DeepSeek Eyes $45B Valuation in First-Ever Funding Round

DeepSeek, the Hangzhou-based AI research lab that stunned the global AI industry in early 2025 with its resource-efficient large language models, is pursuing its first external funding round at a target valuation of $45 billion, according to people familiar with the matter reported by Bloomberg. If completed, the raise would transform DeepSeek from a largely self-funded offshoot of quantitative hedge fund High-Flyer Capital Management into one of the most valuable AI startups on the planet — without having spent years accumulating venture backing the way its Western rivals have.

What's New on the Table

Until now, DeepSeek has operated almost entirely on internal capital provided by High-Flyer, the quant fund founded by Liang Wenfeng, who also serves as DeepSeek's chief executive. That arrangement gave the lab unusual independence — no outside investors demanding growth metrics, no board seats held by outside parties — but it also capped how aggressively the company could scale compute infrastructure and hire internationally.

The decision to seek outside funding marks a strategic inflection point. A $45 billion valuation would place DeepSeek firmly in the tier occupied by Anthropic (valued at roughly $61 billion after its most recent round) and well ahead of where Mistral AI or Cohere currently sit. It would also make DeepSeek one of the highest-valued AI companies globally that is not headquartered in the United States — a geopolitically loaded distinction given ongoing export controls on advanced semiconductors targeting Chinese firms.

The specific size of the capital raise — i.e., how many dollars DeepSeek is actually seeking to bring in, as opposed to the post-money valuation it is targeting — has not been confirmed publicly. Based on comparable rounds at similar valuation step-ups, the primary raise could fall anywhere in the $500 million to $2 billion range, though that figure remains unverified. Investor identity and closing timeline are similarly unconfirmed as of publication.

Why the Numbers Matter

The $45 billion ask deserves context. DeepSeek's core claim to fame is that it achieved near-frontier model performance at a fraction of the training cost traditionally associated with leading models. Its DeepSeek-R1 reasoning model and the underlying DeepSeek-V3 architecture, released in late 2024 and detailed in the lab's publicly available technical reports, demonstrated that aggressive algorithmic optimization — including mixture-of-experts routing and multi-head latent attention — could dramatically reduce the GPU-hours required per training run. The company's own documentation cited a training cost for DeepSeek-V3 of approximately $5.6 million, a figure that drew widespread skepticism but also widespread acknowledgment that it was structurally lower than comparable Western runs.

That efficiency narrative is the central valuation argument. If DeepSeek can deliver competitive intelligence at lower marginal cost, its gross margin profile on any future inference or API business could be substantially superior to rivals burning hundreds of millions of dollars on compute annually. Investors pricing the round at $45 billion are effectively betting that this cost advantage compounds — that DeepSeek can out-margin, not just out-engineer, the field.

There is a countervailing risk, however. U.S. export controls have restricted DeepSeek's access to the latest NVIDIA H100 and H200 GPUs. The lab has publicly acknowledged relying on older A100 hardware and domestic alternatives. Scaling from research efficiency to full commercial deployment — the kind required to justify a $45 billion valuation — may require compute access that current export restrictions complicate. Any future tightening of those controls represents a material risk factor that prospective investors will need to price.

Separately, DeepSeek's open-weight model releases have earned it enormous goodwill in the developer community but do not, on their own, generate revenue. The company has not publicly disclosed a commercial monetization roadmap, making the valuation more akin to a bet on strategic optionality — API services, enterprise licensing, or domestic Chinese government contracts — than on proven revenue multiples.

BlockAI News' Take

The optics here cut in two directions simultaneously, and that tension is the real story.

On one hand, a $45 billion first-round valuation for a lab that has never taken outside money — and that built its reputation on not needing to spend lavishly — is a striking number. It suggests that the efficiency-as-moat thesis has genuinely landed with sophisticated capital. The fact that DeepSeek is only now coming to market, after already demonstrating technical credibility, gives it negotiating leverage that most early-stage AI startups simply do not have. Liang Wenfeng is not selling a vision; he is selling a track record.

On the other hand, the Web3 and decentralized-AI communities should pay close attention to what this round is not doing. DeepSeek is raising through a conventional private equity structure, concentrating governance in a single founder-led entity with deep ties to a Chinese quant fund. There is no on-chain transparency, no token-based alignment mechanism, no community ownership layer. At a moment when projects like Bittensor, Nous Research, and various decentralized compute networks are arguing that AI development can be structured differently, DeepSeek's raise is a reminder that the most technically credible frontier labs are still gravitating toward the oldest capital structures available.

That is not a criticism — it is a signal. If $45 billion in centralized, opaque, geopolitically sensitive AI capital feels uncomfortable to you, the question worth asking is whether decentralized alternatives can match the engineering output that justified that number in the first place. The answer to that question will define a significant portion of the next AI cycle.

Watch for disclosure of the lead investor's identity and jurisdiction — that detail alone will tell observers more about DeepSeek's strategic alignment than any technical paper — and monitor whether U.S. Commerce Department export control revisions scheduled for mid-2026 review affect the round's closing conditions.

How we report: This article cites primary sources, regulatory filings, and on-chain data where available. BlockAI News uses AI tools to assist with research and first-draft generation; every article is reviewed and edited by a human editor before publication. Read our full How We Report page, Editorial Policy, AI Use Policy, and Corrections Policy.