Chinese AI developer DeepSeek is approaching the close of a funding round of approximately RMB 100 billion, more than double its original target, as its latest model V4.1 Flash continues to reshape how enterprises worldwide evaluate the trade-off between AI capability and cost.
Funding Round Doubles in Two Weeks, Industrial Capital Leads
According to multiple transaction insiders cited by Caixin Magazine, DeepSeek is set to complete its second external funding round in the near term, with the amount increased from a previously planned RMB 50 billion to approximately RMB 100 billion. The proposed investments from new and existing shareholders each account for roughly half of the total.The doubling in scale has not affected the overall valuation, with DeepSeek's pre-money valuation remaining at approximately RMB 500 billion.
New investors are predominantly industrial capital, including Stony Creek Capital, SMIC Puhua, CPE Yuanfeng, Legend Capital, and Geely. Existing shareholders from the first round — including CATL, Tencent, IDG Capital, Lisi Capital, and Shixiang Capital — are also participating. CATL has increased its commitment, proposing an additional investment of approximately RMB 10 billion, after contributing about RMB 5 billion in the first round.
The first round, completed in June 2026, raised RMB 51 billion at a post-money valuation of nearly RMB 400 billion, a record for a single round in China's AI industry at the time.
V4.1 Flash: Asymmetric Architecture Delivers Flagship Performance at a Fraction of the Cost
DeepSeek's technical trajectory is equally significant. Released on September 10, 2026, DeepSeek-V4.1-Flash is a 552-billion-parameter Mixture-of-Experts model that employs a novel Causal-Encoder-Decoder architecture with asymmetric activation — just 8 billion parameters active for input and 16 billion for output.This design delivers benchmark results that surpass DeepSeek's own previous flagship, V4-Pro, across multiple dimensions.
On DeepSWE v1.1, a software engineering benchmark, V4.1 Flash scored 74.2, edging out Claude Opus 5 at 74.0 and GPT-5.6 Sol at 73.0.On Terminal-Bench 2.1, which tests AI on real-world compute tasks, it scored 90.6, surpassing GPT-5.6 Sol at 88.8 and Moonshot AI's Kimi K3 at 88.3.Its Codeforces rating of 3471 is the highest in the comparison table, above V4-Pro at 3348 and V4-Flash at 3289.
The cost implications are substantial. DeepSeek has already permanently reduced API pricing for its V4 series to as low as RMB 0.025 per million tokens for cached input and RMB 6 for output, a record low among major global models.Starting September 14, 2026, all deepseek-v4-pro API requests are being routed to V4.1-Flash at V4.1-Flash rates, effectively making the more capable model the default option at no additional cost.
Enterprise Adoption Rebounds as Cost Pressures Mount
In the U.S. enterprise market, DeepSeek's trajectory tells a story of cost-driven pragmatism. According to Ramp, a U.S. corporate spend management platform serving approximately 70,000 businesses, DeepSeek topped its June 2026 software trend list, which tracks first-time enterprise purchases of software vendors.
Ramp's Chief Economist Ara Kharazian described the development as "perhaps the clearest signal" that U.S. companies are seeking lower-cost alternatives to OpenAI and Anthropic. Crucially, these enterprises are paying DeepSeek directly to use its models, rather than deploying open-source versions on their own infrastructure.
The context is one of escalating AI costs. One consulting client reportedly spent $500 million in a single month on Claude licenses after failing to set usage caps. Uber disclosed that it exhausted its entire annual token budget in the first four months of 2026. Salesforce projected approximately $300 million in payments to Anthropic for the year.A Bain survey of 951 enterprises with over $100 million in revenue found that after more than $1 trillion in cumulative AI spending, actual cost savings have consistently fallen short of expectations.
Revenue Growth Validates Commercial Viability
DeepSeek's commercial momentum is reflected in its financials. According to The Information, the company's annualized revenue run rate has surpassed $1 billion, more than double the figure from a few months earlier. CEO Liang Wenfeng disclosed the figure at a recent investor meeting, noting that API price increases of 2.3 to 4.5 times did not cause customer churn.
In the first seven months of 2026, DeepSeek generated approximately RMB 475 million in revenue, roughly ten times its full-year 2025 total. The company is preparing for a potential listing on the Shanghai Stock Exchange's STAR Market, with CITIC Securities engaged as the lead underwriter.
A Genuinely Global User Base
DeepSeek's user distribution stands out among Chinese AI applications for its geographic diversity. According to the a16z Top 100 Gen AI Consumer Apps ranking, DeepSeek ranks fourth globally in web traffic, with users from China (33.5%), Russia (7.1%), and the United States (6.6%) — making it a cross-ecosystem "breakout" player.In the Philippines, Chinese AI chat providers including DeepSeek have reached a 47% market share on Android devices; in Indonesia and Peru, the figure stands at 38%, and in Mexico, 30%.
The Strategic Bet: Domestic Compute for Frontier Training
DeepSeek's funding round carries strategic significance beyond capital. According to reports, the company plans to build a large data center in Inner Mongolia deploying at least 160,000 Huawei Ascend 950DT super-node accelerators — potentially one of the largest known Huawei AI chip clusters. During a recent investor roadshow, Liang Wenfeng reportedly described the plan to conduct frontier model pre-training on domestic compute for the first time as "DeepSeek's biggest gamble."
The company allocates more than 70% of its computing capacity to training and less than 30% to inference, underscoring its continued focus on frontier model development over short-term commercialization.
What It Means for Enterprise Buyers
For global enterprises evaluating AI vendors, DeepSeek's trajectory offers a clear proposition: near-flagship performance at a fraction of the cost of U.S. alternatives. The V4.1 Flash model's combination of benchmark competitiveness, asymmetric architecture efficiency, and aggressive pricing creates a compelling case for cost-conscious buyers — particularly those in price-sensitive markets or high-volume deployment scenarios.
The company's revenue growth, enterprise adoption rebound, and imminent IPO suggest that DeepSeek has moved beyond the "free alternative" phase. The question for buyers is no longer whether Chinese AI can compete on capability, but whether the cost advantage is sustainable — and whether the strategic bet on domestic compute will pay off in the next generation of frontier models.
Sources: Caixin Magazine, Reuters, The Information, Ramp, a16z, DeepSeek API documentation, and other public sources. As of October 14, 2026.