DeepSeek Restarts $7B Fundraise at $66B Valuation Amid Model Delays

Woofun AI reports that DeepSeek has officially restarted its second round of financing, targeting a raise of 50 billion yuan with a pre-investment valuation of approximately 500 billion yuan, aiming to complete signing by late August. This strategic pivot follows a brief suspension triggered by founder Liang Wenfeng’s dissatisfaction with leaked investor meeting transcripts, marking a critical juncture for the company as it navigates intense market scrutiny and competitive pressures in the AI sector.

The financial terms for this second round represent a significant escalation from previous rounds, with the target amount set at 50 billion yuan and the valuation anchored at 500 billion yuan. The timeline for this round began with initial investor engagement in mid-July, but negotiations were abruptly paused at the end of July. The company now intends to finalize the agreement by late August, a compressed schedule that underscores the urgency to secure capital amidst rapidly evolving market conditions and competitive dynamics.

The pause in financing was directly linked to a specific incident on July 26, when DeepSeek suspended discussions due to Liang Wenfeng’s dissatisfaction with widely circulated online comments. These comments primarily revolved around a suspected leaked 'meeting transcript for investors,' which raised concerns about confidentiality and strategic alignment. Despite inquiries from the media, DeepSeek had not provided a response by the time of publication, leaving the communication status ambiguous and highlighting the sensitivity surrounding the company’s internal deliberations.

Historically, DeepSeek’s first round of financing was initiated in April this year and successfully completed in June, raising 50 billion yuan with a valuation exceeding 350 billion yuan. This achievement positioned DeepSeek as the company with the largest financing scale in the history of China's AI large models for its first round. The substantial capital influx established a strong foundation for subsequent rounds, setting a high benchmark for valuation and investor interest in the domestic AI landscape.

The investor composition for the second round includes a mix of standby institutions that 'failed' in the first round and venture capital and industrial capital entities that entered initially and are considering additional investments. Key investors from the first round include the National Artificial Intelligence Industry Investment Fund, Tencent, CATL, Puxuan Capital, NetEase, JD.com, Lisi Capital, IDG Capital, Zhengxin Valley Investment, and Shixiang Capital. The demand for shares remains intense, with the scale of funds expressing investment intentions in the first round exceeding 100 billion yuan, indicating robust market appetite.

Woofun AI data shows that valuation growth is a critical metric, with the pre-investment valuation for the second round increasing by approximately 43% compared to the first round. If this round proceeds smoothly, DeepSeek will have raised over 100 billion yuan in total across both rounds. In contrast, competitor Dark Moon completed its Series F financing in late July with a valuation of approximately 31.5 billion USD (about 210 billion yuan) and immediately opened a new Pre-IPO round at 50 billion USD. Dark Moon aims to complete this financing in August and file for listing within the year, intensifying the competitive race.

Market dynamics reveal volatility in competitor valuations, particularly for Dark Moon. The initial heat for its Series F financing, which started in June, was tepid, with many channels receiving quotas 'selling for several days without anyone wanting it.' Transactions of old shares at an 18 billion USD valuation circulated in the market, reflecting investor caution.

However, the release of the new flagship model Kimi K3, which gained global attention, suddenly made its shares sought after, with some sellers of old shares raising their prices, demonstrating the direct impact of model performance on market sentiment.

DeepSeek faces considerable pressure to train new models, with the official version of DeepSeek-V4, originally scheduled for release in mid-July, only going live for public testing in the latest log of the official API documentation on July 31. Only V4-Flash was showcased, while the official version of V4-Pro has not yet been released, and its specific public testing time remains unannounced. The delayed release highlights the challenges in meeting aggressive timelines while ensuring model quality and performance in a highly competitive environment.

Performance benchmarks and pricing strategy are central to DeepSeek’s competitive positioning. According to the latest Intelligence Index from Artificial Analysis, the intelligence level of DeepSeek-V4-Flash reached 50 points, second only to Zhiyu GLM-5.2’s 51 points in China. DeepSeek continues its 'low cost and high cost-effectiveness' strategy, with V4-Flash output priced at 0.28 USD per million tokens, significantly lower than Zhiyu GLM-5.2’s 4.29 USD. In contrast, OpenAI’s GPT-5.6 series saw an 80% price drop for its entry-level model Luna, but DeepSeek’s cost-effectiveness advantage quickly erased any prior lead held by GPT-5.6.

Market dominance is evidenced by token consumption metrics on the AI model aggregation platform OpenRouter, where the official version of DeepSeek-V4-Flash ranked first in weekly token consumption with 7.1 trillion tokens. This substantial usage volume underscores the model’s appeal to global developers and its ability to capture market share despite competitive pressures. The combination of high performance, low cost, and strong adoption positions DeepSeek as a formidable player in the AI race, with its financing success likely to further solidify its market leadership.