As of November 9, 2026, public information shows that this year's Singles Day produced a split result that matters more than the headline number. Syntun data put transaction volume across major Chinese platforms at 1.0263 trillion yuan for 7 to 31 October, up 2.4% year on year, a modest figure by the standards of previous years. Inside that flat total, AI-related categories moved sharply. Tmall reported a 450% month-on-month rise in token-category transaction value on the first day of its promotional period. JD.com said sales of cloud token products climbed 167% month on month through 30 October, and token packages from more than ten model providers sold 20 times their normal volume within two hours on a 3C livestream. Alibaba said 100,000 monthly memberships for its QwenWork office AI service, priced at 0.01 yuan each in a Taobao Live session, sold out in 30 minutes, equivalent to roughly two trillion tokens.
The numbers behind a flat headline
Read together, these figures describe a shift in what the promotional calendar is selling, and they line up with a broader pattern in category growth. When consumer demand for a category is still forming, promotional spend buys trial rather than share, which is exactly the stage AI capacity is at now. Market size estimates for the category vary widely, so the useful signal is not the absolute number but the direction: demand is coming from buyers who already have a use case, not from curious browsers.
Set against that, the discount mechanism itself is losing leverage. The discount mechanism itself is losing leverage: in the same Syntun survey, about 70% of respondents said the festival loses its appeal if it offers nothing more than price cuts, and more shoppers said they now weigh authenticity guarantees, quality, price protection and after-sales service alongside price. When the discount stops being the differentiator, the platform has to sell something else. This year that something was access to computing capacity.
Why token bundles work as a promotional item
A token bundle behaves unusually well inside a shopping festival. It has no inventory, no logistics and no returns. It can be discounted to a near-zero entry price without destroying the perceived value of the underlying product, because the buyer is paying for future usage rather than a physical good. It also converts a one-off promotional visit into an ongoing relationship: a monthly membership renews, a prepaid credit balance gets drawn down, and both create reasons to return to the app. That is a structurally different proposition from discounting a television.
The comparability problem nobody has solved
There is a catch, and it is a commercial one rather than a technical one. Tokenisation differs across models: the same Chinese sentence can be split into different numbers of tokens by different providers, so a package advertising an identical quota does not represent an identical amount of usable output. Without a common measurement standard, cross-platform price comparison is unreliable, and consumers cannot easily tell whether 20 million tokens from one provider is better value than 15 million from another. Platforms benefit from that opacity in the short term. In the medium term it caps the category, because shoppers who cannot compare tend to default to the cheapest headline number.
What the category mix now looks like
The interesting movement is not that AI products sold well, because novelty categories always sell in their first festival. It is where they sold, and what they displaced in the promotional hierarchy. AI capacity now occupies the slot that used to belong to consumer electronics as the category that gives the campaign a growth narrative, while the categories that carry real volume stayed flat and kept their discount depth. That reshuffling is the part worth planning for, and it changes which category gets the prominent placement next season.
| Category | Signal this season | Driver | Durability |
|---|---|---|---|
| AI tokens and memberships | 450% month-on-month value growth on Tmall day one | Near-zero entry pricing and renewals | Depends on standardising quotas |
| AI hardware | AI PC value up more than 300% in the first hour | Replacement cycle plus local inference demand | Tied to device refresh cycles |
| Core digital goods | Sustained, low-growth baseline | Habitual purchase | Stable |
| Apparel and beauty | Still the volume base | Seasonal need and gifting | Cyclical |
The second row deserves attention because it explains the first. AI hardware and AI subscriptions reinforce each other: a buyer who purchases a device capable of running models locally has a reason to buy token capacity, and a buyer who pays for capacity has a reason to upgrade hardware. That loop is what makes the category more than a promotional stunt. It also means the demand is concentrated among buyers who already have a use case, which limits how far it can spread in a single season.
How promotional rules should adapt
For brands and platforms alike, the practical question is how to price a product with no unit cost, no inventory and an unclear unit of measure, in a calendar built around percentage-off mechanics. Discounting a quota communicates nothing about value, because the buyer has no reference point for what a million tokens produces. The answer is to stop pricing the quota and start pricing the outcome, so that the comparison the shopper makes is between two useful results rather than two numbers on different scales.
| Pricing approach | What the buyer sees | Risk | Best fit |
|---|---|---|---|
| Quota-based | A number of tokens at a discount | Buyer cannot compare across providers | Early adopters |
| Outcome-based | A defined task, such as hours of transcription | Provider absorbs usage variance | Mainstream buyers |
| Bundled | Capacity attached to a device or service | Attribution of value is unclear | Hardware attach |
| Subscription with cap | Monthly allowance at a fixed fee | Overages create bill shock | Regular users |
Outcome-based pricing is the most defensible because it removes the comparability problem from the buyer's side. It also shifts the risk of inefficient tokenisation onto the provider, which is where the incentive to optimise belongs. The trade-off is margin: a provider that prices by outcome has to absorb the variance in how much compute each task consumes. For a category still establishing itself, that is a reasonable price to pay for clarity.
What to do before the next campaign
The operational lesson is to treat AI capacity as a merchandising category with its own rules, not as a digital coupon with a longer description attached to the checkout page. That means giving it a comparable unit, a plausible renewal path and a place in the category review, where it will compete for space against lines with known elasticities. The five steps below are ordered that way, from how the offer is presented to how the revenue is reported after the campaign closes.
FAQ
Is the AI category big enough to change the promotional mix?
Not yet in absolute terms. The headline total grew 2.4%, so AI capacity is still small. What changed is its role: it gives the festival a growth story when discounting no longer produces one.
Why can token packages be discounted so heavily?
Because there is no inventory cost and no returns. The discount buys a subscription relationship rather than clearing stock, so the economics are closer to a software trial than to a product markdown.
What is the main barrier to growth?
Comparability. Tokenisation differs by provider, so identical quotas are not identical amounts of usable output. Without a common unit, buyers default to the cheapest headline number.
Should brands price by quota or by outcome?
By outcome where possible. It removes the comparison problem for the buyer and puts the incentive to optimise usage on the provider, at the cost of some margin variance.
