The storm surrounding DeepSeek's funding round has finally settled.

According to foreign media reports, multiple domestic outlets have confirmed: DeepSeek’s first funding round raised approximately $7 billion, with its valuation peaking at $59 billion. Tencent plans to invest 10 billion RMB, CATL intends to commit 5 billion RMB, and founder Liang Wenfeng personally pledged 20 billion RMB.

This marks the first time since DeepSeek’s founding that it has opened its doors to external capital. Previously, it was entirely sustained by Huafang Quantitative, and Liang Wenfeng had never allowed outside shareholders in. This time, the list of investors is particularly telling—Tencent and CATL appear on the same roster.

In the U.S., OpenAI’s backers are Microsoft and NVIDIA; Anthropic relies on Google Cloud. All operate within a shared technological ecosystem, seeking strategic allies. While Tencent and CATL share a broad industrial direction—betting on the AI industry dividend—their core commercial objectives and application domains are fundamentally different.

They’re buying equity in the same company, but thinking entirely different thoughts. It’s precisely this “cohabiting with divergent intentions” that makes this funding round truly worth analyzing.

Tencent: Just Won One Round, Now Betting on the Opponent

Tencent Hun Yuan Hy3 preview claimed the weekly championship on OpenRouter on May 7th, processing 36.6 trillion Tokens, ranking first in both coding and tool calling. This result demonstrates that Tencent’s self-developed technology hasn’t fallen behind. However, once the funding news broke on June 3rd, the market perception shifted: even with strong benchmark performance from in-house tech, Tencent still chose to invest hundreds of millions to secure a stake in a peer, establishing dual-track parallelism.

Beneath this dual-track strategy lies a tangible gap in consumer-facing ecosystems and commercialization execution. According to QuestMobile data from March this year, Tencent Yuan’s monthly active users stood at only 57.35 million. Douyin’s Daobao reached 345 million, while Qwen hit 166 million.

But for Tencent, the more pressing issue is growth velocity: Yuan added just 8.2 million new users in Q1, while Daobao and Qwen added around 100 million and 126 million respectively. The top ranking of Hun Yuan Hy3 preview on OpenRouter may have been influenced by promotional traffic and concentrated API calls, not purely organic user traction.

Tencent’s Q1 capital expenditures amounted to 37 billion RMB, primarily allocated to AI investments, with 31.9 billion RMB counted as current-period capital spending. Massive investment in AI infrastructure has produced disproportionate returns. Hun Yuan API pricing stands at 1.2 RMB per million Tokens, while Daobao charges only 0.6 RMB. At price, Tencent lacks competitiveness; at scale, it’s being left behind by orders of magnitude. Alibaba has already established ATH Business Group, fully pivoting toward token-based economics. ByteDance’s Daobao quietly launched paid subscriptions on May 4th, offering tiers at 68, 200, and 500 RMB. Competitors are no longer competing on API volume—they’re now racing to monetize usage volume.

Tencent’s 10-billion-RMB investment cannot secure exclusive technology rights to DeepSeek. DeepSeek is open-source—its code is freely downloadable by anyone. What Tencent acquires instead is a “priority participation right”: ensuring it won’t be excluded from the table during the next AI application breakout cycle amid the emerging triopoly of ByteDance, Alibaba, and DeepSeek. Hun Yuan can continue consuming internal traffic, while products like Yuan, QQ, and Tencent Docs remain viable within Tencent’s own ecosystem.

DeepSeek fills the gap in external developer support and B2B enterprise clients. Dual-track parallelism does not signify abandoning self-research—it represents strategic hedging via “self-developed foundation + external positioning” during the consolidation phase of AI oligopoly.

CATL: A Battery Giant Enters Large Models—It’s Not About Code, But Metering

While Tencent’s involvement in DeepSeek remains within the internet sphere, CATL’s entry completely transcends it. A battery manufacturer diving into large models has nothing to do with code—it’s all about metering.

DeepSeek’s fund usage includes “increasing compute investment, optimizing employee compensation, and accelerating commercialization,” with compute investment highlighted as the top priority across multiple media reports. Compute power equals electricity consumption—electricity consumption equals energy storage orders.

Looking at CATL’s actions over the past three months reveals an indirect penetration chain in compute-energy infrastructure.

In April, CATL invested approximately 4.1 billion RMB through its affiliate Zhongheng Technology Investment in Zhongheng Electric, acquiring a 49% stake and indirectly entering Zhongheng Electric’s HVDC business. The actual controller, Zhu Guoding, remained unchanged.

In May, CATL invested roughly 6.5 billion RMB via affiliated entity PJ Millennium to acquire about 38.1% of Century Interconnect, but chose to be a “quiet shareholder”—voting rights remain under founder Chen Sheng.

In June, CATL signaled intent to invest around 5 billion RMB in DeepSeek. Over three months, total outlay approached 15.5 billion RMB. This pace exceeds mere financial experimentation and clearly signals strategic realignment.

CATL posted a net profit of 20.7 billion RMB in Q1 this year, with cash reserves exceeding 410 billion RMB. Investing 5 billion in DeepSeek accounts for only about 1.2% of its cash reserves—insignificant in absolute terms. Yet, three consecutive AI infrastructure-related investments have moved beyond what finance departments typically handle. The ceiling for CATL’s core battery business is clearly visible. By April 2026, China’s new energy vehicle retail penetration rate will surpass 60%, marking the end of the high-growth era. The energy storage market is already caught in a severe price war, with second-tier players such as EVE Energy, Haimchen Energy, and Zhongchu Innovation rapidly eroding market share.

Power demand from AI data centers is this new frontier. In its April 15 earnings call, CATL explicitly mentioned that the acceleration of AI adoption is turning power and energy shortages into a core bottleneck, with data center energy storage demand growing exponentially.

Industry estimates suggest that ultra-large-scale data centers, under high-stability power requirements, require energy storage capacities of 15–20 GWh. As large models shift from training to inference, coupled with the deployment of multimodal models, data center power consumption is growing exponentially. Traditional grids cannot sustain such stable power demands—energy storage systems, backup power, and HVDC become essential.

The collaboration between CATL and SenseTime on the Lingang Intelligent Computing Center energy storage system—17.888 MW—can save nearly 10 million RMB in annual electricity costs. This calculation already makes economic sense. Investing in DeepSeek is equivalent to pre-locking a guaranteed, high-growth end-user customer. DeepSeek’s official website released job postings for the Ulanqab data center in April. Based on recruitment details, the Inner Mongolia facility appears to be in project initiation or construction phase, not yet operational. Its compute expansion plan directly aligns with CATL’s order pipeline.

Liang Wenfeng’s Steering Wheel

Early April saw market rumors placing DeepSeek’s valuation at around $10 billion. Half a year later, the financing proposal pegged it at $59 billion—a staggering increase. Such a leap defies attribution to any single factor. Expectations around DeepSeek V4.1’s launch, expansion of its open-source ecosystem, and overall heat in China’s AI primary market all contributed. Tencent and CATL’s entry provided industrial validation, amplifying valuation premiums. This wasn’t capital inflating fundamentals—rather, capital recognition of existing fundamentals amplified the premium.

Tencent brings imagination around application scenarios; CATL (via affiliates) delivers feasibility in energy infrastructure rollout. But the fundamental driver of upward valuation remains DeepSeek’s own open-source ecosystem and expectations for multimodal product deployment.

Ultimately, DeepSeek first occupied this position—only then did these two industrial giants come knocking.

Tencent imbues DeepSeek with application-layer value; CATL (via affiliates) provides energy-layer value. Yet “hub premium” carries an inherent fragility. Tencent needs DeepSeek’s model capabilities to support super-app ecosystems like WeChat, gaming, and cloud services. CATL needs DeepSeek’s compute expansion to absorb excess energy storage capacity.

Both companies’ demands are rational—but their directions diverge. If Tencent pushes for DeepSeek to prioritize its ecosystem, and CATL demands exclusive energy cooperation, tensions between shareholder interests and DeepSeek’s independent commercialization will inevitably surface.

The upcoming release of V4.1—an enterprise-grade multimodal model—is DeepSeek’s pivotal step from open-source demonstration to independent paid service. This transition requires shareholder support—but also demands freedom from shareholder dependency.

Liang Wenfeng personally poured 20 billion RMB, exceeding the combined investment of Tencent and CATL. External shareholders are limited to fewer than 10 entities, all strategically aligned industrial partners—no pure financial investors. This structure isn’t a formal alliance, merely a loose collection of mutually beneficial parties.

Liang Wenfeng’s massive personal investment has one core goal: keeping the steering wheel firmly in his hands. The funding round is just the beginning—the real test lies ahead. When shareholders begin exercising their “participation rights” and “order rights,” whether Liang Wenfeng’s absolute control can truly hold the ground remains uncertain.

Conclusion

ByteDance and Alibaba are pursuing a “tyrannical vertical integration” path—building their own models, cloud platforms, chips, and applications, swallowing the entire industrial chain. Tencent and CATL take a completely different approach: they don’t build everything themselves but opt for capital binding to an independent hub, jointly betting on third-party nodes.

Liang Wenfeng is the one holding the steering wheel.

This isn’t a matter of superiority or inferiority—it reflects differing industrial maturity between the U.S. and China, leading to distinct strategic paths. While American AI firms compete for GPUs, Chinese AI firms are already battling over electricity, use cases, and supply chain integration.

DeepSeek’s funding, fundamentally, is a costly ticket into the “infrastructure arms race.” The ticket has been purchased. Whether the performance can be completed depends on what each of these spectators actually wants to see on stage.

Source: 36Kr