Alibaba vs. Tencent's battle to become the most successful internet to AI company

Takeaways from recent earnings, the AI strategies becoming more defined

Hi all, I recently received really valuable feedback from a reader. He said he felt I was writing for an audience that already knew the names and the technology, not for people who want to learn about them. I always try to provide context and nuance, so apologies if that hasn’t been coming through lately.

So let’s go back to our bread and butter: understanding Chinese big tech strategy. With Alibaba and Tencent both having just reported quarterly earnings in the past few weeks, this is a good moment to look at how the two are actually playing AI.

What’s becoming clearer with every quarter is that the hyperscalers are trying to integrate AI into their distribution and do all they can to enhance their offerings. As much as the discussion is around how labs are monetizing, for the big tech, the real money still gets made one layer down: providing infrastructure and service add-ons. That’s true for Alibaba, that’s true for ByteDance, and in some capacity it’s true for Tencent.

---

Fortune: Alexander Wells

Same same but different

It’s super easy to overgeneralize and ask “how’s China monetizing AI?” Well, to start, China’s two largest internet platforms are both investing heavily in AI, but their strategies have diverged from the get-go. Despite both trying to sell to enterprise, their ways of pursuing that couldn’t be more different.

Alibaba says it’s rebuilding commerce around AI. But the Qwen app itself hasn’t really taken off, hasn’t changed consumer behavior at any fundamental level, and frankly the agentic activity loop isn’t mature enough yet. Tencent is trying to turn WeChat into an agentic operating system, but it’s still cautious, wary of data breaches and of any agent going rogue inside the most important app in Chinese life.

For both companies, AI isn’t a standalone product. It’s what they’re leveraging to elevate their existing businesses. Or, to be cynical about it, to remain relevant. On the consumer side, this probably isn’t something people will pay for as a SKU. It gets embedded into existing businesses, workflows, and user habits. The win is engagement and stickiness inside the ecosystem.

Alibaba’s answer is cloud plus commerce. Tencent’s answer is WeChat plus agents (until they get a stronger proprietary model?).

You can argue Alibaba’s playbook (sell cloud first, then the Qwen APIs, then services on top) now looks like the cleanest and most lucrative strategy. Alibaba is building from the infrastructure layer upward: Qwen models, Alibaba Cloud, Model Studio, enterprise agents, and on top of that, AI merchant tools and shopping assistants inside Taobao and Tmall. It wants AI to be both a cloud revenue engine and a new shopping interface. The cloud business will likely continue to see notable growth in the coming months as well. As its flywheel takes off, the stronger Qwen models run, the more cloud sales too.

For more big tech coverage, see here.

Tencent is building from the distribution downward, maybe unwillingly, because its cloud has never outcompeted Alibaba’s, and its Hunyuan model continues to lag behind peers too. What it has is the ubiquitous WeChat: Mini Programs, Video Accounts, ads, search, payments, the social graph, and the ability to put agents on top of it all. Tencent doesn’t need users to open a new AI app every day. It needs AI to make WeChat more useful, ads more effective, creators more productive, merchants more efficient, and transactions more seamless.

Alibaba: AI as the new commerce and cloud OS

Kicking off the most recent March quarter earnings call, Alibaba CEO Eddie Wu said: “Alibaba’s full-stack AI investments have progressed from incubation to commercialization at scale. This quarter, we achieved accelerated breakthroughs across models, cloud infrastructure, and applications.” He flagged that Cloud Intelligence’s external revenue growth accelerated to 40%, with AI-related products accounting for 30% of that revenue.

Alibaba’s strategy has been clearer from the start, and some may say it is the more aggressive and more vertically integrated one. Models (Qwen). Infrastructure (Alibaba Cloud). Enterprise products (Model Studio, agent tools). And the commerce layer on top (Taobao, Tmall, Alibaba.com, merchant tools). It’s trying to wire all of this into a single AI-first stack.

In the March quarter remarks, Eddie Wu said: “We see massive potential for agentic AI; we launched multiple enterprise AI agents for office and coding use cases and fully integrated e-commerce capabilities into the consumer-facing Qwen app, deepening synergies between AI and our consumer ecosystem.”

On the other side of the world, Meta has been trying to do something similar, planning to integrate its agentic shopping tool into Instagram. But people underestimate how hard this actually is. An agentic shopping app is not a chatbot that can talk about products. It has to read messy human intent, query a live product catalog, maintain state across multiple steps, process the actual payment, assume liability when something goes wrong, and trigger a return or refund that flows back into real logistics. Each one of those is its own integration. And every integration you don’t own becomes a third-party dependency that can break, reprice, or simply say no.

That’s why when OpenClaw (the open-source AI agent framework OpenAI later acquired) first launched, it had a huge developer moment but never reached normal users. It was a personal AI assistant that ran locally on your own machine and controlled a browser by clicking through pages the way you would. Brilliant in concept, but a bit more brittle in practice, at least hard to scale. The moment a checkout page changed its layout, the agent broke. The moment you actually needed to pay, you were back to handling your own credentials and cards yourself. No matter how many people tried to download it and scale it up, it didn’t work, because the agent layer itself was fine, but everything underneath it (auth, payment, fulfillment, recovery when something went wrong) was still somebody else’s stack.

So credit where credit is due, as much as Qwen has not taken off or disrupted user behavior much yet. It was a genius move. Qwen has access and ownership across the layers, which is why it could actually ship an app that runs end-to-end. Unlike OpenAI needing a Stripe deal, or Meta still figuring out payment partnerships, Alibaba already has shopping, payment, post-purchase service, and logistics sitting in one integrated stack. No third party needs to make it seamless. There is no third party.

The strategically important thing isn’t that Qwen is improving. It’s that Qwen has Taobao and Tmall built within it, and it has Qwen wired into the apps; it becomes a recursive learning loop.

Traditional e-commerce starts at a search box. You roughly know what you want, type some keywords, scroll, compare reviews, ask the merchant a question, place an order, track delivery, deal with returns if it goes wrong.

Alibaba says Qwen Shopping Assistant can help with idea generation, product discovery, in-sale support, order management, and post-purchase service. So this isn’t a smarter search page. It’s an AI sales associate that runs across the whole journey. Lower friction for consumers, potentially higher conversion for merchants, and for Alibaba, a new control point between the user and the product result page.

That last bit is the strategic prize. If the assistant becomes the interface, Alibaba can then reshape how products are discovered and how merchants compete for visibility. Of course, it means how merchants market themselves will change too.

As the old interface was keyword search plus recommendation feed, the new one could be conversational intent plus agentic execution.

The risk is the flip side of the same point, though. If AI compresses the shopping journey, it disrupts how Alibaba actually monetizes discovery. The old model sold attention across search results, displays, and feeds. If the assistant just gives users fewer, more direct recommendations, the new interface has to lift conversion enough to make up for the lost ad inventory. So it’s powerful but not risk-free, as AI can improve shopping and also disrupt how the platform’s existing business models work. Ask Google about it…

Anyway, but the market reacted quite positively to it all. Alibaba shares surged 7% as the company doubles down on AI. Quietly, the company has achieved greater depth and breadth in AI integration than most assumed.

Ok, so you know how I kinda shat on the idea of agentic commerce? How it only made sense for utility needs, not discretionary spending? And the lack of engagement doesn’t make sense for clothing/food, whatnot. So I think that, in the next stage, to really ensure engagement, innovation might need to be part of the interface. Just like Brian Chesky recently said on TBPN, “I think the future is not apps. The future is agents, but I don’t think they’re going to be text-forward. I think they’re going to be really rich user interfaces”…” with e-commerce, you want a very rich user interface. It would be agentic. You can have a conversation with it, but the point is that it has to be more visual.

And on the other hand..well, yesterday I was having coffee with a friend who works in construction and real estate, and he completely challenged my thesis. For businesses like that, and for wholesalers, purchasing is routine. There’s a budget. The repeat purchases are clear, and the SKUs are clear. Agentic commerce makes a lot of sense for B2B buyers, and it can eliminate much of the corruption and handling fees that currently pile up in those flows. It can also optimize pricing for exact ingredients or raw materials based on data, rather than on the moods and preferences that drive consumer consumption. That actually makes sense when he pointed it out to me. Now the question becomes who can build that plug-in layer for companies to adopt.

Goldman Sachs’s Decoding the Agentic Economy report estimates enterprise AI agents could lift global token consumption 24x by 2030 and 55x by 2040. That includes enterprise adoption of agentic commerce.

An awful double chin photo, shameless promo. Nonetheless, thank you for the invite, SocGen.

Which connects to what I keep saying on public forums and panels: AI adoption isn’t actually happening at large scale yet, and it isn’t here to replace jobs. Jobs are many tasks bundled together, and AI can do some of those tasks well. What we’re in is early-stage work-enhancement technology, closer to email than to factory automation, as Goldman illustrates. That also means agent adoption within workflows is nowhere near where it could be. My bet is the next stage shows up in wholesale and B2B sales: better efficiency, fewer mistakes, less corruption, and the tedious middle-manager work compressed down to one or two people checking in between purchase and delivery.

The merchant side may matter more than the consumer side

Merchants already pay Alibaba for traffic, for tools, for services. If AI helps them run their business better, the reason to pay is even more obvious.

Wukong is Alibaba’s AI-native enterprise agent for merchants. Accio, which helps merchants with industry research and ad work, has already taken off. Adoption details are still thin, but the use cases write themselves: product listings, image and video generation, customer service, ad campaign optimization, inventory analysis, promotion design, review management, and figuring out what’s actually likely to sell.

This is where AI produces concrete ROI. A small merchant doesn’t have a full marketing team, design team, copywriting team, and data analyst. If AI helps that merchant create better product pages, better ad creatives, and better customer responses, Alibaba turns AI into a productivity layer for millions of sellers.

Same logic on Alibaba.com’s Accio and Accio Work. Cross-border sourcing is information-heavy, fragmented, and painful: find suppliers, compare prices, understand certifications, negotiate, translate, manage documents, coordinate logistics, handle follow-up orders. Which tbh is really a textbook agentic use case. If Accio Work moves from “AI search for suppliers” to “AI operating system for global trade,” that’s a cleaner monetization path than a consumer chatbot.

Tencent: AI as the invisible upgrade

So we’re all anticipating Tencent’s Penguin reveal next week, but beyond that hype, it’s no secret that Tencent’s management has felt pressure to justify why they aren’t doing that well on AI. A year ago it looked like a cautious approach. Now it kind of looks like being left behind. They’ve been on a hiring spree to build proprietary LLM capabilities, but the core dilemma (to integrate AI into WeChat, or not to) is still haunting them.

At the annual shareholder meeting, Pony Ma addressed the “is Tencent AI falling behind” question directly: “A year ago, we thought we were on board, but then we realized the ship was leaking. Now we feel like we’ve managed to get on, but we can’t even sit down properly. We’re still hoping the ship could go a bit faster.” Ma noted Tencent didn’t have particularly strong foundational AI capabilities early on, but is addressing it through talent, team management, and internal training and reassuring investors that they’re gradually getting on track.

Tencent’s AI strategy from the start was less obvious from the outside because it isn’t really selling AI as a standalone product. What we gave kudos to them for was embedding AI into the businesses that already make money and being open to their non-proprietary models. But that’s becoming simply not enough…

Anyway, its near-term justifications have all been about how AI has bolstered existing revenue streams. Ads are the cleanest example of them doing a good job at leveraging AI to enhance $. Marketing services revenue grew 20% YoY, helped by AI-driven ad recommendation upgrades and stronger closed-loop marketing inside Weixin. That’s the near-term ROI. If better models improve targeting, bidding, creative optimization, and conversion prediction, advertisers spend more. Tencent doesn’t need consumers to pay RMB20 a month for an AI assistant. It needs advertisers to see higher ROI.

Tencent also disclosed that AIM+, its automated campaign management tool, powered about 30% of total marketing services spending. AIM+ is essentially Tencent’s version of the AI-assisted ad-buying layer: instead of advertisers manually tuning every campaign, the system optimizes toward performance goals. Structurally similar to what Meta has been doing with Advantage+: the platform absorbs operational complexity, advertisers provide budget and objectives, AI does the optimization. And this brings us back to its superpower. Tencent has the unfair advantage that Weixin, Video Accounts, Official Accounts, Mini Programs, Mini Shops, and payments all live inside one ecosystem. The more closed-loop the transaction path, the better the AI can measure and optimize.

Tencent’s e-commerce path is social, not search-first

Alibaba’s e-commerce AI strategy starts from shopping intent. That’s always been Taobao’s DNA. Tencent’s starts from social attention, which is WeChat’s DNA, and its ecosystem remains its stronger differentiator in agentic AI, as it remains the most natural gateway for agentic tool access.

In recent months, they’ve launched several AI products: Hunyuan, Yuanbao, CodeBuddy, WorkBuddy, QClaw, and more. Nothing has been adopted widely yet.

In Alibaba’s world, users go to Taobao or Tmall because they want to buy something. AI helps them find the product and complete the transaction.

In Tencent’s world, users are already inside Weixin: chatting, reading, watching, following creators, searching, joining groups, scanning QR codes, using Mini Programs. Commerce emerges from that activity. Furthermore, Mini Programs can potentially evolve into modular ‘skills’ that AI agents invoke to execute real-world tasks in WeChat. Without installing separate apps- and we all know that mental hurdle.

The puzzle piece not to overlook is Mini Programs and Mini Shops. And we even signaled this last time we wrote about Tencent.

Mini Shops GMV continued to grow rapidly. Tencent added brand merchant incentives, coupon sharing for repeat buyers, and creator-to-merchant matchmaking. That’s social commerce infrastructure, and AI can improve every part of it: content recommendation, creator matching, product targeting, search ranking, ad optimization, customer service.

Video Accounts time spent grew more than 20% YoY, helped by recommendation model upgrades. Weixin Search query volume grew more than 25% YoY, helped by LLM-powered ranking and broader AI search coverage. These aren’t just engagement numbers; they can be interpreted as commerce infrastructure numbers. Because more time in Video Accounts means more distribution for creators and merchants. Better Weixin Search means easier discovery of products, services, content, and Mini Programs. Better AI ads mean more profitable advertiser spend. Growing Mini Shops mean more transaction-related service revenue.

Tencent’s e-commerce strategy is not to become Taobao, but to make Weixin/ WeChat a better place for (agentic) commerce to happen naturally.

The real Tencent strategy: agents inside WeChat

So we ended the last section on this. Tencent embraced OpenClaw; it pushed out its own version, blablabhlah. Yes, because Tencent’s AI direction is agentic AI.

Tencent has Hunyuan, Yuanbao, CodeBuddy, WorkBuddy, QClaw. It also has Weixin, QQ, WeCom, Mini Programs, QQ Browser, payments, and cloud infrastructure. The plan isn’t to build one chatbot. It’s to make AI agents useful across the whole ecosystem.

This is where Tencent’s distribution edge really helps because a standalone AI agent has to ask for everything: permissions, integrations, APIs, user trust. Tencent already has the user relationship, the chat interface, the payment rails, the enterprise messaging layer, and a massive Mini Program ecosystem.

The most important framing from Tencent is that Mini Programs can eventually be “skills” that agents call. A Mini Program is already a lightweight app inside Weixin: book services, sell products, process payments, manage memberships, handle deliveries, connect to offline merchants. If agents can call Mini Programs as tools, Weixin becomes an agent execution environment.

That means the user doesn’t just ask a chatbot for advice. The agent eventually helps complete the task. Book the appointment. Order the product. Apply the coupon. Track the delivery. Summarize the group chat. Create the Mini Program. Generate the ad campaign. Pull the document from WeCom. Send the follow-up.

That’s the agentic version of WeChat. (More on this in QClaw’s overseas push.)

Which is why Tencent shouldn’t be judged only on whether Hunyuan beats Qwen or DeepSeek on benchmarks. Benchmarks matter, and they’re trying to play catch-up, but it’s not the whole story. Tencent may need the world’s best model if it has the most useful execution layer; it needs the best, most efficient model that is made natively for its use cases.

And this brings me back to something we wrote nearly a year ago as Alibaba is trying to create a better AI shopping experience. If executed correctly, Tencent may actually create the more powerful AI operating environment.

---