Part III: The Spring Festival Gala Is China’s AI Super Bowl
battle of agents, tools, multimodal workflows, and infrastructure, turning prime time TV into habit loops
Ok, so now we’re finally at the last piece of this series.
There have been headlines everywhere about “春节AI模型大战,” the Spring Festival AI LLM showdown. But despite all the newly released models, it didn’t feel like a benchmark war. It felt like showmanship and real-life usage demos, engineered around the one moment in China’s calendar where attention is both concentrated and national.
[Ethan Choi wrote up an incredibly detailed post on China’s AI labs, models and capabilities]
Earning trust
I’m going to state something obvious that people still underestimate. The CCTV Spring Festival Gala isn’t “just a show.” It’s the Super Bowl for Chinese businesses.
There’s a reason companies fight for sponsorship, product placement, and a 5-second close-up of a business leader. That kind of screen time that isn’t even an ad. It’s signaling. It’s reach. It’s credibility and legitimacy. It’s a serious flex.
Being part of the show actually sends a message to the people (老百姓s - the average uncle and aunties) that says: this brand is trustworthy (and, for that matter, this technology is trustworthy).
And the Spring Festival compresses real life into one performance after another. The goal is for it to be relatable for everyone: from grandkids visiting grandma to farmers harvesting to industrial robots on the assembly line. These are intentional; nothing is more intentional than the way they planned the themes for the gala. And because everything is so deliberate. How AI is showcased was also deliberate.
Stress test
But as much as the Spring Festival is a big celebration. It’s an opportunity for communities to get together offline, but often then that sharing extends online. So in that way, the modern-day Spring Festival also becomes a stress test for the nation’s infrastructure, digital infrastructure, the economy, and everyone’s sanity.
Mind you, you dont just go back to your parents’ home (with all your aunties, uncles, and cousins) for ONE Thanksgiving/Xmas dinner; you literally go for like a week, at least. The sheer reach is incomparable to any other time of the year.
And for many, it is the only week they get to see their loved ones who may live across the country. Thus, Spring Festival is also the largest human migration in the world each year, and all of that adds up to one of the biggest spending weeks of the year in China.
So if a product can turn holiday chaos into a habit, it wins mindshare after the holiday.
Now, holiday scenarios are inherently multi-step. Family coordination. Travel planning. Meal logistics. Content creation. Social obligations. The kind of “help me do this” that requires a system to decompose a task, call tools, handle exceptions, and still land the plane. It’s not a prompt-response game. It’s a multi-step delivery, something agentic.
And that is why we saw Alibaba, Baidu, Tencent, and ByteDance all go gaga on getting people on their apps over the break with their huge red envelope battles. And the startups all releasing agentic models - Moonshot/ Minimax/ Zhipu.
So if YOUR AI app/tool/product can turn holiday chaos into a habit, it wins mindshare after the holiday. The gala thus becomes a funnel through which hundreds of millions of people are simultaneously exposed to your brand and jump on to try it.
Multimodal has moved from “wow” to “control.”
For the past year, the multimodal conversation has been dominated by wow. “Look how cinematic.” “Look how lifelike.” (Or how slop) “Look how it can imitate.” That’s fine, but it misses what businesses actually need.
Serious and scaled businesses don’t pay for wow. They pay for control and predictability. And as we’ve written about how Kling is courting indie film studios over the year, Seedance’s big reveal shows that AI-made film is actually capable of doing more than “sloppy” dopamine hits.
The shift in this window is that multimodal “control” is starting to matter more than multimodal “wow.” The value isn’t generating something impressive once. The value is generating something repeatable. Being able to reference tone, faces, motion, even music, and reducing prompt roulette enough that output becomes a workflow, not a slot machine.
And the holiday window is useful once again for this to go viral, because it concentrates content creation and sharing. And in this case, no one won the gala more than ByteDance.
ByteDance is the clearest winner
If you want one company that embodies this shift from “model” to “delivery,” it’s ByteDance. Whether it was during the gala or offline, they won CNY.
When a platform tries to pull AI into the mainstream through a Super Bowl-scale event, the model itself isn’t the only thing being tested. Infrastructure is being tested. Distribution is being tested. The entire conveyor belt is being tested.
How did the ByteDance stack matter? Well…
Doubao is the consumer surface.
Volcano Engine is the cloud conveyor belt.
Seed2.0 is the “doer” layer positioned around tool reliability and orchestration.
Seedream and Seedance are the multimodal engines that turn AI into content workflows.
As the official AI partner to the Spring Festival Gala, Doubao secured a prime marketing position and was cited as having logged 1.9 billion interactions, as ByteDance leveraged the nation’s most-watched broadcast to accelerate mass adoption. Volcano Engine served as the cloud partner for the 2026 Gala, but Doubao eclipsed it in visibility through repeated on-screen placements and prominent content integration.
You could see it everywhere: red envelope game moments, MC shoutouts, and quite literally powering all the AI visual art. This didn’t feel like a passive “product placement.” It felt like ByteDance turning the Gala into a distribution surface for an interactive behavior.
Then you see the same playbook spill even into offline life. I went to an offline lantern festival, and if you purchase the entry ticket on Douyin, it’s half off. The discount itself isn’t the story. The company really captured all touchpoints. The story is the habit loop. They’ve maxxed out on distribution online and offline.
And on top of all this, Seedance took home the cup. During the CNY break, the latest Seedance 2.0 drop shocked the world with its multimodal model, and it reinforced the ByteDance thesis in the most direct way: when a platform with that kind of reach wires AI directly into creation workflows, adoption doesn’t need a traditional go-to-market battle. It rides the existing creator stack and distribution flywheel.
That’s also the business model implication people often miss. For ByteDance, models aren’t just features. They’re levers that expand inventory. Cheaper, faster creation increases the supply of content minutes. More minutes expand monetizable attention, whether through ads, commerce, or creator monetization. In this vertical, “winning” is less about who has the best prompt and more about who owns the loop from creation to distribution to feedback to monetization.
So what now?
So what does all of this imply?
It implies that the big tech competition is increasingly a delivery competition. Not “who is smartest,” but who can deliver outcomes at scale, reliably, and cheaply enough that users come back after the holiday glow fades.
It also implies that the societal layer is not optional if you want to scale. When the Spring Festival Gala becomes an AI demo day, AI gets normalized not through policy memos but through mass entertainment. People called the 2026 CCTV Spring Festival Gala “首个AI春晚,” “First AI Spring Festival Gala,” and that phrasing captures the dynamic where the shoecase actually normalizes AI at the national scale.
If Part 1 of this series was about how China can close the capability gap, and Part 2 was about why monetization is brutal in an involutionary market, then Part 3 is about the actual playbooks these AI companies took during the holiday and how they leveraged the gala.
So over the next 90 days, I’ll be curious to see whether the holiday habits translate into retention. Whether agent workflows break less often as iteration loops tighten. Whether multimodal control becomes commercially repeatable. Whether the infrastructure layer proves to be the real conveyor belt behind adoption.
Because if the Gala is China’s AI Super Bowl, then the real winners aren’t the ones who ran the flashiest ad. They’re the ones who got you to talk about them months later again and again, and the apps that got you to come back the next day, and the day after that, when it’s no longer a holiday red envelope frenzy.
Alright, I’ve hit my CNY quota, guys. Let’s move on from the Spring Festival now.
Part I: The Gala, the Suburbs, and the “Months Behind” Myth in LLM Labs
Hello! Over the last week, X has gone into a frenzy about the CCTV Spring Gala’s robotics shows. And tbh they were truly impressive in their agility, flexibility, and mobility. I’m resurfacing some old robotics pieces I’ve written and put under “Physical AI” on AI Proem. But I’m not going into details about that today, because even though these robots a…
Other relevant reads:
Arrival of The Horse. China’s Two Viral AI Drops Explained
Hi all, there are six days until Lunar New Year eve, and most of East Asia is on holiday, except for the AI labs. Like one by one is rushing to get something out.
Rise of China's Robotics Industry: from Manufacturing Arms to Embodied AI
Hi all, as I mentioned, I finally got my Manus AI invite, and I was stoked to try it out. Manus’s research capabilities supported this deep-dive piece, and I must say it's comprehensive and efficient. Even though the pro version of OpenAI limits deep research to twice a day (I guess that’s good; it limits people’s abuse of it), Manus was quite impressiv…












Curious whether there are cultural differences between the US and China approaches to advertising, in this respect: were there any advertisements by Chinese AI/tech companies during the Spring Festival that tried to "diss" their competitors, the way we saw during the US Super Bowl from Anthropic, directed at OpenAI? Or did Chinese companies focus only on their own strengths and value added across the consumer value chain?