The company that changed the conversation around the cost of advanced AI is now raising fresh capital as it expands its models, customer base and ambitions for a public listing. 

[For more news, click here]

DeepSeek is approaching a new stage of its growth. The company drew global attention after showing that a relatively young AI developer could produce competitive models without spending at the levels associated with some of the industry's biggest laboratories. It is now nearing a funding round that would value the business at about $74 billion, with roughly $7.4 billion being raised. The round is expected to close before the end of August, while preparations are also underway for a possible listing on Shanghai's Star Market in 2027. 

The scale of the financing marks a notable shift from the way DeepSeek was built. Founder Liang Wenfeng funded much of its early development through his hedge fund, High-Flyer, allowing the company to focus on research without following the usual venture-backed path. As its models gained a much wider audience, the business began moving beyond research into infrastructure, products and commercial services. Its first major external funding round earlier this year valued DeepSeek at more than $50 billion, and the latest round would give it considerably more capital to support that expansion. A potential IPO would take the company another step away from its early identity as a closely held AI lab and toward a much larger technology business. 

Building the Business Behind the Models 

The commercial side of the business is beginning to show some substance. DeepSeek's annual recurring revenue has reached about $500 million, with API access making up a large part of its business. The growth matters because it shows that businesses and developers are paying to use its models inside their own products rather than simply testing them out. Its low-cost models remain an important part of the offering, while newer systems are being developed for more demanding work such as coding, tool use and AI agents. API pricing has also been revised as the company adjusts to growing demand. DeepSeek is increasingly managing its models as commercial services with different levels of capability and use, rather than relying on low prices alone to attract customers. For companies deciding which AI systems to build into their operations, the calculation is increasingly about the combination of capability, reliability and operating cost. 

DeepSeek's efficiency advantage remains important in that calculation, but maintaining it at scale is a different challenge from demonstrating it in a model release. A growing customer base requires more computing capacity, while more capable systems require continued spending on research and engineering. The company has been expanding its workforce and working on its own inference chip as it builds out the infrastructure behind its models. The new capital will help fund that work, while also putting more weight behind whether DeepSeek can continue delivering competitive AI without allowing the cost of running and developing it to rise as quickly as its ambitions. 

Competition is making the timing more important. Alibaba, Moonshot AI, ByteDance and other developers are continuing to improve their own systems, giving businesses more alternatives as they decide which models to use. DeepSeek's early lead gave it a strong position in the conversation around affordable AI, but that position has to be earned again with each generation of models. Its advantage will matter most if developers continue to find that its systems are capable enough for serious workloads and inexpensive enough to justify wider deployment. For enterprise customers, the novelty surrounding R1 matters less than whether DeepSeek can deliver a dependable product at a cost that makes sense over time. 

A potential Shanghai IPO would eventually provide a clearer view of how well that transition is going. A private valuation reflects what investors are willing to pay for expected future growth, while public markets give investors a much closer look at revenue, costs and the pace at which a company is using capital. For DeepSeek, that scrutiny will be particularly relevant because efficiency sits at the center of its reputation. A listing would also give the company another potential source of funding as it expands its model portfolio and infrastructure. The $74 billion figure is therefore best understood as a measure of expectations around what DeepSeek could become, rather than a final judgment on the business today.  

The company has already shown that there is room to challenge the industry's assumptions about the cost of advanced AI, and its growing commercial business suggests that the technology has found paying customers. The funding gives DeepSeek more room to build, but it will also give investors a closer look at whether the company's cost advantage can hold as the business gets bigger. 

Related Articles 

Nvidia Earnings Put the AI Infrastructure Boom to the Test 

Rethinking GPU Cloud Infrastructure as Inference Workloads Rapidly Expand 

VAST Data and AMD Expand AI Infrastructure Alliance to Tackle the Industry’s Inference Bottleneck