The artificial intelligence lab is finalizing a funding round of $7.5 billion that would value the company at roughly $74 billion by the end of October. Annual recurring revenue for DeepSeek recently surpassed $1 billion, a sharp increase from less than $500 million a few months ago. Demand for services remained strong even after CEO Liang Wenfeng raised API prices between 2.3 and 4.5 times, helping the API business reach an 82.9% gross margin through July.

To manage resource constraints, the company allocates 70% of its compute capacity to training and 30% to inference. DeepSeek uses Nvidia RTX 5090 gaming GPUs to run inference for smaller models, which frees up high-end hardware to ease the shortage of training resources. This approach has contributed to tightening retail availability of high-end consumer GPUs.

Sign in to suggest edits

Key sources

  1. SOURCE@jukan05“DeepSeek allocates 70% of its compute capacity to training and 30% to inference”x.com
  2. SUPPORT@jukan05“RTX 5090s have dried up in the retail market”x.com
  3. SUPPORT@choblin29“raised API prices by as much as 2.3x–4.5x, yet Liang Wenfeng told investors its customer count didn’t fall”x.com
  4. SOURCEmarketbrief.now
  5. SUPPORT@rohanpaul_ai“Liang told investors that DeepSeek’s internal tests show its smaller models can run well on graphics chips designed for gaming”x.com
  6. SOURCEhuggingnewshuggingnews.com
  7. SOURCEmarketbrief.now
  8. SOURCEhuggingnewshuggingnews.com
Markdown