Original · Unique Research · 2025-11-12
Editorial note: This is a complete English edition of the historical article and interview. Biographical details, product milestones, claims of being first or globally unique, state-of-the-art performance, and the reported 100-fold productivity gain and reduction to one percent of previous costs reflect the source’s reporting and Su Guoli’s statements, not independently verified current results. The source supplies no comparable benchmark protocol or cost breakdown for those performance claims. The discussion of “2P” concerns products for professional practitioners, while “2C” concerns consumer products. Relative references such as April, October and “currently” are preserved in their historical context. Predictions, analogies and the “eight of ten measures” saying express the speaker’s views, not measured market shares.
If there is one especially exciting line from this wave of AI, it may be Su Guoli's observation that defining a new form of content consumption is an opportunity that comes along only once in decades.
Most teams are watching performance benchmarks, model parameters and Sora2 demos. He is watching something else: when video truly becomes an AI native medium, will it rewrite both how people spend their time and how they enjoy it?
A member of the generation born in or after 1995, Su Guoli has sold two companies. In 2023, he built Tripo from the ground up at VAST; Tripo later became a leading global AI 3D tool. After 2024, he took on an even more ambitious challenge at Sand.ai: competing directly for the vast amount of time users spend with video.
In his opinion, there are roughly only two ways for AI to benefit mankind:
One is “Save time”—automating repetitive work. The GPT series, he believes, will eventually become the most powerful general-purpose productivity tool in human history.
The other is “Enjoy time”—when people watch short videos and dramas or play games, can AI improve the experience itself, rather than merely sharpen recommendations and make the feed more addictive?
Sand chose the latter. From the company's first day, Sand.ai has been preparing for it: 2P is a certain opportunity, but 2C offers a boundless horizon.
When Models Begin to Contain Stories
For video, the watershed Su Guoli points to is Sora2.
Not because of how realistic its images are, but because it is the first video model to truly internalise the ability to construct a complete narrative loop.
In the past, video tools were essentially 2P: what was given to you was cheaper shooting, more efficient editing, and more intelligent keying. You have to be a practitioner first before it can help.
The arrival of models such as Sora2 suggests another possibility: the model itself can understand a video's setup, development, turn and resolution, and sustain a complete sequence on its own. The barrier between creators and ordinary people has fallen by another order of magnitude. Video creation is beginning to move slowly from the 2P world toward the 2C world—from building tools to creating a new generation of content itself.
Sand set out on this path earlier, and with greater conviction, than outsiders may realise.
In April that year, they released the autoregressive video model Magi-1, which they said remained the world's only high-quality autoregressive video model. In October, they released Gaga-1, which generates audio and video together and, according to the company, achieved global state of the art in character performance.
These names sound technical, but for Su Guoli, they are just milestones on the road:
First define what the next-generation content platform should look like, then work backward to identify the technical prerequisites—the parts that others will not build, or will not build well, if we do not build them—and solve them one by one. Magi-1 and Gaga-1 are only the first pieces in that chain.
What really excites him is seeing creators adopt these technologies as everyday production tools. One team, he said, has already used Gaga-1 to produce short dramas at 100 times the previous efficiency and roughly one-hundredth of the former cost.
In that world, the ability to shoot or edit is no longer the most important barrier. The landscape of content production is being redrawn: the old divide was between those with equipment and those without it; the new divide will be between those who can tell compelling stories and those who can design more engaging premises, characters and pacing.
Once models can tell stories, people can no longer contribute merely by pressing the shutter.
Good Companies, Exceptional Companies and AI-Native Organisations
Thinking clearly about the technical direction is just the first step. The real difficulty lies in the organisation.
Su Guoli has a very vivid metaphor:
A poor company is a limping company—stronger in technology but weaker in product, or formidable in business while its technology can only chase others from behind.
A good company is an all-rounder—technology, product and business are all strong.
Top companies are a third type: they integrate all their insight and resources to define a product that is structurally revolutionary. The model contains product functions, while the product itself naturally contains a commercial flywheel.
In his eyes, OpenAI is the benchmark in this sense:
ChatGPT is not merely an API with a front-end skin. Its underlying capabilities, product form and path to commercialisation fit together naturally. As the model improves, the product becomes more useful; as the product becomes more useful, the commercial flywheel becomes more complete; and that flywheel in turn feeds the model.
Therefore, what Sand has been doing sounds a little old-fashioned, but it is especially scarce in AI entrepreneurship:
Vertically integrating cross-functional teams, so technology, product, operations and business examine problems together from the outset instead of each writing its own OKR;
Continuously strengthening foundational organisational capabilities requires not only the ambition to be first, but also the patience to smooth out a complex system piece by piece.
In the traditional Internet era, the core of making products is to understand users;
In the AI era, he adds another half to that principle:
Building products now requires understanding not only the user, but also the model. The opportunity lies beyond the current boundary of model capabilities but within the boundary of user needs.
That may sound abstract, but it explains the many Agent products that flare up and quickly disappear:
If your killer feature becomes a standard capability of mainstream models six months later, then you have built little more than a temporary layer for transferring capability.
A product that endures must build a new structure on top of the model's foundation: a new interaction paradigm, workflow or form of content.
Agentic: a Feature Along the Way, Not an End-State Label
Over the past year, Agentic AI has been one of the hottest buzzword terms.
Su Guoli's judgement was very calm:
Agentic is better understood as a key characteristic that gradually emerges as model capabilities advance toward AGI, rather than as a label that can independently constitute an industry.
Many teams rush to put the Agent label on their products—letting a model call tools, write scripts and chain workflows on its own, making it appear more intelligent ++.
But in his view, the greatest risk is that the complex architecture you worked so hard to assemble today may become a default capability of tomorrow's foundation model.
For entrepreneurs, then, the question is no longer “Should I build an agent?” but two connected questions:
As models evolve over the next year or two, which capabilities will become basic infrastructure, like water, electricity and gas?
How far beyond those utilities must I go to reach problems that will remain within the sphere of human value creation even as models improve?
The importance of Agentic is not whether you use tool calling capability. It is that Agentic forces you to redraw the division of labour between people and models—which parts of mechanical reasoning the model handles, and where humans contribute value judgements and choices about worldview.
This is also an implicit reminder from Su Guoli to AI creators:
Do not turn yourself into an agent who simply works harder and pushes more relentlessly.
Your value is not in clicking a few more buttons for the model, but in the decisions you make on its behalf and the long-term responsibilities you assume.
Day1 Global and China's Virtual Manufacturing Industry
On globalisation, his judgement is direct: the new generation of Chinese companies is naturally Day1 global.
Especially in 2C markets, content products such as video can cross languages and cultures: humour, excitement, suspense and reversals often share underlying structures across countries.
2P business will be more regional, and it is necessary to dig deep into local industries, regulations and customer relationships;
2C products compete for people’s attention worldwide and look for shared aspects of human nature.
In his opinion, the greatest advantage of Chinese entrepreneurs is that they are close enough to the Chinese Internet market.
If we only look at the richness of online content consumption, he used a very classic statement:
If the world's richness of online content consumption amounted to ten measures, China alone would hold eight.
In fact, our virtual manufacturing industry is not inferior to the physical manufacturing industry:
China has many short-video platforms, subject areas and operating playbooks, while growth hacking, content operations and A/B testing have been pushed to an extreme. For AI video teams, this is both a brutal competitive environment and an enormous training ground—a place to iterate, make mistakes and correct them at the greatest speed, before taking a product forged in the world's most competitive market to a global audience.
The challenge is also similar:
Whether physical or virtual, moving into the premium end of global markets requires a long transition from low-end supply to high-end brand.
Whether Chinese teams can complete that journey in AI video depends on two things:
Whether they possess sufficiently world-class foundational capabilities;
And whether they have the courage to make a product others feel compelled to use, even if they cannot fully explain why it feels better.
A Higher Floor and a More Distant Ceiling
The theme of this conference is “Pioneering Intelligence | The Individual Era.”
If Su Guoli's interpretation is summarised in one sentence, it is probably: AI is raising the floor and ceiling at the same time.
The floor is rising because AI is making more ordinary individuals relatively versatile and self-sufficient:
There is no need for a complete film crew. One person can complete the work that used to require more than a dozen people with a few prompts;
Lacking modelling, special-effects or post-production skills is no longer an inherent weakness; models can help fill those gaps.
The ceiling is pushed away because the real super individual has gained greater leverage:
In the past, your creativity may have been held back by execution: you could imagine an immense universe, but lacked the budget to film it;
Today, the cost of execution is collapsing. You can experiment freely, start over and rebuild from scratch.
Worlds that once existed only in notebooks, drafts and late-night conversations now have a chance to appear on screen in full for the first time.
For AI creators, the biggest opportunity is:
Models are democratising creative capability, pushing the barrier for newcomers and amateurs to reach the top to an unprecedented low.
You no longer need ten years of industry experience to enter. In a sense, as long as you dare to tell a story that others have not told, you will have a chance to be seen.
But the challenge is equally sharp:
You must ask yourself a question again. In a world where AI can also write scripts, storyboard, shoot and edit, what is my value?
Is it a deeper understanding of human emotions?
A deeper understanding of a particular vertical?
The ability to construct a more complex worldview?
Or a personality and taste, formed through sustained creative work, that others cannot copy?
This is not a question that can be answered in one night, but it is an unavoidable act of self-examination for everyone who wants to go further in the individual era.
Let AI Help You Master AI
At the end of the conversation, when asked how individuals or small teams should prepare over the next 1–3 years to seize the AI dividend, Su Guoli gave an answer that sounded a little lazy but was actually very honest:
Everyone should put this question to a large model in light of their own circumstances—let AI help you master AI.
At first, this sounds like handing the question back to the tool. Look deeper, however, and it reflects a characteristically AI native mindset.
Do not treat AI as a technology that must be fully learned before it can be used. From the outset, treat it as a co-author of your thinking.
You can let it help you diagnose:
Which parts of your current workflow can already be automated, and which still need you for now?
Is your real comparative advantage thinking, expressing, organising resources, or looking at the world from a special perspective?
If you break down what you do today, which parts will models eventually replace, and in which parts will stronger models make you more valuable?
No one can give everyone the same standard answer to those questions.
But if you are willing to keep talking with the model and with yourself, and to keep correcting course through practice, you are already doing the most important thing—not chasing AI, but rewriting your own coordinate system together with AI.
Sand.ai is trying to rewrite how humans consume video content;
And for every individual, the question more worth thinking about may be:
In a world where videos can tell stories by themselves and models can act by themselves,
What kind of irreplaceable story do you want to contribute to this era?
Selected Interview Q&A
Q1: First, briefly introduce yourself and Sand.ai?
Su Guoli: I am Chris Su Guoli, a serial entrepreneur born in or after 1995 who has previously sold two companies. In 2023, I formally committed myself to AI entrepreneurship, first building Tripo, a world-leading AI 3D tool, at VAST, and later joining Sand.ai to explore AI-native 2C product forms. Sand.ai was founded in early 2024. Its long-term vision is to accelerate the broad benefits of AI for humanity, and it has focused firmly on video generation from the start.
Q2: Why did you lock in AI video + 2C + entertainment from the beginning?
Su Guoli: There are probably two paths to making technology broadly beneficial: Save time and Enjoy time. Language models represented by GPT will ultimately become the strongest general-purpose productivity tools, corresponding to Save time. Before brain-computer interfaces arrive, video is the highest-density form through which humans receive information, so from the beginning we decided to build 2C entertainment around AI video technology.
Q3: What are Sand.ai’s most important products and technological breakthroughs today?
Su Guoli: We first defined what the next-generation content platform should look like, then worked backward to identify the most critical technical dependencies and tackle the parts that, if we did not do them, others would not do or would not do well. There have been two interim achievements: the high-quality autoregressive video model Magi-1, released in April, which remains the only one of its kind globally; and Gaga-1, released in October, which generates audio and video together and achieved global state of the art in character performance.
Q4: How far away is the consumer product you envision?
Su Guoli: Technical breakthroughs are continuing to advance. Gaga-1 is only one milestone. We can already see the outline of the ideal 2C product, and it is getting closer to an MVP form.
Q5: What originally motivated you to pursue AI entrepreneurship?
Su Guoli: It is very simple: I like to create. AI is a sufficiently fundamental variable to open an exceptionally large space for everyone who likes to create, which is why I am willing to work in this field for the long term.
Q6: In your opinion, what is the most critical breakthrough in the field of generative AI recently?
Su Guoli: In video, it is of course Sora2. It is the first model in the field to truly internalise closed-loop narrative ability. That means the gap between creators and ordinary people has narrowed by another order of magnitude, marking the transition of video creation from 2P to 2C.
Q7: What was your team's reaction when Sora2 came out?
Su Guoli: We had already formed a clear direction and made relevant predictions quite early, so Sora2 did not surprise us. It helped confirm the feasibility of one direction we had been exploring and provided early validation that AI video could define a new kind of 2C product.
Q8: Under the rapid iteration of technology, how do you balance technological innovation and business implementation?
Su Guoli: Organisationally, we have been vertically integrating cross-functional teams while strengthening foundational capabilities. Simply put, a bad company is a limping company with obvious shortcomings. A good company is an all-rounder in technology, product and business. A top company integrates all its insight and resources to define a structurally revolutionary product with a built-in commercial flywheel.
Q9: Which company exemplifies that top tier for you?
Su Guoli: OpenAI. It is the kind of company whose model contains product functions and products contain commercial closed loops. The three evolve together in a tightly connected way—a path we strongly endorse.
Q10: How do you view and practice the new paradigm of Agentic AI?
Su Guoli: We have already carried out substantial exploration. Agentic is fundamentally a key characteristic that gradually emerges as model capabilities advance toward AGI. The key is to understand the boundaries of model capabilities. The killer feature of many interim agent products may become standard functionality in foundation models within six months or a year.
Q11: In the AI era, what is the most important change in making products?
Su Guoli: In the past, when making products, it was more about understanding users; now, on the basis of understanding users, we also need to understand models. To do something beyond the boundaries of model capabilities and within the boundaries of user needs, the cognitive requirements of the founding team are particularly high.
Q12: What do you think of the difference between 2P and 2C in terms of globalisation?
Su Guoli: 2P is highly regional and requires going deep into the industries and customers of a particular region. 2C is more about finding common ground and serving global users with a single product. We clearly devote more of our energy to the latter.
Q13: Do you have any special experience with transnational data privacy, ethics and cultural differences?
Su Guoli: We do not claim any special insight here. We mainly learn from and refer to pioneers so that we can avoid repeating their mistakes.
Q14: What do you think about the role of Chinese AI enterprises in the global market?
Su Guoli: The new generation of Chinese companies should all be Day1 global. I think Chinese entrepreneurs' greatest advantage is their proximity to China's Internet market. If we consider only the richness of online content consumption, then, to borrow the classical saying, of the world's ten measures China alone holds eight. Our virtual manufacturing industry is no weaker than our physical manufacturing industry.
Q15: What about the challenge?
Su Guoli: The challenge is similar: gradually moving from the low and middle ends of the market to the high end. Whether physical or virtual, a business that wants to secure a premium global position has to endure that upgrading process.
Q16: How do you understand the theme “Pioneering Intelligence | The Individual Era”?
Su Guoli: I think there are two parallel changes: on the one hand, super individuals get greater leverage through AI; on the other hand, non-super individuals also become relatively versatile because of AI, and they can handle many things independently.
Q17: Can you give a specific example of your empowerment of individual creators?
Su Guoli: There are already creators using our newly released Gaga-1 to make short dramas. If you simply quantify it, the production efficiency is about 100 times the original, and the cost is only one percent of the original.
Q18: In your opinion, what is the biggest opportunity for AI creators in the AI track?
Su Guoli: The opportunity lies in models democratising creative capability. Newcomers and amateurs alike have the chance to become leading creators; this is no longer something only a small number of people can afford to pursue.
Q19: What about the most severe challenge?
Su Guoli: The challenge is to understand your relationship with AI. Given that AI already exists and will continue to evolve rapidly, creators need to reconsider where their true value lies instead of competing with models for the same work.
Q20: Looking forward to the next 1–3 years, what advice do you give to individuals and small teams?
Su Guoli: I actually think everyone should ask a large model this question in light of their own circumstances—let AI help you master AI. Different people and different starting points will naturally have different optimal solutions. The key is to develop the habit of thinking and iterating together with AI as soon as possible.