Why Is ByteDance Thriving While Alibaba Struggles in the AI Race?

In the first half of last year, I went all-in on Alibaba.

The reasons were solid: cloud + models + chips + application ecosystem — among China's major tech giants, Alibaba has the most complete setup, comparable to Google in the US. With the AI wave coming, there was no reason for it not to win.

But from the second half of last year until now, I've grown increasingly disappointed.

---

How did this disappointment come about?

1. Tool-calling capability is shockingly poor

Several of my projects initially used the Qwen API. Qwen's early foundational model capabilities were indeed good.

But after the Agent trend took off, Qwen exposed its real weakness — its tool-calling capability is shockingly poor, far worse than MiniMax and GLM. Its reputation doesn't match its actual performance.

2. DingTalk hasn't adapted to a change as significant as OpenClaw

OpenClaw has been out for a long time now. Feishu natively integrated it, that goes without saying. Even WeChat has already integrated it. Yet DingTalk, which focuses on the B2B market, still shows no movement. Alibaba's response speed leaves me speechless.

3. Alibaba is absent from the battle for the super-app gateway

Fighting for the AI super-app gateway is like scrambling for a ticket to the next era.

Domestically, Doubao is far ahead; globally, ChatGPT dominates, with Gemini catching up fast.

But Alibaba, a "top player," hasn't even beaten DeepSeek, let alone Doubao.

The Qwen app being hard to use is one thing, but it hasn't improved much over such a long time — responsiveness, voice, interaction, it's comprehensively behind Doubao.

Many automakers working on in-car intelligence have switched their preferred partner from Alibaba to ByteDance. The daily call volume for Doubao in vehicles is already quite substantial.

4. The Spring Festival Gala sponsorship revealed a lack of strategic resolve in one detail

This year's Year of the Horse Spring Festival Gala sponsorship rights were given to Doubao.

This isn't about money; it's about competing for the national mindshare in the AI era. Qwen just gave it up.

Instead, I saw Qwen ads in elevators, featuring an unfamiliar tennis player as the spokesperson, monotonous and repetitive, with questionable effectiveness.

Before the Spring Festival, they ran a "download the Qwen app for food delivery subsidies" campaign — many people around me said they uninstalled it right after claiming the subsidy. They boosted daily active users, hit their KPIs, but users didn't stick around at all.

5. Chasing leaderboards for KPIs isn't building a product

The recently hyped video generation model Happy Horse ranks high on leaderboards, but actual user feedback is very poor.

Chasing leaderboards, gaming the data, hitting KPIs and calling it a day — is this the state of Alibaba's AI team now?

Coupled with the recent major reshuffle of the R&D leadership, it's almost certain there are major internal problems.

---

Where's the problem?

My assessment: The core decision-makers don't have a deep enough understanding of actual AI business, and have been misled by subordinates with pretty leaderboard rankings and KPIs.

Resource allocation is misdirected, agentic capability development is lagging, and execution has severely deviated.

Alibaba's core decision-makers are more familiar and skilled with new retail. They understood the food delivery wars, but they might not have really figured out how to fight this AI battle.

---

So, reduce the position?

I'm still weighing it.

Alibaba's cloud foundation and infrastructure still hold value and won't disappear in the short term.

But if the investment thesis was "China's Google in the AI era," these recent signals are already loosening that logic.

I might adjust the position, but there's no rush — still observing.