Key Takeaways

- DeepSeek's annualized revenue hit $400-500 million, primarily from cloud API sales
- The company seeks to raise $7.4 billion at a $74 billion valuation, roughly 148x revenue
- Gross margins of 70-80% persist despite pricing 27x cheaper than OpenAI's comparable models
DeepSeek, the three-year-old Chinese AI company that rattled Silicon Valley with its low-cost models, has built an annualized revenue base between $400 million and $500 million. The company now plans to raise $7.4 billion at a $74 billion valuation ahead of a Shanghai STAR Market listing, according to The Information.
The bulk of that revenue comes from API sales. Enterprises pay for cloud-based access to DeepSeek's models, and at pricing that remains a fraction of U.S. competitors. In February 2025, DeepSeek charged $2.19 per million output tokens versus OpenAI's $60 for its o1 model. Nearly 27 times cheaper.
How does DeepSeek maintain 70-80% gross margins at those prices?
Infrastructure improvements. The company has figured out how to process more queries with fewer chips, reducing the per-query cost of running its V4 flagship model. This is the core of DeepSeek's pitch: you can build and operate frontier AI models without the massive compute budgets that OpenAI and Anthropic require.
The numbers suggest the pitch holds up. A 70-80% gross margin on API access, while charging dramatically less than competitors, implies either vastly more efficient inference or significantly lower training costs. Likely both. DeepSeek claimed its V3 model cost roughly $5.6 million to train, compared to the hundreds of millions spent by U.S. labs.
What does a $74 billion valuation mean for AI investors?
At 148 times annualized revenue, DeepSeek's target valuation sits in rarefied territory. For context, OpenAI's October 2024 funding round valued it at roughly 100x revenue. The premium likely reflects two bets: that DeepSeek will scale revenue quickly and that its efficiency advantages compound as the company grows.
The company has hired investment banks to prepare for the STAR Market listing. It aims to file this year and complete the IPO in 2027. Middle Eastern investors are a particular target, The Information reported.
This marks a sharp turn for founder Liang Wenfeng. The Wall Street Journal noted he previously resisted external capital and prioritized frontier research over commercialization. The pivot suggests that even open-source AI developers cannot escape the capital demands of compute.
Enterprise adoption despite security concerns
Businesses have adopted DeepSeek's models because the cost savings are hard to ignore. But national security and data privacy concerns persist. Chinese law requires companies to cooperate with state intelligence requests, creating potential exposure for enterprises using DeepSeek's cloud APIs.
The workaround: self-hosting. Companies can run DeepSeek's open-weight models on their own infrastructure, avoiding data transmission to Chinese servers. Consumer use of DeepSeek's apps creates greater data-sharing risks. Enterprise finance teams evaluating the cost savings should factor in the compliance overhead of self-hosting versus the risk profile of cloud APIs.
Why this matters for fintech and finance teams
AI inference costs directly affect the ROI of any automation project. A model that performs at GPT-4 levels but costs 27x less changes the math on document processing, fraud detection, customer service, and dozens of other finance use cases. DeepSeek's efficiency gains are not academic.
But the geopolitical dimension adds friction. Finance teams operate under strict regulatory requirements. Using a Chinese AI vendor, even a cost-effective one, introduces compliance questions that U.S. or European alternatives do not. The question is whether the cost delta justifies the additional due diligence.
Logicity's Take
DeepSeek's numbers force a difficult conversation for finance teams. The 27x cost advantage is real, and the 70-80% gross margins prove it is sustainable. But the compliance burden of self-hosting may eat into those savings. Teams evaluating AI inference providers should benchmark DeepSeek's V4 against Anthropic's Claude, OpenAI's GPT-4o, and Google's Gemini on actual finance workloads. The winner depends on your risk tolerance and your data residency requirements.
Frequently Asked Questions
What is DeepSeek's current revenue?
DeepSeek's annualized revenue recently reached between $400 million and $500 million, according to The Information. The majority comes from API sales for cloud-based model access.
When will DeepSeek go public?
DeepSeek plans to file for an IPO on Shanghai's STAR Market this year and complete the listing in 2027. The company has hired investment banks to prepare the offering.
How much is DeepSeek trying to raise?
DeepSeek seeks to raise 50 billion yuan, approximately $7.4 billion, at a valuation of 500 billion yuan, or roughly $74 billion.
Why is DeepSeek so much cheaper than OpenAI?
DeepSeek's infrastructure improvements allow it to process more queries with fewer chips. The company claims to have trained its V3 model for about $5.6 million, versus hundreds of millions for competitors, and maintains 70-80% gross margins despite lower prices.
Is it safe for enterprises to use DeepSeek?
Using DeepSeek's cloud APIs creates data privacy exposure under Chinese law. Enterprises can reduce risk by self-hosting DeepSeek's open-weight models on their own infrastructure, though this adds operational complexity.
Earlier coverage of DeepSeek's IPO plans with additional context on the STAR Market listing process
Need Help Implementing This?
If your team is evaluating AI inference providers for finance workloads, we can help you benchmark costs, assess compliance requirements, and build a vendor selection framework. Reach out to Logicity for a consultation.
Source: PYMNTS | / PYMNTS
Huma Shazia
Senior AI & Tech Writer
Produced with AI assistance and reviewed by the Logicity editorial team. Learn more in our Editorial Policy.






