DeepSeek AI Statistics and Facts (2025)

Though often overshadowed by US companies like OpenAI, DeepSeek AI exploded onto the international scene in early January 2025 with its large-scale, cost-efficient models.

With its roots in Chinese quantitative finance, it focuses on efficiency and open-source innovation, drawing attention from around the world.

In this article, we present key statistics and facts about DeepSeek’s rapid rise and examine how it stands against dominant American AI players.

What Is DeepSeek AI?

DeepSeek AI is an independent artificial intelligence research lab operating under the umbrella of High-Flyer, a top Chinese quantitative hedge fund.

Founded in 2023 by Liang Wenfeng, DeepSeek has focused on building advanced large language models (LLMs) and coding tools—prioritizing pure research, openness, and algorithmic efficiency over immediate commercialization.

Top 14 Key DeepSeek AI Statistics and Facts

• Pricing: Priced at 1/30th of similar OpenAI models, costing $2.19 per million output tokens versus OpenAI's 01 model at $60.00.
• App Store: Number #1 in the Apple App Store (27th Jan 2025) - ahead of ChatGPT on the #2 spot
• App downloads: 2.6 million downloads (28th Jan 2025)
• Age: Only 20 month old (founded May 2023 in China
• Team Size: ~200 employees at Deepseek vs. OpenAI’s 3,500
• Model Scale: Up to 671B parameters (DeepSeek-V3)
• Training Costs: ~1/10 of comparable Western models (Wired, Jan 2025)
• Open Licensing: Most models are MIT-licensed (Business Standard, Jan 2025)
• Tech Impact: DeepSeek’s latest AI model triggered a global tech selloff, risking $1 trillion in market capitalization. (Bloomberg, Jan 2025)‍
• Global Recognition: January 2025 launch drew international attention (and caused a 13% pre-market drop to Nvidia stock)
• Research Papers: 68 research papers on arXiv
• Chatbot Arena Rank: #4 with a score of 1357, just after ChatGPT-4o on the #3 place
• X.com followers: 349,800 followers (27th Jan 2025)
• Users: Estimated 5-6 million users (Deepseek user analysis)

‍

“Unlike many Chinese AI firms that rely heavily on access to advanced hardware, DeepSeek has focused on maximizing software-driven resource optimization.”
— Marina Zhang, Associate Professor at the University of Technology Sydney

How Many Users Does DeepSeek AI Have?

We estimate Deepseek has an total user-base of between 5-6 million users worldwide based on a cross-data analysis.

But as of 28th January 2025, there is no public data available on the exact number of users DeepSeek AI has. However, based on available Google Play Store download numbers and its Apple App Store rankings (number 1 in many countries as of January 28, 2025), it is estimated to have been downloaded at least 2.6 million times - a number that is rapidly increasing due to widespread attention.

‍

OpenAI vs. DeepSeek AI – What Is the Difference?

Despite both companies developing large language models, DeepSeek and OpenAI diverge in funding, cost structure, and research philosophy.

While OpenAI’s API might cost around $15 per million input tokens and $60 per million output tokens, DeepSeek-R1’s comparable cost is $0.55 and $2.19, respectively.

DeepSeek also employs pure reinforcement learning (RL) in some of its models (like R1-Zero), whereas OpenAI leans heavily on supervised and instruction-based fine-tuning. The result: DeepSeek’s models are more resource-efficient and open-source, offering an alternative path to advanced AI capabilities.

DeepSeek AI Timeline

• May 2023: DeepSeek AI is founded by Liang Wenfeng, transitioning from High-Flyer’s Fire-Flyer AI research branch.
• November 2023: Release of DeepSeek Coder, the lab’s first open-source code-focused model.
• Early 2024: Introduction of DeepSeek LLM (67B parameters) and subsequent price competition with major Chinese tech giants.
• May 2024: Launch of DeepSeek-V2, praised for its strong performance and lower training cost.
• Late 2024: DeepSeek-Coder-V2 (236B parameters) appears, offering a high context window (128K tokens).
• Early 2025: Debut of DeepSeek-V3 (671B parameters) and DeepSeek-R1, the latter focusing on advanced reasoning tasks and challenging OpenAI’s o1 model.

Top 18 Key Statistics About DeepSeek AI

Here's the most important facts about DeepSeek:

1. 1. Founded in May 2023: DeepSeek launched as a spin-off from High-Flyer hedge fund, prioritizing fundamental AI research over quick profit—much like early OpenAI.
   2. Over 100 Billion RMB Hedge Fund: High-Flyer’s peak at ~$15 billion in assets gave DeepSeek robust funding, enabling high-level experimentation without immediate revenue pressure.
   3. 10,000 Nvidia H100 GPUs: DeepSeek preemptively gathered these chips, then focused on software-based efficiency to compete with larger Western labs when export controls tightened.
   4. Predominantly Recent Graduates: Most DeepSeek researchers finished their degrees in the past two years, fostering rapid innovation through fresh perspectives and minimal corporate baggage.
   5. Full Reinforcement Learning for R1-Zero: DeepSeek relies on RL over extensive supervised fine-tuning, producing advanced reasoning skills (especially in math and coding).
   6. Multi-Head Latent Attention (MLA): This subdivides attention mechanisms to speed training and enhance output quality, compensating for fewer GPUs.
   7. Mixture-of-Experts (MoE): Only a targeted set of parameters is activated per task, drastically cutting compute costs while maintaining high performance.
   8. 671 Billion Parameters in DeepSeek-V3: Rivaling top-tier Western LLMs, it still costs far less to train due to DeepSeek’s resource optimizations.
   9. $5.5 Million Estimated Training Cost: DeepSeek-V3’s expenses are much lower than typical for big-tech models, underscoring the lab’s efficient RL and architecture choices.
   10. Distilled Model Variants: “R1-Distill” compresses large models, making advanced AI accessible to those with limited hardware.
   11. MIT-Licensed Releases: DeepSeek grants free rein for adaptation and commercialization, attracting global contributors to improve its models.
   12. $0.55 per Million Input Tokens: DeepSeek-R1’s API slashes costs compared to $15 or more from some US competitors, fueling a broader price war in China.
   13. Major Impact in China’s AI Market: DeepSeek’s price competition forced Alibaba, Baidu, and Tencent to lower their rates, spurring wider AI adoption.
   14. Emphasis on Fundamental Research: Rejecting a pure application focus, DeepSeek invests in “moonshot” strategies, reminiscent of early OpenAI’s bold ambitions.
   15. Countering US Export Controls: Despite chip embargoes, DeepSeek innovates with custom GPU communication and memory optimizations, challenging the policy’s effectiveness.
   16. Patriotic Drive: Researchers often view their work as boosting China’s global AI standing, blending national pride with scientific rigor.
   17. Global Coverage: Wired and Forbes spotlighted DeepSeek’s breakthroughs, validating its model efficiency and open-source approach.
   18. Enormous Future Potential: DeepSeek’s continued push in RL, scaling, and cost-effective architectures could reshape the global LLM market if current gains persist.

FAQ

Is DeepSeek AI free?

DeepSeek’s core models are open-sourced under MIT licensing, which means users can download and modify them at no cost. However, the company does charge for its API usage, with rates starting at $0.55 per million input tokens.

Is DeepSeek a Chinese company?

Yes, it was founded in May 2023 in China, funded by the High-Flyer hedge fund. DeepSeek maintains its headquarters in the country and employs about 200 staff members.

Is DeepSeek AI good?

DeepSeek’s latest model, DeepSeek-R1, reportedly beats leading competitors in math and reasoning benchmarks. With up to 671 billion parameters in its flagship releases, it stands on par with some of the most advanced LLMs worldwide.

Is DeepSeek safe?

DeepSeek hasn’t faced major security controversies, but concerns about censorship may arise given it’s Chinese-owned. The firm says it’s more focused on efficiency and open research than on content moderation policies.

How did DeepSeek achieve competitive AI performance with fewer GPUs?

They adopted innovations like Multi-Head Latent Attention (MLA) and Mixture-of-Experts (MoE), which optimize how data is processed and limit the parameters used per query. As a result, DeepSeek gets more mileage out of its ~10,000 H100 chips.

What makes DeepSeek’s models cheaper to train and use than US competitors’?

DeepSeek leverages reinforcement learning to reduce the need for constant supervised fine-tuning. They also undercut mainstream prices, with APIs costing $0.55 per million input tokens—compared to $15 or more from other providers.

Why does DeepSeek focus on open-source releases despite potential profit losses?

By offering models under MIT licensing, DeepSeek fosters community contributions and accelerates innovation. This approach builds brand recognition and a global user base, often leading to broader long-term opportunities.

How did a hedge fund background influence DeepSeek’s approach to AI research?

High-Flyer’s financial success—at one point surpassing 100 billion RMB—provided ample funding for computational and experimental needs. DeepSeek’s data-driven philosophy also echoes the quantitative mindset behind hedge fund operations.

Could DeepSeek’s efficiency methods reshape US-China tech relations?

Yes, because the firm’s advanced optimizations challenge the effectiveness of chip export restrictions.

• If DeepSeek can reliably produce top-tier models with fewer GPUs, it might undermine assumptions about how to limit China’s AI progress.