DeepSeek Nears $12 Billion Funding Round Led by Tencent and CATL

What Bloomberg found

DeepSeek is close to closing a round of at least RMB 80 billion (about $12 billion), according to a Bloomberg report published on October 6, 2026, and picked up by TechNode, citing people familiar with the matter. Tencent and battery maker CATL are among the largest investors in the round.

The company was initially seeking about RMB 50 billion, but investor interest pushed commitments beyond that target. Based on term sheets already signed, the final amount could approach RMB 100 billion, though details may still change before the round closes.

DeepSeek did not respond to Bloomberg's request for comment during the Chinese national holiday, according to TechNode. The company plans a corporate restructuring after completing the funding, as preparation for a possible initial public offering (IPO) in early 2027.

Who's putting in the money

Tencent is one of China's largest technology companies, with a presence in gaming, messaging (WeChat), and cloud computing. CATL is the world's largest maker of batteries for electric vehicles. The combination of both names in the check signals that appetite for DeepSeek doesn't come only from the pure AI ecosystem, but from industrial and consumer conglomerates that want a direct stake in the sector.

Neither source details the percentage of the company being sold or the round's implied valuation. TechNode also doesn't say whether other international investors (outside China) are on the cap table, which is relevant given the history of restrictions on foreign capital in cutting-edge Chinese AI.

Why this matters to anyone choosing a third-party model

DeepSeek has established itself as a benchmark among open-weight models, with competitive API pricing. A good share of Brazilian teams running LLMs in production today test or use DeepSeek's API precisely for its cost-benefit ratio, whether through its own API or by hosting the weights on providers such as Together AI, Fireworks, or their own infrastructure.

A round of this size changes nothing in the code of anyone who already integrates the API tomorrow morning. But it's a sign of financial staying power: new money usually translates into more GPU capacity, more research, and, in theory, less pressure to cut token prices below sustainable levels, something that was already a concern for part of the market.

What changes (and what doesn't) in practice

For those who depend on DeepSeek in production, three points are worth watching over the coming months:

• Price and SLA stability: large rounds often come with product reorganization; it's worth monitoring the changelog and status page before signing an enterprise contract.
• Corporate restructuring: the move toward a 2027 IPO tends to bring more governance and financial reporting, something that's currently scarce among closed Chinese AI companies.
• Geopolitical risk: capital from Tencent and CATL reinforces DeepSeek's ties to Chinese conglomerates, which could weigh on compliance assessments for Brazilian companies operating in the US or with clients sensitive to data origin.

None of these points is a reason to switch providers today, but they are exactly the kind of information that should factor into an architecture decision that relies on a single model vendor. Teams that use DeepSeek as the central piece of a pipeline (not just as a fallback) gain one more argument for keeping provider abstraction in the code, through layers like LiteLLM or a custom wrapper, so they can switch models without rewriting the application.

What's still open

TechNode and Bloomberg make clear that the numbers could still change before the round closes, and that DeepSeek itself has not confirmed the amount. There are also no public details about the company's valuation, the specific use of the money (training new models, international expansion of the API, infrastructure), or an exact IPO timeline beyond "early 2027."

For Brazilian developers, the practical recommendation is the same as always when betting on any third-party model: follow the official documentation, test with real data from your use case, and avoid lock-in that would force you to rewrite your integration if the company changes its pricing or access strategy after closing the round.