DeepSeek’s reported $1 billion annualized revenue surge is primarily due to significant price hikes, especially during peak hours and for certain token types, rather than genuine growth in user volume. While the company manages resource allocation by prioritizing training workloads, many users have migrated to cheaper providers, highlighting that the revenue increase reflects pricing strategy changes more than actual usage growth.
The video discusses the recent surge in DeepSeek’s reported annualized revenue, which crossed the $1 billion mark, and clarifies that this increase is primarily due to a significant price hike rather than genuine growth in user volume. Earlier in the year, DeepSeek was known for its low prices and high-quality AI outputs, attracting many users who shifted their workloads to the platform to save costs. However, in August, DeepSeek quietly implemented a new pricing structure that dramatically increased costs during peak hours and for certain token types, leading to a sudden and substantial rise in user bills.
The key point is that while DeepSeek’s revenue doubled from around $500 million to $1 billion, this was largely driven by price increases rather than an equivalent increase in usage. The company introduced peak and off-peak pricing, with peak hours priced significantly higher, and increased charges on “cash input” tokens, which are heavily used by coding agents and AI development tools. This pricing strategy caused some major users, like the tool Open Code, to reconsider their reliance on DeepSeek’s API due to the steep cost increases.
Despite the price hikes, DeepSeek was not operating at a loss before the changes; the company reported a healthy gross margin of 82.9% earlier in the year. The price increase was more about managing resource allocation and traffic shaping, as over 70% of DeepSeek’s GPU capacity is dedicated to training models rather than serving API requests. The peak pricing discourages heavy usage during China’s office hours, effectively prioritizing training workloads over API demand during those times.
The video also highlights that the price hike was partially rolled back shortly after, with the introduction of a new model version (V4.1 Flash) that reduced some costs, and the attempted retirement of the Pro model was reversed due to user backlash. Additionally, because DeepSeek’s models are open-source under an MIT license, many users migrated to cheaper providers hosting the same models, which means the model’s popularity grew but DeepSeek’s direct API revenue did not necessarily follow.
In conclusion, the video urges users to be cautious when interpreting revenue growth headlines, especially when price changes are involved. True growth comes from increased usage, not just higher prices. For heavy users, DeepSeek might still be the best option due to its cash hit rates, but lighter users should shop around among providers. The $1 billion revenue figure is more a reflection of a fundraising narrative than a pure growth story, and understanding the distinction between price-driven and volume-driven revenue is crucial for anyone relying on DeepSeek’s services.