Why DeepSeek Skipped the Chinese New Year Red‑Packet Rush: A Cost‑Benefit Analysis

Rational Choice Under Two Constraints

During Chinese New Year, major AI model providers launched large red‑packet campaigns to acquire users, while DeepSeek quietly released a 1 M token context window.

Capital Constraint: User Net‑Value Model

For an AI service, acquiring a user can be expressed as:

Net Profit = Revenue_from_user – Inference_cost – Customer_Acquisition_Cost (CAC)

Red‑packet users often generate near‑zero usage after the coupon is spent, resulting in low or negative net value. Large firms can offset this loss with other ecosystem revenue, but a capital‑constrained company like DeepSeek would likely incur negative overall profit.

DeepSeek’s parent, Huanfang Quantitative, manages roughly ¥70 billion in assets, with projected 2025 earnings in the tens of billions. By contrast, ByteDance and Alibaba plan AI‑related capital expenditures in the trillions, indicating a large gap in funding efficiency for user acquisition.

Compute Constraint: Inference vs. Training Resource Competition

Assume total computable capacity C, with a portion T continuously allocated to training. The remaining capacity for inference is C – T. Because training and inference clusters differ in hardware and scheduling, the relationship is directional rather than a precise subtraction.

If N new users each consume P compute units per month for inference, the constraint is: N × P ≤ C – T DeepSeek is still developing a new architecture (1 M context window, next‑generation training). The leftover inference capacity is limited, so large‑scale user acquisition would crowd out training resources and slow model iteration.

User Quality Segmentation

Users can be grouped by expected net value:

Low‑value “grab‑and‑go” users – ~70% of users, net value negative or negligible.

Light‑usage retainers – ~25% of users, low but positive net value.

High‑frequency paying users – ~5% of users, relatively high net value.

When the proportion of low‑value users is high, the aggregate expected net profit is severely reduced, regardless of the exact percentages.

Different Objective Functions, Different Rational Behaviors

Doubao (ByteDance) optimizes for monthly active users (MAU) to feed its traffic‑driven business.

Yuanbao (Tencent) optimizes for social penetration within the WeChat ecosystem.

DeepSeek’s public behavior—continuous paper publishing, open‑source releases, no subscription model, very low API pricing—suggests a research‑driven objective rather than user‑scale KPIs. Its parent’s capital support allows operation without monetizing C‑end users.

Limitations of the DeepSeek Path

Funding risk : Ongoing reliance on Huanfang’s capital injections is vulnerable to market fluctuations affecting the parent’s returns.

Consumer moat : Individual users face low migration costs; without ecosystem lock‑in, products can lose users quickly. Enterprise/API users have higher switching costs.

Open‑source cost : Open‑sourcing builds community trust but also enables competitors to iterate on the same codebase, reducing commercial barriers.

Long‑term viability depends on delivering noticeable technical leaps in each major version. Sustained technical advantage would preserve developer goodwill and justify continued funding; a stall would leave DeepSeek at a disadvantage compared to capital‑rich rivals.

Note: Financial figures for Huanfang Quantitative are based on public reports and serve only as magnitude references; user‑segment ratios are illustrative assumptions; the model is a simplified framework and does not constitute financial advice.