From July 19th to 23rd, SIGGRAPH 2026 was held in Los Angeles, USA. This annual event, which brings together leading scholars, artists, and engineers in the fields of computer graphics and visual computing from around the world, attracted nearly 10,000 attendees from 69 countries. Rendering, geometric modeling, animation simulation, and generative AI—almost all the cutting-edge explorations on how to use computers to create the visual world were presented in these five days.

This year, under the spotlight of this grand event,

VAST, a leading global general artificial intelligence company

(Developed the Tripo series of large 3D models) and became one of the most frequently mentioned Chinese teams at the conference:

He joined Disney, NVIDIA, and Bolt Graphics in the lineup of four keynote speakers at this year's conference.

Five papers selected for Technical Papers

It covers mesh generation, 3D Gaussian generation, texture generation, interactive 3D asset generation, and cross-topology animation.

0a4074f03176e1a2bdcaf67483900067.png

VAST Chief Scientist Dr. Cao Yanpei's Keynote Speech

The team also won the top prize in the conference's signature Real-Time Live! segment.

046527c88b3872ce445da55a11fcef7f.png

VAST and World Labs, founded by Fei-Fei Li, jointly hosted a hackathon.

And the World Model Industry Night. Concurrent with the conference,

VAST Chief Scientist Cao Yanpei Selected for MIT Technology Review TR35 China List

Several events combined to make this company, which was only three years old, attract a lot of attention.

If there's one segment of SIGGRAPH that best represents the most noteworthy voices in the field of computer graphics this year, it's the Keynote address. This year, there were only four Keynote presentations, from Disney, NVIDIA, Bolt Graphics, and VAST.

Dr. Cao Yanpei led the algorithm team to the Main Stage

He delivered a speech entitled "The Teapot Test: First It Tested Showing. Now It Tests Making".

686a8cb3c7925773c099add365c9bf41.png

There's an old story behind this question in computer graphics: In 1975, Martin Newell, a pioneer in computer graphics, created a teapot model by hand measurements. This was one of the earliest 3D models in human history and later became the standard test object for evaluating rendering capabilities in the entire field of graphics. For half a century, the teapot test has consistently examined whether a computer can realistically represent a three-dimensional object.

However, Cao Yanpei believes that in the era of generative AI, this question that has lasted for half a century needs to be redefined.

Today, generative models can skip rulers, graph paper, and manual input. Input a sentence, and the model can generate a teapot with a smooth surface, realistic lighting, and the ability to be viewed from different angles—a result that looks complete enough.

Cao Yanpei then made three requests of the teapot: to put it into a game, to manufacture it using a 3D printer, or to use it to train robots to pour tea. None of these requests could actually be fulfilled. The reason is that there wasn't actually a teapot in the image.

9f86a1144c9488b6f9399d5c531527a1.png

VAST Chief Scientist Dr. Cao Yanpei's Keynote speech. Speech link: https://www.youtube.com/watch?v=zZ8fbEeaG-8

The model can generate a large number of beautiful images, but it may not have a reliable mesh, editable topology, realistic geometric details, movable joints, and physical properties that can be read by the simulator.

As Cao Yanpei said: You can look at it from all sides, but you can't really touch it.

Fifty years ago, the teapot test assessed whether a machine could display an object. Today, the new teapot test asks: Can a machine truly create an object?

The creation mentioned here means that the generated results can enter the game engine, be 3D printed, and become objects for robot training and physical simulation.

This is actually the second time VAST has been included in the SIGGRAPH Keynote speaker lineup.

Three years ago, VAST founder and CEO Song Yachen shared the stage with world-class entrepreneurs such as Jensen Huang, becoming the first Chinese entrepreneur to deliver a keynote speech at SIGGRAPH. At that time, generative 3D was still in its early stages, and the industry's most direct vision was to significantly lower the barrier to entry for modeling.

Three years later, the same company took to the main stage again, and the discussion had expanded from how to generate a model to how to build a full-stack 3D production system and how to drive a continuously operating world.

In the Real-Time Live! segment, VAST won the top prize.

In SIGGRAPH's Real-Time Live! segment, teams need to demonstrate their technical effects in real time in front of a real audience. Any lag or mishap will be exposed, making it one of the most anticipated segments of the entire conference.

f4afbce1e10823e426f822c0cf9f8eca.png

This year, a total of 8 projects were selected for this segment, ranging from a major presentation such as Disney bringing the animated character Olaf to life as a robot, to Runway showcasing a single image generating a real-time AI character.

The VAST team's project, "Create Interactive 3D Assets in Seconds!", demonstrated how generative AI can create fully textured, skeletally bound, and directly animable 3D assets in seconds. Audience prompts submitted on the spot instantly generated corresponding assets and were displayed in a shared virtual world in real time.

d012bc759d8e3125a5330cc5cec36ac5.jpg

The prompt submitted by the user at the event generated a treasure chest with "SIGGRAPH 2026" written on the front.

The capabilities truly demonstrated in this demonstration are far more complex than simply generating a model from a given sentence.

The system needs to understand user input, generate a 3D structure, complete texture mapping, predict reasonable skeletons and skinning, enable the character to move, and finally put multiple results into a scene that can run in real time.

cd6e0e26319148d72fc81b51e3343e38.jpg

If any step goes wrong, the final presentation may fail.

This project won the top prize in the Best in Show—Real-Time Live! category, as selected by the judging panel.

73d6bd2103546a683875448cbf914793.jpg

Audience members participate in the creation process, and the generated 3D props will randomly appear on the track. Cars that hit the props will score points, and the car with the higher score within 30 seconds will win.

Winning the top prize in such a low-tolerance on-site setting proves that VAST's capabilities, as described in its paper, are indeed capable of running without editing in front of thousands of people.

Sharing the stage with World Labs, a public dialogue with the world's top world model teams.

Another notable trend at SIGGRAPH 2026 is that the boundary between large 3D models and world models is rapidly disappearing.

During the conference, NVIDIA showcased its advancements in neural rendering, simulation, and the Cosmos world model, and positioned 3D graphics as a crucial foundation for physical AI to understand, generate, and simulate the world.

World Labs also positions itself as a spatial intelligence company, hoping to build models that can perceive, generate, reason, and interact with the three-dimensional world.

VAST's position in this global technology discussion has also become clearer.

From July 18th to 19th, prior to the opening of the conference,

VAST and World Labs, founded by Fei-Fei Li, jointly hosted a hackathon called Worlds in Action.

The event attracted over 500 creators to build prototypes of next-generation AI-native 3D interactive experiences within two days. The judging panel also included industry experts from organizations such as Lionsgate, Sony, Meta, and the NBA. The prize pool, combined with the opportunity to showcase the products at SIGGRAPH, made it a significant draw.

2a1020d506a5b0d6360f9080b17aa8bf.png

On July 20, the two parties jointly launched an industry networking night called World Models & GenAI Mixer, which brought together about 200 practitioners from AI platforms, basic model companies, visual effects studios, virtual production teams and creative agencies. VAST, World Labs and Groupcore Technology also held a dialogue on AI-driven next-generation 3D content production.

For a Chinese company, being able to co-host an event at a forum like SIGGRAPH with World Labs, a company founded by top academics, is itself a sign of peer recognition. This speaks volumes about its industry standing, more so than a single product launch or ranking achievement.

This has also changed VAST's role: it is no longer just showcasing a 3D generation technology developed by a Chinese team, but has begun to act as an important participant in the global field of AI 3D basic models, discussing the next stage of technological boundaries and industry direction with leading companies.

7 月 19 日至 23 日，美国洛杉矶，SIGGRAPH 2026 如期而至。这场汇聚全球计算机图形学与视觉计算领域顶尖学者、艺术家与工程师的年度盛会，吸引了来自 69 个国家的近万名参会者。渲染、几何建模、动画仿真、与生成式 AI，几乎所有关于如何用计算机创造视觉世界的最前沿探索，都在这五天里集中呈现。

而今年，这场盛会的聚光灯里，

全球领先的通用人工智能公司 VAST

（研发了Tripo系列3D大模型）成为大会上出现频率最高的中国团队之一：

，与迪士尼、英伟达、Bolt Graphics 一同进入本届大会四场主题演讲阵容；

，覆盖网格生成、3D 高斯生成、纹理生成、可交互 3D 资产生成和跨拓扑动画。

VAST 首席科学家曹炎培博士 Keynote 演讲

团队还在大会标志性的 Real-Time Live! 环节拿下最高奖

VAST 与李飞飞创立的 World Labs 联合举办黑客松

和世界模型行业之夜。大会同期，

VAST 首席科学家曹炎培入选《麻省理工科技评论》TR35 中国区名单

几件事叠加在一起，让这家成立仅三年的公司备受瞩目。

如果说 SIGGRAPH 有什么环节最能代表这一年图形学领域最值得所有人都听一听的声音，那就是大会的 Keynote 环节。今年的 Keynote 只有四场，分别来自迪士尼、英伟达、Bolt Graphics，以及 VAST。

曹炎培博士带领算法团队登上 Main Stage

，带来题为《The Teapot Test: First It Tested Showing. Now It Tests Making》的演讲。

这个题目背后有一段计算机图形学的老典故：1975 年，计算机图形学先驱 Martin Newell 用手工测量的方式，做出了一个茶壶模型，这是人类历史上最早的 3D 模型之一，后来成为整个图形学领域检验渲染能力的标准测试对象。半个世纪以来，茶壶测试考验的始终是计算机能否把一个三维物体逼真地呈现出来。

但曹炎培认为，到了生成式 AI 时代，这道延续了半个世纪的考题需要被重新定义。

如今，生成模型已经能够跳过尺子、方格纸和手工录入。输入一句话，模型便可以生成一只表面光滑、光影逼真，还能从不同角度观察的茶壶，这样的结果看起来已经足够完整。

曹炎培随后向这只茶壶提出了三个要求：把它放进游戏，用 3D 打印机制造出来，或者用它训练机器人倒茶。这三个要求都无法真正完成。原因在于，画面背后根本没有一只真正存在的茶壶。

VAST 首席科学家曹炎培博士 Keynote 演讲。演讲地址：https://www.youtube.com/watch?v=zZ8fbEeaG-8

模型可以生成大量精美图像，但它未必拥有可靠的网格、可编辑的拓扑、真实存在的几何细节、能够运动的关节，以及可以被仿真器读取的物理属性。

正如曹炎培所说：你可以围着它看，却无法真正触碰它。

五十年前，茶壶测试的是机器能否将一个物体展示出来。今天，新的茶壶测试开始追问：机器能否真正创造一个物体？

这里所说的创造，意味着生成结果可以进入游戏引擎，被 3D 打印，成为机器人训练和物理仿真的对象。

其实这是 VAST 第二次进入 SIGGRAPH 的 Keynote 演讲阵容。

三年前，VAST 创始人兼 CEO 宋亚宸曾与黄仁勋等世界级企业家同台，成为第一位在 SIGGRAPH 发表主题演讲的中国企业家。当时，生成式 3D 仍处于早期阶段，行业最直接的想象是大幅降低建模门槛。

三年之后，同一家公司再次站上主舞台，讨论的问题已经从怎样生成一个模型，扩展到怎样建立全栈 3D 生产体系，以及怎样驱动一个持续运行的世界。

在 Real-Time Live! 环节，VAST 拿到最高奖项

SIGGRAPH 的 Real-Time Live! 环节，团队需要在真实观众面前，实时演示技术效果，任何卡顿或翻车都无处遁形，这也是整场大会里关注度最高的环节之一。

今年共有 8 个项目入选这一环节，既有迪士尼把动画角色 Olaf 实体化为机器人这样的重量级展示，也有 Runway 展示的单图生成实时 AI 角色。

VAST 团队带来的项目《Create Interactive 3D Assets in Seconds!》，现场演示了如何用生成式 AI 在数秒内做出完全带纹理、已绑定骨骼、可直接动画的三维资产，观众当场提交的 prompt 能够即时生成对应的资产，并实时呈现在共享的虚拟世界中。

现场用户提交的 prompt 生成一个宝箱，正面写着 SIGGRAPH 2026。

这场演示真正验证的能力，远比输入一句话生成一个模型复杂。

系统需要理解用户输入，生成三维结构，完成纹理映射，预测合理骨骼和蒙皮，让角色产生动作，最后将多个结果放进一个可以实时运行的场景中。

其中任何一个环节出错，最终展示都可能失败。

这个项目拿下了由评审团评选的 Best in Show——Real-Time Live! 环节的最高奖项。

现场观众参与创作，生成的 3D 道具会随机出现在赛道上，赛车撞到道具即可得分，30 秒内分数更高的一方获胜。

能在这样一个容错率极低的现场环节拿到最高奖，证明了 VAST 在论文里描述的能力，是真的能在没有剪辑的情况下、当着几千人的面跑起来的。

与 World Labs 同台，全球顶级世界模型团队的公开对话

SIGGRAPH 2026 的另一个明显趋势，是 3D 大模型与世界模型之间的边界正在快速消失。

英伟达在大会期间集中展示了神经渲染、仿真和 Cosmos 世界模型等进展，并将 3D 图形学视为物理 AI 理解、生成和模拟世界的重要基础。

World Labs 也将自身定位为一家空间智能公司，希望构建能够感知、生成、推理并与三维世界交互的模型。

VAST 在这场全球技术讨论中的位置也变得更加清晰。

大会开幕前的 7 月 18 日至 19 日，

VAST 与李飞飞创立的 World Labs 联合举办了一场名为 Worlds in Action 的黑客松

，吸引了 500 多位创作者在两天内搭建下一代 AI 原生 3D 交互体验原型，评审团里还出现了狮门影业、索尼、Meta、NBA 等机构的行业专家身影，奖金池加上 SIGGRAPH 现场展示机会构成了不小的吸引力。

7 月 20 日，双方又联合发起了一场名为 World Models & GenAI Mixer 的行业社交夜，聚集了约 200 位来自 AI 平台、基础模型公司、视效工作室、虚拟制片团队和创意机构的从业者，VAST、World Labs 与群核科技三方还就 AI 驱动的下一代三维内容生产展开了一场对谈。

对一家中国公司来说，能在 SIGGRAPH 这样的场合与 World Labs 这样由学界顶流创立的公司联合发起活动，本身就是一种同行认可的体现，这比单次产品发布或榜单成绩更能说明其行业位置。

这也让 VAST 的角色发生了变化：它不再只是向外展示一项中国团队做出的 3D 生成技术，而是开始作为全球 AI 3D 基础模型领域的重要参与者，与头部公司共同讨论下一阶段的技术边界和产业方向。

He will be featured at the SIGGRAPH main venue keynote, joining Disney, NVIDIA, and Bolt Graphics in the lineup of four keynote speakers at this year's conference.

登上 SIGGRAPH 主会场 Keynote，与迪士尼、英伟达、Bolt Graphics 一同进入本届大会四场主题演讲阵容。

This Chinese company took center stage at the prestigious international conference SIGGRAPH.

这家中国公司站上国际顶会SIGGRAPH的C位

Lately, I've often been asked by friends how they evaluate the combination of large-scale models with major domestic and international mobile phone manufacturers such as Apple and Huawei, as well as the AI phones that OpenAI, Doubao, Jieyue Xingchen, and even DeepSeek may plan to launch.

Is an iPhone killer on the horizon? What opportunities will the emergence of AI smartphones bring to the development of the AI industry both domestically and internationally? Against the backdrop of a global tech war, where do domestic industries have a chance to win this race?

More than a decade ago, I personally participated in a technology and product trial directly planned and led by senior leaders of Huawei. Almost at the same time that Hinton, Fei-Fei Li, and others on the other side of the ocean were exploring and ushering in the golden age of artificial intelligence, China's technology leaders had already begun exploring the industrialization of artificial intelligence based on their own business strategies.

Huawei's initial motivation for embarking on this technological exploration was to find its strategic positioning for the future after the mobile internet era. The ultimate result of this exploration was the launch of the first-generation Honor Magic at the end of 2016, which was widely recognized at the time as the world's first AI phone.

I have no intention of advocating for Huawei or arguing about who had the world's first AI phone. The important thing is to look back at the origins of that technological exploration more than a decade ago, and the birth of the Magic Phone, to revisit its thought process and procedures, and to discuss its successes and lessons learned.

This can provide a reference for the development of the current domestic AI industry, especially for smart devices represented by smartphones, as well as related industries such as robotics.

Starting point: The concept of "application calling application"

In late 2010, at the Shanghai World Expo's Ship Pavilion, Xu Zhijun, then Vice Chairman of Huawei and Rotating Chairman of the EMT, and I had a discussion.

With the rapid development of the mobile internet and the emergence of various apps, what will the future of mobile phones look like?

Back then, there weren't as many apps as there are today, and enthusiasts were used to downloading apps from third-party app stores like Wandoujia. Also, the storage and chip capabilities of mobile phones at that time were far less advanced than they are today.

However, Huawei's top management has already keenly and clearly predicted a scenario:

As user-facing apps become more widespread, user behavior will be increasingly digitized, making it possible for networks and devices to record and understand user behavior.

As users use more diverse and frequent applications, the granularity of user behavior data becomes finer, leading to increasingly higher accuracy in this analysis. Data networks and various hardware devices are becoming increasingly adept at understanding users, making better ways for devices to interact with users possible.

The conclusion of the discussion at the World Expo was that: in the future, standardized software will be replaced by irregular software; and future mobile phones will transcend the stage of smartphones and become intelligent phones.

Xu Zhijun immediately decided to develop this intelligent system and create a "smartphone." Later, he further summarized this idea and logic.

He believes that in the era of mobile internet, the core essence of smartphones is "using applications to call other applications".

I also joined Huawei and began to lead and participate in this technical project that explores the ability to "call applications of applications". The name of this technical project has a strong historical background, and it is called iMax.

This was the strategic starting point for the world's first AI phone. However, in 2010 and for a long time afterward, Huawei executives and I were not used to using the term "AI phone," and instead used another term, "intelligent phone," to describe the upcoming AI phone.

This project was never about making a better phone from the beginning, but about exploring a completely new strategic path after the smartphone era, represented by the mobile internet.

There will be a smarter, more intelligent system that understands users better, which will call upon various service capabilities to proactively provide services to users.

In December 2016, Huawei's Honor brand officially launched the Honor Magic, equipped with a smart system called Magic Live. It supports scene awareness and more intelligent interaction. While fully protecting user privacy and data security, it can continuously learn and understand user habits and connect different apps to proactively serve users.

Standing on the stage, Honor CEO Zhao Ming used the term "Intelligent Phone" for the first time in public. A year later at the CES exhibition in the United States, Huawei Consumer Business CEO Yu Chengdong also began to use the term.

The technological exploration behind the first AI Phone

I often get curious friends asking me why Huawei was able to make an AI phone more than a decade ago, with a mix of curiosity and skepticism.

It was no coincidence. In 2012, Hinton and Sutskever's convolutional neural network won the ImageNet competition, and the deep learning boom was about to begin.

At that time, there was also a surge of interest in deep learning in China's academic and industrial circles.

In particular, research institutions both domestically and internationally, such as NEC Research Institute in Japan, have cultivated a group of Chinese AI scientists. Later, they joined companies like Baidu and ByteDance, or founded their own AI companies, continuing their research in areas such as recommendation systems, facial recognition, and autonomous driving.

Huawei has also recruited a number of top AI scholars, including Ren Zhengfei himself, who mentioned basic academic topics in AI such as machine learning in his public speeches 10 years ago.

However, as an industry giant, Huawei's focus from the beginning was on how to implement its technology and how to integrate it with its business. The challenge was that the technology, which now seems to have a clear answer, was a huge problem in 2011.

Google's landmark paper, "Attention is All You Need," was published in 2017, six months after the Magic phone was released. Two years later, OpenAI embarked on its path of developing large language models, attempting to achieve better results by brute-force training data with massive computing power.

Previously, Huawei and I had already developed some basic technical ideas and paid attention to the "fastweights" that Hinton mentioned in his early years, considering it a kindred spirit—it was this concept that inspired and influenced Google's famous paper "Attention is All You Need".

There are mathematical connections between everything, and intelligent systems can find these interconnected red lines.

In 2012, during a report to Huawei's senior management, the IMAX team used lines from the opening of the American TV series "Touch" to describe the interconnectedness of everything in the world. The show's narration specifically noted that these lines originated in China.

The iMAX project team's mission was to build an intelligent system before ChatGPT, or even before the Transformer architecture, officially emerged.

In late 2012, the largest available internet data came from several social media platforms. Based on this data, the IMAX team trained a rudimentary "language model" system, which also included some specific hand-crafted rules.

Some of the ideas and practices in this regard are very similar to those later adopted by OpenAI on ChatGPT.

Obtaining such a system is just the beginning of the challenges and difficulties of creating an AI phone. How to fit such an intelligent system into a phone presents an even greater and more difficult technical, engineering, and market challenge, encompassing several aspects:

In 2012, the storage and chip processing capabilities of smartphones at the time were insufficient to support and accommodate such a large and complex system, and could not contain so many rules.

Apple's first phone with an NPU chip and AI algorithms was released in 2018, and Apple used it only to solve facial recognition issues.

The Magic phone, released at the end of 2016, uses more complex algorithms, including determining whether to display information on the screen based on the user's eye movements, and deciding whether to directly generate calendar reminders, navigation information, and purchase information based on the user's communication.

Today, some of these features are still retained in Huawei's high-end flagship phones.

Some of them have even become role models for the "AI phones" that are currently being released.

The challenges extend beyond the combination of hardware and systems. From a market perspective, the timing for this type of phone to be recognized and accepted by users and the market is far from ripe; and there are also significant challenges in internal organizational communication.

The process of the iMAX system evolving into the Magic phone was actually a process of exploring the edge of survival.

Huawei's forward-thinking strategic vision and awareness, along with its strong supply chain and robust software and hardware teams, enabled this project to transform from a strategic concept into a real product, allowing the Magic phone to enter the market 10 years ago.

The essence of AI phones: the phone "immerses" itself in a large model, rather than the model being stuffed into the phone.

Three months ago, I came across a new book, *Project Maven*, which tells the story of a Marine Corps colonel who, despite internal and external pressures, pushed the U.S. military to adopt AI technology to transform its combat methods. The book mentions the technological and organizational challenges brought about by AI, which are very similar to the progress of Huawei's iMAX project, something I have personally experienced.

Looking closely at the current attempts of several major manufacturers and AI model companies to develop AI smartphones, while admiring the courage of these entrepreneurs, we also deeply feel the challenges of their paths.

Given the current technological and capital conditions, Apple, Huawei, OpenAI, and various major model manufacturers have joined forces to try and explore true AI smartphones.

The end of smartphone usage seems to be fast approaching.

But as Magic's progress and ending demonstrate, combining a truly intelligent AI system with hardware is fraught with challenges, from design to chips and systems, and especially the business model.

The hardware, market, and business model challenges that Magic once faced, as well as the explorations it made, are worth reflecting on and learning from for today's AI smartphone explorers.

The essence of AI phones lies in enabling the AI system to exist proactively yet silently, understanding users and human nature. It's about immersing the phone in a "large model," rather than stuffing the "model" into the phone.

Even with better software algorithms and intelligent models today than they were 10 years ago, technology, engineering, and product development remain key factors.

If mobile phone hardware is outsourced, it's difficult to talk about "immersing" it in a large model.

In terms of products, given the extremely immature algorithms and hardware at the time, Magic not only grasped the strategic direction of AI phones but also encountered some practical engineering challenges in integrating AI algorithms into mobile phone hardware. These included issues such as how to protect user data and privacy, and how to implement the various rules governing AI algorithms in complex scenarios.

For AI hardware to be more closely integrated with user data, it involves user data security and privacy protection. After the release of Magic, it drew a strong backlash from an internet giant, who demanded the removal of the Magic Live system, citing user data concerns. At the time, a Huawei executive personally responded to inquiries, presciently proposing the idea that data belongs to the user and should serve the user.

Judging from the final results of the iMAX and Magic projects, how to build a business model for AI phones is the most worthwhile topic to discuss.

Currently, major smartphone manufacturers and internet apps have formed a perfect alliance in terms of business models.

App developers rely on major mobile phone manufacturers to ensure their user base, while major mobile phone manufacturers rely on apps to generate profits far exceeding those of mobile phone hardware. The core of this model is bidding for rankings.

Magic's AI phone approach of "application calling application"

It has bridged the gap between the division of labor between apps and hardware platforms, while also eroding the advertising-based business model of major mobile phone manufacturers.

Whether this approach to smartphones was worthwhile during the golden age of smartphone growth over the past decade is a matter of Huawei's own judgment and choice. Since apps themselves have already generated substantial profits for major smartphone manufacturers, why create an application on top of an application, thus revolutionizing themselves?

In terms of sales volume, the Magic phone was negligible within the entire Huawei ecosystem. Afterwards, Honor retained the Magic brand but shut down the Magic Live system, the application that could access other apps.

Over the past year, OpenAI has made many commercialization attempts, among which advertising still has the greatest potential.

Before the release of Magic, I left Huawei and the iMAX project and went on to work for two large financial giants in the field of fintech, focusing on big data and artificial intelligence. I also continued to review, analyze and reflect on the iMAX project.

The corresponding consideration is that for AI-powered smartphones to achieve success, they cannot be separated from AI-driven innovation in application services.

While serving users, AI smartphones must provide opportunities for value creation and also offer greater opportunities for commercial success through application innovation.

Current AI smartphones, including the Magic from 10 years ago, have not yet ventured into this field.

近来，我经常会被身边的朋友问到，如何评价苹果、华为等国内外手机大厂与大模型的结合，以及OpenAI、豆包、阶跃星辰、甚至DeepSeek可能计划推出的AI手机。

iPhone Killer是不是会马上来临？AI手机的出现会对国内外AI产业的发展带来怎样的机会？全球科技热战的背景下，国内产业赢下这场竞赛的机会在哪里？

十几年前，我曾亲身参与了一场由华为公司高层领导直接策划、领导的技术、产品试验。几乎就在大洋彼岸的辛顿、李飞飞等人探索和开启人工智能黄金时代的同时，中国的技术领导者们已经从自己的业务战略出发，开始了人工智能产业化的探索。

华为公司开始这场技术探索的初衷，是在移动互联网时代后，寻找自己未来的战略定位。这场探索的最终结果，首代荣耀Magic在2016年底推出，这是当时被外界公认的全球首款AI手机。

我无意替华为出头，争竞谁是世界首款AI手机。重要的是，回顾10多年前的那场技术探索的源起，以及Magic手机的诞生，重温其思路和过程，讨论其成败教训，

可以为当前国内AI产业发展，尤其是对智慧手机为代表的智慧设备，以及机器人等相关行业的发展提供借鉴。

起点：“应用调用应用”的构想

2010年底，在上海世博会船舶馆的现场，时任华为公司副董事长和EMT轮值主席的徐直军曾经和我有一场讨论，

面对移动互联网的高速发展，各种APP层出不穷的出现，未来的手机样态会是什么？

当时的APP数量还不像今天这样种类繁多，爱好者们还在习惯用豌豆荚这样的第三方应用商店去下载应用，当时的手机存储和芯片能力也远不像今天这样强。

但是，华为公司高层们已经敏感、清晰地预测到了一个场景：

用户端APP应用会不断普及，会让用户的行为被不断数据化，使得网络和终端记录、了解用户行为成为可能。

用户使用应用的门类、频率增多，用户行为数据颗粒变细，这种分析的准确度也越来越高。数据网络和各种硬件设备，有能力变得越来越了解用户，设备与用户更好的交互方式将会成为可能。

世博会现场的那场讨论的结论是：未来规则化的软件，将会被不规则的软件替代；未来的手机会超越智能手机smartphone的阶段，变成智慧手机。

徐直军当场拍板，把这个intelligent的系统做出来，做一台“智慧手机”。后来，他进一步将这个思路和逻辑做了总结，

他认为移动互联网时代之后，智慧手机的核心要义是“以应用调用应用”。

我也加入了华为公司，开始领导、参与这项探索能够“调用应用的应用”的技术项目，这个技术项目的名称有着浓厚的时代痕迹，就叫iMax。

这就是世界首款AI手机的战略起点。只不过，在2010年以及此后的很长时间里，华为高层和我自己都还不习惯用AI手机这个词，而是一直使用另外一个名词“intelligent phone”，来描述即将出现的AI手机。

这个项目从一开始就不是为了做一台更好的手机，而是为了探索移动互联网为代表的智能手机时代之后的全新战略路径——

会有一个更聪明、更智慧，更了解用户的系统，来调用各种服务能力，主动为用户提供服务。

Analysis and field notes on China’s technology ecosystem.

Latest articles

This Chinese company took center stage at the prestigious international conference SIGGRAPH.

A 10-year-old AI phone revealed a secret that no major mobile phone manufacturer dares to admit today.

Jia Yangqing has started a new business, building an AI army that has increased the speed of GLM-5.2 by 534%.

Mysterious dark horse - Moke Robot surges to the top of WorldArena rankings, achieving 810E Godhood with only 32 True Martial Arts cards.

For thousands of robots to enter crowds, they first need to "clearly see what's right next to them within a meter."

1.1 trillion! Shandong's "Optical Module King" debuts on the Hong Kong Stock Exchange, opening below its IPO price.

AI teammates are joining the company; Tencent Cloud aims to restructure the next generation of R&D workflows.

Shenzhen-based catering robot company secures 100 million yuan in funding, with China Merchants Group investing.

Breaking News! The United States bans the import of Chinese humanoid robots.

Zhiyuan's full-modal large model has reached number one globally! Surpassing Google and Nvidia, it outperforms major companies.

Less is more; now it's the big model's turn to prove it.

A 9-meter-tall behemoth "comes back to life"! On the opening day of the Shenzhen Natural History Museum, a biomimetic dinosaur from across the border ignited the crowd.

Shenzhen's semiconductor equipment "little giant" raises nearly 1 billion yuan.

Alibaba Tech Employees' Summer Event: Human vs. Machine Showdown, AIGX New Product Launch, and Chat Between Two Cities

Learn to squeeze toothpaste with robots! ZhiZaiWuJie launches its first implicit tactile world model.