- China’s “price killer” prepares major API price hike less than three months after slashing fees
- Surging global demand exposes the limits of cheap AI access amid tightening compute supply
DeepSeek, the Chinese AI startup that shook the global AI market with aggressively low pricing, is preparing to raise API fees significantly, signaling a shift away from its ultra-low-cost strategy as computing capacity struggles to keep pace with demand.
On August 6, DeepSeek announced through its developer platform that it plans to “substantially” increase API service prices in the near future.
The company did not disclose the exact increase or implementation timeline but advised developers and enterprises to plan usage carefully and manage account balances accordingly.
The announcement marks a dramatic reversal less than three months after DeepSeek positioned itself as the industry’s “price killer” by pushing AI model costs to unprecedented lows.
In late May, DeepSeek announced a permanent 75% reduction in API pricing for its V4-Pro model, bringing output costs down to as low as 6 yuan ($0.83) per million tokens.
The move effectively reset expectations for high-end AI inference pricing and intensified global discussions over whether the AI industry was entering a prolonged price war.
Now, DeepSeek itself appears to be stepping back.
Mismatch between demand and supply
The underlying reason is straightforward: demand has outpaced computing supply.
Data illustrates the scale of the imbalance. Between 2025 and early 2026, DeepSeek’s daily active users surged from 120 million to 200 million, a 66.7% increase.
Over the same period, however, its available computing capacity grew by only 8.3%.
The mismatch has already affected service reliability. DeepSeek experienced three major service disruptions in May alone.
When the V4 preview version launched, a surge of 1.7 million new users overwhelmed the system, causing widespread outages that took hours to resolve.
At the same time, usage from developers has exploded.
Global AI model aggregation platform OpenRouter data showed that DeepSeek V4 Flash ranked first in weekly model usage, consuming 36.5 trillion tokens in a single week.
Developer platform OpenCode reported that on August 1 alone, DeepSeek V4 Flash processed 8 trillion tokens through its platform.
DeepSeek previously introduced peak-and-off-peak pricing in July, doubling prices during high-demand periods. But the adjustment was apparently insufficient.
The planned broad-based price increase suggests the company is officially moving away from its extreme low-cost positioning.
Industry-wide price increases
DeepSeek is not alone in making such adjustments.
Several leading Chinese AI companies, including Zhipu AI, Moonshot AI and MiniMax, have raised API prices this year.
Zhipu has increased API prices three times within the year, while still seeing usage expand by 400%, suggesting that demand remains strong despite higher costs.
Goldman Sachs research has noted that such pricing adjustments do not necessarily indicate weakening demand.
Instead, they reflect continued appetite for AI services, tightening compute resources and a gradual shift in the industry from aggressive price competition toward more sustainable pricing models.
From price war to compute constraint
DeepSeek’s planned price increase creates a striking contrast with its recent global expansion.
Just one week earlier, V4 Flash had reached the top of global model usage rankings, demonstrating how an extreme cost-performance strategy could rapidly attract developers worldwide.
OpenRouter data showed that nearly half of DeepSeek’s users came from overseas markets, including the United States and Europe.
But the same strategy that helped DeepSeek gain global adoption also created pressure on its infrastructure.
The company’s challenge is no longer simply competing against rival AI models. It is ensuring that it has enough computing resources to serve the demand it successfully created.
DeepSeek’s pricing reversal offers a small but revealing snapshot of the broader AI industry transition.
The market is moving away from an era when companies competed primarily by subsidizing usage and toward a new phase where access to computing power, infrastructure efficiency and sustainable economics determine competitive advantage.
In the AI race, cheap intelligence can attract users quickly — but maintaining that advantage requires the most expensive resource of all: compute.
