DeepSeek has restarted its second funding round after pausing the process in July, according to reports, as the Hangzhou-based artificial intelligence start-up seeks nearly $8 billion in fresh capital and a valuation of almost 500 billion yuan, or about $74 billion. The move marks a renewed push by one of China's most closely watched AI developers to secure the resources needed for large-scale computing infrastructure and to compete in an increasingly expensive global race to build and deploy advanced models.

The fundraising effort had been paused earlier in the summer after remarks the company made to investors were leaked. The company has now returned to the market with a target that would place it among the most valuable privately held AI start-ups in China. The round is not yet complete, and the final size, valuation and investor list could still change. Monolith Management, an investor in Beijing-based Moonshot AI, the developer of the Kimi line of AI models, is reportedly in talks to contribute. The reported interest from an investor with exposure to another prominent Chinese model developer highlights how capital is moving across the country's AI ecosystem as investors look for positions in the sector's leading players.

Funding target and valuation

DeepSeek is seeking nearly 500 billion yuan in valuation, equivalent to about $74 billion or £55 billion. The reported target is a significant step up from the company's first funding round, which concluded earlier in the summer at a valuation of about 350 billion yuan. A jump of that scale would reflect both the rapid growth of China's AI market and the premium investors are placing on developers that can train high-performance models at lower cost than many of their international peers.

The reported raise of nearly $8 billion, or about £6 billion, would give DeepSeek substantial resources at a time when AI companies are spending heavily on compute, talent and data. The company has not publicly confirmed the terms of the round, and the details reported so far are based on people cited in media accounts. Even so, the size of the reported target underscores how quickly capital requirements have risen in the AI sector, where leading developers are competing for scarce chips, energy and engineering talent.

Data centre and compute plans

DeepSeek is seeking capital as it reportedly plans to construct a massive data centre in Inner Mongolia, part of a broader initiative to gain access to 1 gigawatt of computing power. That capacity could include leasing capacity from third parties as well as building its own infrastructure. Some of the funding raised in the second round would reportedly be used to finance the Inner Mongolia data centre.

A 1 gigawatt target is substantial. AI data centres require enormous amounts of electricity, cooling and networking equipment, and access to reliable power has become a critical constraint for companies training and serving large models. Inner Mongolia has attracted interest from technology companies because of its energy resources and cooler climate, both of which can help reduce the operating costs of power-intensive computing facilities. For DeepSeek, the project would represent a move to secure infrastructure at scale rather than relying entirely on external cloud capacity.

Pricing signal

Alongside the funding process, DeepSeek told users last Thursday that it intended a substantial rise in prices, without specifying its upcoming pricing plans. The notice is notable because the company has built its reputation on training high-performance models inexpensively and offering them at low cost. That approach challenged prior assumptions by Western companies that have been ploughing hundreds of billions of dollars into infrastructure for training and deploying models.

DeepSeek currently charges $0.14 per million input tokens and $0.28 per million output tokens. By comparison, Moonshot's latest Kimi K3 is priced at $3 per million input tokens and $15 per million output tokens, while Anthropic's latest Fable 5 is priced at $10 and $50 respectively, according to the reported comparison. The gap illustrates how aggressively DeepSeek has priced its services. Any substantial increase would test how much of its user base is attracted by low prices and how much is drawn by model quality and reliability.

Pricing is a central competitive lever in the AI API market. Developers and enterprises choose models based on a combination of performance, latency, context length, reliability and cost. A company that has competed partly on price can raise rates if it believes its models offer enough value, but it risks losing price-sensitive customers to rivals. The planned increase also comes as AI companies face higher costs for inference, power and infrastructure. DeepSeek's decision to signal higher prices without giving details may be intended to prepare users for a change while preserving flexibility in how it structures new tiers.

Low-cost model strategy under pressure

DeepSeek's low-cost positioning has been a defining feature of its public image. The company has been known for training high-performance models at lower expense and offering access at prices well below those of larger international rivals. That strategy helped it attract attention and users, and it challenged the view that cutting-edge AI would inevitably require enormous capital spending by every participant.

However, the economics of AI are shifting. Training larger models and serving millions of users require substantial compute, and the cost of acquiring and operating infrastructure has risen. Companies that initially subsidized usage to gain market share may need to adjust prices as usage grows. DeepSeek's reported data centre plan and its price signal can be seen as related responses to the same underlying pressure: to compete at scale, the company needs both more capital and a path to covering its costs.

The reported funding round also comes as investors reassess the AI sector. After a period of rapid enthusiasm, venture and strategic investors are paying closer attention to revenue, margins and infrastructure requirements. A company with a strong technical reputation but limited public financial disclosure may face questions about how it will turn low-cost model development into a sustainable business. The second round, if completed, would provide a clearer signal of investor confidence.

Competitive landscape

DeepSeek is one of several Chinese AI developers competing for talent, users and capital. Moonshot AI, the developer of the Kimi models, is among its prominent peers, and the reported interest from Monolith Management shows that investors are building exposure across multiple companies in the sector. China's AI market includes large technology platforms, well-funded start-ups and research-backed teams, all of which are racing to release models and attract enterprise customers.

Globally, the AI funding environment remains intense. Companies such as Anthropic and OpenAI have raised billions of dollars and are investing heavily in data centres and chips. DeepSeek's reported $74 billion valuation would place it in a rarefied group, even if it remains smaller than the most valuable U.S. AI companies. The comparison with Moonshot and Anthropic on token pricing shows that competition is not only about model quality but also about the cost of running models at scale.

The Inner Mongolia data centre plan, if confirmed, would also tie DeepSeek more closely to China's domestic energy and computing infrastructure. Access to 1 gigawatt of computing power, whether owned or leased, would give the company more control over capacity and potentially lower long-term costs. At the same time, large infrastructure projects carry execution risks, including construction delays, equipment shortages and regulatory approvals.

Key facts at a glance

• DeepSeek has restarted its second funding round after pausing in July, according to reports.
• The Hangzhou-based AI start-up is seeking nearly $8 billion, or about £6 billion.
• It is targeting a valuation of nearly 500 billion yuan, about $74 billion or £55 billion.
• Monolith Management, an investor in Moonshot AI, is reportedly in talks to contribute.
• DeepSeek plans a massive data centre in Inner Mongolia as part of efforts to access 1 gigawatt of computing power.
• Some of the new funding would reportedly go toward the data centre.
• The company told users last Thursday to expect a substantial price rise, without giving details.
• DeepSeek currently charges $0.14 per million input tokens and $0.28 per million output tokens.
• Its first funding round closed earlier in the summer at a valuation of about 350 billion yuan.

Source: Silicon UK News