DeepSeek Nears $7.4B Round at $74B Valuation as 2027 STAR Market IPO Takes Shape

DeepSeek — the Chinese lab whose open-weight models upended global API pricing and forced every frontier competitor to respond — is once again at the center of the AI capital story. According to multiple reports from China Money Network, the South China Morning Post, and the Wall Street Journal, the company is closing in on a roughly 50 billion yuan (about $7.4 billion) funding round at a pre-money valuation of around 500 billion yuan — approximately $74 billion. The round is targeting an end-of-August close, and it is explicitly designed as a pre-IPO stepping stone: DeepSeek could file for a public listing as early as the end of 2026, with a debut targeted for 2027 on Shanghai’s Nasdaq-style STAR Market.

The Deal on the Table

The structure of the round, as reported by SCMP and later confirmed in outline by the WSJ, reads like a who’s-who of Chinese strategic capital. Returning backers include Monolith Management, Shixiang Capital, CATL (the world’s largest battery maker), Tencent, JD.com, and NetEase — a coalition that spans hedge funds, internet platforms, and heavy industry. New investors are in talks to join: CPE, Legend Capital, and the semiconductor-focused Stony Creek Capital, alongside GigaDevice-backed funds and Hefei state vehicles.

That last detail matters. GigaDevice is one of China’s leading memory and MCU chipmakers, and Hefei — the capital of Anhui province — has become one of the most aggressive state investors in China’s semiconductor and AI supply chain, having previously bankrolled BOE Technology and NIO. Their arrival in DeepSeek’s cap table signals that this is not just a financial bet: it is an industrial-policy alignment, tying China’s most famous open-model lab directly to the national push for compute self-sufficiency.

The fresh $7.4 billion comes just weeks after DeepSeek’s first-ever external raise — also roughly $7.4 billion — which closed in June at a post-money valuation in the region of $52 billion. If the new round closes at the reported terms, DeepSeek’s valuation will have risen more than 40% in a single quarter, one of the fastest markups on record for a company of this scale.

What the Money Buys: 1GW of Compute

The stated use of proceeds is unusually concrete for an AI fundraise. The bulk of the capital will fund roughly one gigawatt of new compute capacity — an enormous buildout for a lab that has historically run lean, extracting frontier-level performance from clusters far smaller than those of OpenAI or Google.

The timing is not incidental. DeepSeek’s model lineup has grown increasingly demanding: the V4 family spans the MIT-licensed V4 Flash (284B parameters) and V4 Pro (1.6T-parameter MoE with 1M-token context), and the just-released V4-Pro-0813 pushed agent benchmarks like Terminal Bench 2.1 (82.7) and Cybergym (76.7) well past its April preview. Serving, fine-tuning, and training that generation of models at scale requires exactly the kind of capacity this round is designed to buy.

The remainder of the capital, per SCMP’s sources, will go toward intensifying the talent war inside China — where DeepSeek competes for researchers not only against Alibaba’s Qwen, Tencent’s Hunyuan (which just open-sourced the 770B Hy4), and Zhipu’s GLM, but also against the deep pockets of every hyperscaler building out agentic AI.

Revenue Reality Check: $500M ARR and Climbing

For a company being marked at $74 billion, revenue is the obvious question. By WSJ’s reporting, DeepSeek’s annualized revenue has reached roughly $500 million — a figure The Information corroborated in July. That makes the implied valuation multiple roughly 150x ARR, a number that would be aggressive even by 2021 standards, let alone in a market that has spent 2026 digesting what “AI revenue quality” actually means.

The bull case rests on trajectory rather than the current base. $500 million ARR flowing from API sales of models that are — critically — open-weight, is a position no Western lab occupies. DeepSeek monetizes while the ecosystem it seeds does free distribution, fine-tuning, and devrel on its behalf. Every V4 deployment, every vLLM/SGLang recipe, every fine-tune in production somewhere is simultaneously a zero-cost moat and a lead-generation funnel back to DeepSeek’s own hosted API.

The IPO Path and What It Means

The strategic prize is the listing. Shanghai’s STAR Market has spent two years preparing for a DeepSeek-class IPO: regulators relaxed profitability requirements, streamlined review for “hard tech” companies, and — per multiple reports — changed listing rules weeks before DeepSeek’s plans solidified. A 2026 filing with a 2027 debut would make DeepSeek one of the largest STAR listings ever, and the first true test of whether China’s onshore market can price a frontier-AI asset.

The contrast with US peers is stark. Anthropic is reportedly approaching a $1 trillion pre-IPO valuation and SpaceX is targeting what could be 2026’s largest listing — but both are US stories, priced in dollars, under US disclosure rules. DeepSeek’s listing would be the first frontier-lab IPO priced in yuan, for a domestic investor base, under a regulatory regime that has made no secret of wanting national champions on national exchanges.

Risks and Open Questions

Three risks stand out. First, the round’s terms have already shifted once: DeepSeek briefly paused the raise in late July after viral social-media posts, resuming in August with Bloomberg reporting a reopened round seeking close to $8 billion. Terms “subject to change” is doing real work in every report. Second, export-control pressure is escalating — the US Commerce Department is drafting rules to close the “remote access” loophole that allegedly let Moonshot train on US chips in Thailand, and any precedent set there could complicate DeepSeek’s own compute procurement even with domestic accelerators ramping. Third, the open-weight model that made DeepSeek famous also caps its pricing power: V4 Pro’s $1.32/$3.96 per million tokens undercuts Western rivals but also invites commoditization from its own ecosystem.

None of these seem likely to stop the round. By every indication, DeepSeek closes August with roughly $7.4 billion more in the bank, a $74 billion mark, a gigawatt of compute on order, and a date circled on the 2027 calendar.