AI is taking over the world by storm, and AI generated humans are becoming more popular. What does this mean for the ecosystem? 

In this episode of Impulso Podcast, we sit down with Jensen, co-founder and CEO of Topview.ai, an AI-powered video creation platform.

With over 13 years of experience at Taobao, Jensen has been at the forefront of AI applications, especially in e-commerce and digital media. As AI-generated video becomes more sophisticated, Jensen shares how the shift from small, targeted AI models to large, multimodal models has transformed the industry since 2022. 

From completely AI generated videos, to AI humans in the videos – how will AI change the future of content creation? Or even the videos we see on a daily basis?

Also Available on Spotify and Apple Podcast.

[AI-generated Transcript in Chinese]

[00:00:00] Elody
欢迎来到新一期的 Impulso Podcast。今天我们邀请到了 Jensen 作为嘉宾。
Jensen 是 Topview.AI 的联合创始人兼 CEO。

[00:00:08] Jensen
很高兴来到 Momentum Works 在新加坡的办公室。我们与江玕进行过许多合作,并且双方团队之间有着深厚的相互支持。
我曾在阿里巴巴工作了 13 年,过去两年里,我们创立了一家专注于 AI 生成视频的公司。我们利用数字人技术创建 AI Avatar,并将其应用于电商营销场景。Topview.AI 是我们的核心产品。今天,我将和大家分享一些关于 AI 生成视频的见解。

[00:00:56] Jianggan
那以下是我的问题。你刚才提到你在阿里巴巴工作了 13 年,然后又创业了两年。在这 15 年里,你见证了人工智能技术的发展。你有什么从业经验或者关于这几年来AI技术的关键突破想要分享吗?

[00:01:21] Jensen
嗯,让我想想。如果要把 AI 分成几个主要阶段,一个是 2022 年之前的 AI,另一个是 2022 年之后的 AI。
在 2022 年之前,AI 主要依赖小型、针对性的模型,例如使用深度学习来解决特定问题。
而在 2022 年之后,进入了大模型时代,所有 AI 生成的内容都由大语言模型完成。它成为了一个通用模型,可以解决更广泛的问题。
基本上,自 2022 年以来,市场迎来了爆发,特别是在 ChatGPT 发布后,整个行业跃升到了一个新水平。
目前,AI 发展速度极快,基本上每六个月就会出现一次大的更新。实际上,我认为这个进程正在加速,可能行业内每三个月就会有一次新的重大突破。

[00:02:39] Jianggan
这个行业的竞争越来越激烈了。作为从业者,你会感到焦虑吗?

[00:02:46] Jensen
这其实是件好事,也许有时候确实会有压力。不管是投资人、客户,还是我们自己,我们经常会讨论这些问题。
对于做 AI 应用的公司来说,这是一个巨大的利好。比如说,去年我们使用 OpenAI 的服务,每月可能需要支付 10,000 美元,现在只需要 1,000 美元。

[00:03:11] Jianggan
是的,成本确实大幅下降了。

[00:03:12] Jensen
举个例子,我们最近发现 Gemini 的表现非常好,所以我们切换到了 Gemini
然后过一段时间,GPT-4.5 发布了,我们可能又会换回去。
现在市场上有很多选择。

[00:03:27] Jianggan
是的,选择变多了。

[00:03:27] Jensen
更多的选择、更低的成本、更好的效果和更快的速度——这对 AI 创业和应用发展来说,是最好的时代。

[00:03:39] Jianggan
所以,真正焦虑的其实是那些做大模型的公司。

[00:03:41] Jensen
我觉得,每个人都有自己的焦虑和应对方式。

[00:03:48] Elody
那么,你自己会感到焦虑吗?随着上游选择的增多,进入这个行业的人也越来越多了。

[00:03:57] Jensen
当然,竞争确实激烈。这就是为什么我说,每个人都有各自的竞争方式。
例如,在 AI 应用层面,一个互联网产品的成功取决于对用户需求的理解,以及解决用户痛点的能力,从而在市场上占据优势。
未来,大型 AI 模型将会像云计算一样,成为一种公共服务。
就像我们不会因为选择某个云计算平台而决定一家公司的成败一样,企业的成功与否最终取决于它的业务,而不是使用哪家 AI 模型。

[00:04:31] Jianggan
我们来谈谈 AI 在视频领域的应用,包括你们目前正在做的项目。
过去一年里,每次我去你的办公室,都会发现一些新的变化和进展。

[00:04:48] Jensen
是的,我相信今年的变化会更大。
我们的技术最近有了一次重大升级,可能和你上次来访时已经完全不同了。

[00:04:56] Jianggan
但那次才是两个月前啊。

[00:04:59] Jensen
哈哈,可能是三个月吧。好吧,好吧。
回顾过去一年视频技术的发展,Sora 的发布是一个重要的转折点。
OpenAI 作为 AI 领域的领导者,聚集了最顶尖的人才、最好的算法和最优质的数据,展示了视频生成的可行路径。
但说实话,他们后来进展缓慢,一整年都没有推出任何新的突破性产品。
与此同时,市场上的商业化进展却很快。
中国的公司,比如 Kuaishou-KlingMiniMax,也进入了视频生成领域,国外的 Runway 也在竞争。
中国的许多 AI 大模型公司,比如 ByteDance,也纷纷加入竞争。
Kuaishou-Kling 让视频生成变得更低成本、更稳定,并达到了可商业化的标准。
这是一波浪潮,最初是 Sora 引发的,然后在去年七八月份,Kuaishou 推动了第二波。
六个月后,第三波浪潮到来(指deepseek)。

[00:07:17] Jianggan
所以,去年大家基本上都没休息,对吧?

[00:07:19] Jensen
是的,每个人都很忙,哈哈。

[00:07:23] Elody
问题是,许多竞争对手都推出了新产品。在过去两个月——以及接下来一个月里,你们也会开发新的产品吗?你们是如何决定开发哪些产品的?

[00:07:43] Jensen
回到我之前的观点,AI 应用的核心在于市场洞察、客户需求,以及如何有效利用 AI。这三点是关键。
我们的策略是专注于营销和 ToB(企业服务)。
我们把经典的 4P 营销理论融入了 AI 框架。传统的 4P 理论包括 Product(产品)、Price(价格)、Place(场景)和 Promotion(营销)
我们重新定义为:

  • People(数字人)
  • Product(产品)
  • Place(产品使用的场景)
  • Promotion(营销策略)
    我们的所有内容都基于这四个要素。

[00:12:50] Jianggan
那么,未来 AI 生成内容和真实视频的区别会越来越小,对吧?

[00:12:55] Jensen
是的,也许我们曾经畅想的元宇宙,最终会由 AI 来实现。
在 AI 之前,所有虚拟环境都需要人工设计,因此不够真实。
但现在,AI 可以自动生成完整的虚拟环境和交互式数字人,使得元宇宙的概念更加可行。

[00:13:24] Jensen
目前,在某些情况下,我们仍然可以分辨 AI 生成的视频,比如光影效果可能稍显不自然。
但在未来六个月内,AI 生成的视频可能将变得难以分辨,即便是业内专家也无法区别 AI 生成的内容和真实视频。

[00:14:04] Jianggan
那么下一个问题,你认为 AI 生成的内容正变得难以分辨。
对于营销人员和电商从业者来说,他们应该如何看待 AI?
你认为在 AI 赋能业务价值的过程中,主要的阻力是什么?

[00:14:27] Jensen
我认为这是一个关键问题。
去年,我们与一些客户,特别是做市场营销的客户交流时,他们对 AI 抱有非常高的期望,甚至觉得 AI 无所不能。
但到年底,许多人感到不满意,因为很多概念验证(POC)项目并没有如预期般成功。
这在一定程度上是因为他们的期望过高。去年 AI 的热度达到了巅峰,之后大家的期望自然有所回落。
现在,我们看到新一波 AI 应用潮正在兴起。但与上次不同的是,人们对 AI 的局限性有了更清晰的认识,因此态度更加谨慎。尽管如此,他们对 AI 工具的满意度、接受度以及在实际工作中的使用频率都比以前更高。我认为这是一种螺旋式上升的过程——每一轮的技术发展都会让人们更清楚 AI 的能力和边界。

[00:15:23] Elody
这很有道理。那你认为 AI 的未来是什么?

[00:15:27] Jensen
我们的团队正在逐步收敛我们的产品方向。
过去,我们探索了多个方向,但现在我们已经找到了一类用户,他们对某些功能的满意度非常高。
我们发现,某些特定功能的应用转化率非常高,而且使用频率也很高。
通过数据分析,我们锁定了五个核心功能,其中有一个表现最突出,用户使用率和付费转化率最高。
所以现在,我们正在专注于优化和深化这个功能。
一旦找到一个强劲的功能点,我们就会精细化打磨它,并扩展它的应用场景,将它整合到更广阔的商业策略中。
当前,我最关注的是如何尽快将我们的最新技术推广给更多用户,以抢占更大的市场份额。

[00:16:16] Jianggan
最后一个问题。

[00:16:19] Elody
是的。你刚才提到,目前的重点是数字人与产品的互动。
那么,展望未来,你是否认为 AI 生成的数字人能够无缝地与完整的 3D 沉浸式环境交互?
这是否是 AI 的最终目标?

[00:16:42] Jensen
是的,这是我们的长期愿景,但这并不意味着我们现在就会立即去做它。
过早地进入这个领域可能会让自己陷入困境。
关键在于时机——知道何时迈出下一步。
目前,完整的 3D 集成可能还需要几年时间。
在此期间,研究人员会继续推进基础科学研究。
如果出现突破——无论是来自开源社区还是来自商业研究团队——这些能力最终都会逐步被行业采用。
我认为这个过程是渐进式的,而不是一蹴而就的。
关键是要准确判断技术成熟的时机,以便在正确的时间点推出合适的产品——就像投资一样,哈哈。

[00:17:27] Jianggan
你在杭州生活多年了,对吧?

[00:17:28] Jensen
是的,虽然我不是杭州本地人,但我在杭州生活了很多年。
我也在北京、上海、深圳等地待过。
杭州是一个高度市场化、现代化的城市,可能是中国最适合创业的城市之一。
最近,我认识的许多创业者都在杭州申请政府补贴。
这个过程非常透明——你不需要去拉关系,也不需要请客送礼。
只要符合条件,政府就会直接把资金拨到你的账户里。
这和许多地方的做法完全不同,哈哈。

[00:18:10] Elody
是的,我也听说过类似的故事。
前段时间,我和一些来自义乌的朋友吃饭,他们提到,义乌当地政府正在积极引导企业申请补助。
他们甚至专门招募了一批官员,帮助企业了解政策、申请补贴。
对于许多不是杭州本地的年轻创业者来说,杭州的补贴政策是一个很大的吸引力。

[00:18:31] Jensen
是的,其中一个关键因素就是杭州政府在资金支持上的透明度。

[00:18:34] Jianggan
杭州的确资金充裕。许多地方也有补贴,但透明度因地而异。

[00:18:36] Jensen
没错。杭州的申请流程非常直接、清晰。

[00:18:39] Jianggan
你刚才提到这是杭州的一个优势。那么,还有其他原因让杭州成为 AI 发展中心吗?

[00:18:43] Jensen
政府支持是一方面,但还有其他原因。
比如说,在 AI 和技术领域,杭州的生态非常强大。
许多研究机构,包括浙江大学,都为行业提供了大量人才。
此外,阿里巴巴的影响力也很大,形成了一种技术驱动的商业文化。
这些因素共同塑造了杭州独特的 AI 创业环境。

[00:18:57] Elody
你的许多竞争对手也在杭州。你认为这种产业聚集效应对你们有帮助吗?

[00:19:04] Jensen
在一定程度上是有帮助的。
比如,深圳有智能硬件生态,很多做硬件的公司聚集在那里。
同样,杭州的 AI 生态也很强大,很多公司都在做大模型和视频生成。
但每个城市都有自己的特点。
北京有许多大型 AI 研究实验室,而上海在企业解决方案方面有很强的优势。
所以,我认为杭州的生态是一个优势,但它并不是唯一决定成功的因素。

[00:19:26] Jianggan
那么,你认为最重要的因素是什么?

[00:19:28] Jensen
这是一个多因素的组合。
杭州政府的开放政策、浙江大学的人才供应,以及阿里巴巴的产业影响力,共同形成了这个环境。
此外,杭州的创业文化也很强,吸引了许多 AI 初创公司。

[00:19:58] Elody
以上这些你提到的因素也是你们选择新加坡作为海外扩张的第一站的原因吗?
还是有其他考虑?

[00:20:03] Jensen
是的。新加坡是一个非常适合国际化业务的地方。它很好地平衡了东西方文化。在工作方式、法律流程、企业治理结构等方面,它更偏向西方。但同时,它与中国及亚洲其他地区有着紧密的联系。此外,许多中国创业者的英语能力不错,这使得新加坡成为一个自然的全球化桥梁。

 

[AI-generated Transcript in English]

[00:00:00] Elody
Welcome to a new episode of Impulso Podcast. Today we have Jensen here as our guest.
Jensen is the cofounder and CEO of Topview.AI

[00:00:08] Jensen
I’m glad to be at the Momentum Works’ office in Singapore. We have had a lot of cooperation with Jianggan as well as significant and profound mutual assistance. I used to work at Alibaba for 13 years.  Then in the last two years, we started a business that specializes in the AI-generated video. We will create AI Avatars with digital human technologies and use them in e-commerce marketing scenarios. Topview.AI is our main product. Today, I will share with you some of my insights about AI-generated videos
[00:00:56] Jianggan
Then that’s my question. You just said you worked at Alibaba for 13 years and then started a business for two years. During those 15 years, you witnessed the development of technologies over the years. What kinds of experience do you want to share or what kinds of  key breakthroughs do you want to share about AI.

[00:01:21] Jensen
Well, I see.Then if we divide AI into several major stages, one is AI before 2022 and the other is AI after 2022. Before 202, it was all about small, targeted models. Well, like using deep learning to solve a specific problem. Then after 2022, the LLM era came. So all AI-generated content are done with big models. It becomes a universal model that solves the general problem. Basically, this market boomed after 2022, especially after the release of ChatGPT. Basically, there is a large version every six months. In fact, I think the process is accelerating again. There would be a new version every three months in that industry.
[00:02:39] Jianggan
The industry is becoming more and more competitive. As a practitioner, do you feel more anxious?

[00:02:46] Jensen
It’s better. It’s good, and maybe it’s also true. Whether it’s investors, clients, or ourselves, we often discuss these issues.
Yes, it’s a huge advantage for companies working in AI applications. For example, if we had used OpenAI last year, we might have expected to pay $10,000 per month. Now, it only costs $1,000.

[00:03:11] Jianggan
Yes, the cost is significantly lower.

[00:03:12] Jensen
For example, we recently found that Gemini performs exceptionally well.
So we switched to Gemini.
Then, after a while, GPT-4.5 was released.
We might switch back again.
There are now many choices available.

[00:03:27] Jianggan
Yes, there are many options.

[00:03:27] Jensen
More choices, lower costs, better results, and faster speeds—this is the best opportunity for AI entrepreneurship and application development.

[00:03:39] Jianggan
So, actually, it’s the large model developers who are feeling the pressure.

[00:03:41] Jensen
Well, I think everyone has their own anxieties and ways of dealing with them.

[00:03:48] Elody
Do you feel anxious? As more upstream choices become available, more people are entering the field.

[00:03:57] Jensen
Yes, many. That’s why I say everyone has their own approach.
For example, in the application layer, an internet product’s success depends on understanding user needs and addressing pain points effectively, leading to better market positioning.
Looking ahead, large AI models will become a public service, much like cloud computing.
I never say, “Choosing a particular cloud provider this year will determine my company’s success,” right?
Ultimately, success depends not on the cloud itself but on the business it supports.

[00:04:31] Jianggan
Let’s talk about some applications in the video field, including the projects you are working on.
Over the past year, every time I visited your office, I noticed changes and new developments.

[00:04:48] Jensen
Yes, I believe this year will bring even bigger changes.
Our technology has recently been upgraded, so it might be completely different from when you last visited.

[00:04:56] Jianggan
That was just two months ago.

[00:04:59] Jensen
So maybe three months, hahaha. OK, OK.
Looking back at the past year’s developments in video technology, Sora’s release was a key turning point.
OpenAI has always been a leader in AI, setting the direction by assembling the best people, the best algorithms, and the best data.
They showed Sora was feasible, paving the way for true video generation.
But to be honest, they have since fallen behind, not releasing anything groundbreaking for an entire year.

Meanwhile, monetization strategies have been similar across markets.
Chinese companies like Kuaishou-Kling, and MiniMax have entered the video generation space, alongside international companies like Runway.
Some of China’s original large model companies, such as ByteDance, have also joined the competition.

Kuaishou-Kling has now made video generation more cost-effective, stable, and commercially viable, meeting industry standards for production and monetization.
This was the first wave, starting with Sora. The second wave, led by Kuaishou, happened around July or August last year. About six months later, the third wave arrived.

[00:07:17] Jianggan
So, it seems no one really took a break last year, right?

[00:07:19] Jensen
Yes, everyone was busy. Haha.

[00:07:23] Elody
The problem is, many competitors have launched new products. Over the past two months—and in the coming month—will you also be developing new products? How do you decide which projects to pursue?

[00:07:43] Jensen
Going back to my earlier point, the key to AI applications is understanding market

insights, addressing customer needs, and leveraging AI effectively. These are the three critical factors. Our strategy focuses on excelling in marketing and ToB (business-to-business). We’ve integrated the classic 4P marketing theory into our framework. Traditionally, the 4Ps are Product, Price, Place, and Promotion.
We’ve adapted them as follows:

  •   People (digital humans)
  •   Product (the actual item)
  •   Place (scenarios where the product is used)
  •   Promotion (marketing strategies)

All our content aligns with these four elements.

[00:08:31] Elody
I see, so you’re explaining this framework.

[00:08:33] Jensen
Yes. In our video generation approach, we incorporate elements related to people,

products, and scenarios. For example, a person may be shown holding a product or wearing an accessory. Instead of focusing on price, we prioritize human representation, which AI can enhance. Based on this 4P framework, we’ve recently developed two new features. We can demonstrate their effects later in this video.

[00:10:16] Elody
So, the idea is that viewers can see the product from different perspectives, right?

[00:10:19] Jensen
Yes, they can hold it, wear it, or interact with it naturally.

[00:11:01] Jensen
Yes. For example, two years ago, I predicted that our digital human technology would reach this level around this time of year, and now we have achieved it.
Similarly, last year, we anticipated AI-generated video would advance in a particular direction, and we saw that happen.
At that time, we had internal discussions analyzing market trends and technological developments.
We correctly predicted the rise of open-source AI video models and how competition would intensify with more players.
The only thing we didn’t expect was that OpenAI itself didn’t progress as quickly as we thought. Haha.

[00:11:44] Jianggan
What about predicting the future? What trends do you foresee?

[00:11:47] Jensen
Now, I am analyzing advancements in AI technology, particularly in video generation.

[00:11:51] Jianggan
Then, what stocks should we buy?

[00:11:53] Jensen
Haha, well, that’s a different kind of prediction. Maybe we’ll discuss that in a private group.

[00:11:57] Jianggan
Okay.

[00:11:58] Elody
Speaking of monetization, what about paid AI services?

[00:12:02] Jensen
We are now seeing a transition from generating just the speaker’s mouth movements to generating full facial expressions, and now even full-body animations with gestures.
I believe that in the future, just like Li Fei’s team is working on world models, we will see digital humans seamlessly integrated into real-world 3D environments.
Currently, everything is still in 2D, but once true 3D integration is achieved, it will be a major breakthrough.
We’ve already seen AI progress from generating just a talking mouth to a fully animated body. Now, it’s moving toward fully interactive, AI-generated 3D worlds.

[00:12:50] Jianggan
So, would this be considered the Metaverse? Do you think AI will make the Metaverse more realistic?

[00:12:55] Jensen
Yes, perhaps what we once called the Metaverse will ultimately be realized through AI.
Before AI, everything had to be manually designed and animated. That’s why early virtual environments lacked realism.
Now, AI can generate entire digital environments and interactive virtual humans, making the concept of the Metaverse far more achievable.

[00:13:06] Jianggan
Another question: You mentioned technological leaps in AI-generated content. How can we distinguish between real and AI-generated videos?

[00:13:24] Jensen
Right now, it’s still possible to tell the difference in some cases. If you’ve watched AI-generated videos long enough, you may notice subtle repetitive patterns.
For example, lighting and shadow effects on faces may still feel slightly unnatural.
It’s similar to how, in the early days of photo editing, people could recognize heavily edited images.
However, within the next six months, AI will likely reach a point where the difference becomes indistinguishable, even to experts in the field.
It will be extremely difficult to tell apart AI-generated videos from real footage.

 

[00:14:04] Jianggan
So, the next question. You believe AI-generated content is becoming indistinguishable from real content.
For marketers and e-commerce professionals, how should they view AI?
What do you think are the main obstacles in using AI to generate value?

[00:14:27] Jensen
I think this is a key issue.
Last year, when we talked with some clients, especially those in marketing, they had very high expectations for AI.
They thought AI could do everything.
By the end of the year, many were dissatisfied, as many proof-of-concept (POC) projects did not succeed as expected.
This was partly because expectations were too high. At that stage, AI hype peaked, and later, expectations naturally dropped.

Now, we’re seeing another wave of AI adoption.
This time, there’s been real technological progress, so interest is picking up again.
However, unlike last time, people are now more cautious because they understand AI’s limitations.
Despite this, their satisfaction with AI tools and their acceptance of AI in their workflow are improving.
I see this as an iterative process—each wave of adoption refines people’s understanding of what AI can and cannot do.

[00:15:23] Elody
That makes sense. What do you think is the future of AI?

[00:15:27] Jensen
Our team’s focus has gradually narrowed.
Previously, we explored multiple directions, but now we have identified a user segment with high satisfaction.
We found that certain features lead to high in-app purchase conversion rates and frequent usage.
By analyzing user data, we identified five key features.
Among them, one stands out with the highest engagement and conversion.
So now, we are focusing on deepening and optimizing that feature.

Once we identify a strong feature, we refine it, expand its application, and integrate it into our broader strategy.
Right now, my priority is ensuring that our latest advancements are quickly adopted by more users so we can capture more market share.

[00:16:16] Jianggan
The last question.

[00:16:19] Elody
Yes. Earlier, you mentioned that your current focus is on digital humans and their interactions with products.
Looking further into the future, do you believe that AI-generated digital humans will be able to seamlessly interact with fully immersive 3D environments?
Is that the ultimate goal?

[00:16:42] Jensen
Yes, that is our long-term vision, but it doesn’t mean we need to pursue it immediately.
Rushing into it too early could be detrimental.
The key is timing—knowing when to take the next step.
Right now, full 3D integration may still be a couple of years away.
In the meantime, researchers will work on the foundational science.
If breakthroughs occur—whether through open-source projects or proprietary research—those capabilities will gradually become available to the industry.
I see this as an iterative, step-by-step process rather than a sudden leap forward.
The key is understanding when the technology is mature enough for practical application—just like investing. Haha.

[00:17:27] Jianggan
You have lived in Hangzhou for many years, right?

[00:17:28] Jensen
Yes, although I’m not originally from Hangzhou, I have lived there for many years.
I’ve also spent time in Beijing, Shanghai, and Shenzhen.
Hangzhou is a highly market-oriented and modern city.
It might be one of the most business-friendly cities in China.

Recently, many entrepreneurs I know have applied for government subsidies in Hangzhou.
The process is very transparent—you don’t need to network or pull strings.
You apply, and if you qualify, they quietly transfer the money to your account.
This is very different from how things work in many other places. Haha.

[00:18:10] Elody
Yes, I’ve heard similar stories. Recently, I had dinner with some friends from Yiwu.
They mentioned that Yiwu’s local government is actively recruiting officials to help businesses navigate subsidy policies.
For many young entrepreneurs who aren’t originally from Hangzhou, the subsidy policies are a major attraction.

[00:18:31] Jensen
Yes, and one key factor is the transparency of the funding process.

[00:18:34] Jianggan
There’s a lot of funding available. Many places have subsidies, but transparency varies.

[00:18:36] Jensen
Exactly. Hangzhou’s process is very straightforward.

[00:18:39] Jianggan
Earlier, you mentioned one factor. What’s another reason Hangzhou stands out?

[00:18:43] Jensen
Government support is one factor, but there are others.
For example, in terms of AI and technology, Hangzhou has a strong ecosystem.
Many research institutions, including Zhejiang University, contribute to the talent pool.
There’s also the influence of Alibaba, which has created a tech-driven business culture.
The combination of these factors has led to a unique environment for AI innovation.

[00:18:57] Elody
Many of your competitors are based in Hangzhou.
Do you think clustering with other AI companies is an advantage?

[00:19:04] Jensen
It helps to some extent.
For example, Shenzhen has a strong smart hardware ecosystem, making it a great place for hardware startups.
Similarly, Hangzhou has developed a strong AI ecosystem, with many companies focusing on large models and video generation.
That said, each city has its own strengths.
Beijing has a concentration of large AI labs, while Shanghai has a strong presence in enterprise solutions.
I think Hangzhou’s ecosystem is beneficial, but it’s not the only factor driving success.

[00:19:26] Jianggan
So, what do you think is the most important factor?

[00:19:28] Jensen
It’s a combination of factors.
The openness of the government, the talent coming from Zhejiang University, and the influence of Alibaba have all played a role.
There’s also a strong culture of entrepreneurship, which has attracted many AI startups.

[00:19:58] Elody
Is that why your first overseas expansion was in Singapore?
Or were there other reasons?

[00:20:03] Jensen
Singapore is a great place for international business.
It strikes a balance between Eastern and Western business cultures.
The work habits, legal processes, and corporate governance structures are very much aligned with the West.
At the same time, there are strong ties to China and other parts of Asia.
Additionally, many Chinese entrepreneurs speak English well, which makes Singapore a natural hub for expanding globally.

[00:20:39] Jianggan
This has been an interesting discussion.
Over the years, we’ve had many conversations with Jason and his team.
There’s so much happening in AI that it’s difficult to explain everything in words—you really have to see it in action.

That’s why we regularly visit Jensen’s office in Hangzhou as part of our business immersion programs.
If you want to experience it firsthand, come and join us.

Thanks for joining this episode of the Impulso Podcast. See you next time!