I was recently at the China-Indonesia Digital Economy Forum in Jakarta. It was an event focused on growing AI collaboration between Indonesia and China.
And that’s an important topic. It’s on my short list.
Everyone is talking about the China-US relationship for AI. Usually with political rhetoric. But I’ve been watching the China-SE Asia relationship. That’s really ground zero for the China AI tech going global. SE Asia is usually the first stop. Then it’s usually Latam and Europe.
So here are my three take-aways from the forum. And thanks to Tencent for flying me to the event and providing my hotel.
Lesson 1: We Are Seeing Real Progress in AI Cloud Computing Between China and SE Asia.
Cloud is the natural first step for most businesses in SE Asia in AI. Cloud is an easy way to access the new AI tools. It’s much easier than building. And Chinese AI Cloud businesses are particularly aggressive in the region right now. You can get pretty good deals.
For example:
- Alibaba AI Cloud is very active in SE Asia, especially in the Philippines. And especially in AI-focused ecommerce services for merchants and brands. They are arguably the market leader for IaaS right now in the region.
- Tencent AI Cloud has been expanding pretty aggressively in SE Asia for the past year.
- Huawei Cloud is definitely expanding but the focus is more about hardware and data centers.
- Baidu Cloud doesn’t seem to be doing much in SE Asia, which is surprising.
The event I point to for growing China-SE Asia cooperation is the cloud migration of GoTo last year.
In June 2025, Indonesian-based GoTo (previously Gojek and Tokopedia) did a big migration of their on-demand services to Tencent Cloud. In practice, that meant moving +1,000 micro services over. And apparently this big switch happened within 4 hours and 52 minutes.
The GoTo-Tencent project was a pretty symbolic event. GoTo is the quasi-national champion of digital for Indonesia. Although a co-founder was just sentenced to jail last week.
In the near term, this type of AI cloud project is about optimizing operations and consolidating data.
- You want everything on one consolidated platform. And getting one centralized data depository is a pre-requisite for building out new tools.
- Keeping all the data in local cloud servers is also increasingly important (i.e., data sovereignty).
- Stability is also a big issue.
And over the long-term, this is about creating a platform for intelligence capabilities.
- You want to be on a platform that is aggressively rolling out new AI tools and capabilities. This can be advanced frontier models as well as low-cost, downloadable models.
- AI Cloud is a good way to deploy new tools quickly with minimal risk. Then you can bring your key capabilities on premise over time.
One of the interesting services Tencent Cloud has been taking international is its famous mini programs. WeChat is still the world’s only real super-app because of the huge suite of mini programs. And Tencent has not really tried to extend this outside of China since 2015.
But now Tencent Cloud is offering this capability as a service to businesses. They will help you build your own mini programs. That’s how the Abu Dhabi government rolled out its local super-app for government services. “Mini programs as a service” is a pretty unique cloud offering.
I’ve been writing a lot about Tencent going international this past year. And I interviewed their VP Karl Xu. Here are the details.
Lesson 2: Open-Source AI is Creating Win-Win Collaborations. And Accelerating Local Innovation.
In 2025, we had the “DeepSeek moment”, when a small team in Hangzhou dropped the first low cost, open source, open weight, downloadable LLM. It shocked the world and changed the strategy for LLMs.
This was followed by lots of other China models adopting the same approach. One year later, we have:
- DeepSeek V4, which has the flagship Pro and the efficiency-optimized Flash tiers. They also have the reasoning-focused DeepSeek-R2 series.
- Alibaba’s Qwen series.
- There is the Qwen 3 / Qwen 3.7 series, which includes specialized Qwen3-Coder variants and Qwen3-VL for vision tasks.
- Zhipu AI’s GLM-5 and GLM-5.2
- These have been in the news a lot this month.
- Moonshot AI’s Kimi K2.5 and Kimi K2.7 Code
- The MiniMax-M2.5 and MiniMax-M3
- Tencent’s Hunyuan series
- Tencent has transitioned from a fully closed ecosystem to offering highly competitive open weights architectures.
- Baidu’s ERNIE open series
- Baidu has historically maintained a strict proprietary wall around its ERNIE foundation models. But it has also pivoted to open-source specific tiers to retain developer adoption.
- ByteDance’s Seed series
- ByteDance open sources its foundation architectures under the Seed banner. This is distinct from its closed proprietary Doubao system.
- Xiaomi’s MiMo series
- Xiaomi manages its open weights ecosystem through the MiMo AI team. This group builds models designed to bridge software development with smart home hardware ecosystems.
- Meituan’s LongCat series
- Meituan publishes its near-frontier open weights models under the LongCat AI initiative. The architecture gained initial fame on public router platforms under the stealth name Owl Alpha.
- They recently made the news because their new 1T parameter model was trained on Huawei (not Nvidia).
- Huawei’s openPangu series
- Huawei released the initial piece of its next-generation open architecture to the developer community on June 30, 2026.
- This family features the recently launched openPangu 2.0 Flash model, with the flagship openPangu 2.0 Pro model arriving shortly after.
Nvidia is making big moves in open source in the US. But outside of that, it’s mostly China players in this space right now.
And we do see a spectrum between fully and partial open source / open weight models. Some are also more low-cost focused than others. And some are more international vs. domestic focused than others. DeepSeek is still arguably the leader.
This is all great for developers and businesses in Southeast Asia. Really for developers and businesses everywhere (note: a shocking number of Silicon Valley startups use DeepSeek).
The low-cost aspect was the original differentiator versus the frontier models (OpenAI, Anthropic).
But now the focus is shifting towards the greater control and data sovereignty you get with downloadable open-source models. Businesses are getting much more concerned about outsourcing their alpha to frontier models, which seek to commoditize them.
The net result of all of this is lots of local innovation. That’s the ROI to watch in Indonesia and SE Asia.
Lesson 3: Agents, Agents, Agents. But Scaling is a Challenge.
Agents are the big story of 2026.
Imagine a world with 8B humans and 800B agents. Plus, +20B robots.
That’s what is coming. And that means lots of change and disruption. Likely starting with content creation and agentic commerce.
The big challenges with deploying agents are currently interconnections, variable costs and scaling.
- The interconnections agents require are complicated and still emerging, especially outside of China. For agents to operate online, they need MCPs and other protocols. To interact with merchants and each other. And also, for payment, identification and governance. The payment rails are a big problem.
- Agents also have significant variable costs. Generative AI isn’t cheap like traditional software. Increasing usage resulting in rapidly increasing costs for businesses. That’s true of GenAI in general. And it turns out to be much more true for agents (who never sleep). They use way more compute than humans. So, the costs of agentic activity are substantial. This is another reason why most models will need to be locally hosted (i.e., downloaded) using your own compute. It’s also another reason why low-cost models (i.e., Chinese) and better for most activities.
- Scaling agents is the biggest problem. Even if you solve the cost and connectivity problems, getting these activities to scale within a business means they have to be integrated into the workflows. You have to rebuild the workflows in the high-impact areas, before you can scale up to thousands of agents. And getting the quality and governance right is a pre-requisite to scaling. This is what people are struggling with right now.
For China, I am mostly following the agents of Alibaba and Tencent and I expect Tencent to become the leader. I recently wrote about their new WorkBuddy and CodeBuddy agents. These are their flagship agent products this year.
This is where Tencent (and China) have an advantage. Tencent (and Alibaba, ByteDance and Baidu) all have robust ecosystems. Deploying and scaling agents within these ecosystems is much easier than operating between businesses. For example, WeChat already has the identity, payment and governance systems in place. I’m expecting agents in China to advance much faster than in the West.
This should be a boon for SE Asia, because the China agents are going international right now. And again, SE Asia is the first stop. Tencent’s QClaw is already available outside of China. And WorkBuddy is launching this month.
In my own business, I am mostly using WorkBuddy and Claude Cowork right now. Those are the two leading models for people who don’t want to try to set up OpenClaw on their own.
***
Overall, those are my take-aways. It was a pretty good event. Lots of interesting questions were raised. And China-SE Asia AI collaborations are an important topic.
Cheers, Jeff
Disclosure. I have had a paid consulting relationship with Tencent in the past twelve months.
——-
I am a consultant & keynote speaker on how to increase digital growth and strengthen digital AI moats.
I am the founder of TechMoat Consulting, a consulting firm specialized in increasing digital growth and strengthening digital AI moats. Get in contact here.
I write (a lot) about digital growth and digital AI strategy (3 best selling books, +2.9M followers on LinkedIn). There is a free book and email newsletter below.
My Moats and Marathons book series is a framework for building and measuring competitive advantages in digital businesses.
Note: This content (articles, podcasts, website info) is not investment advice. The information and opinions from me and any guests may be incorrect. The numbers and information may be wrong. The views expressed may no longer be relevant or accurate. Investing is risky. Do your own research.
