners can we continue to lower technology barriers and jointly pave the way for intelligent technology to become accessible to people in their everyday lives.