DeepSeek Prepares for a STAR Market IPO as Its Valuation Nears $70B
DeepSeek, the Hangzhou-based AI model developer that upended global assumptions about the cost of frontier AI, has hired CITIC Securities to underwrite a planned initial public offering on the Shanghai Stock Exchange's STAR Market, according to two people familiar with the matter cited by Reuters on September 9, 2026 — a move that signals the company's transition from a research-driven laboratory into a capital-market-facing enterprise.
Neither DeepSeek nor CITIC Securities responded to requests for comment at the time of publication. The listing timeline, fundraising size, and offering valuation remain undetermined, the sources said. The engagement of an underwriter, however, represents a concrete procedural milestone: under Shanghai Stock Exchange rules, formal sponsor retention precedes guidance filing, regulatory review, and China Securities Regulatory Commission registration — meaning a public debut remains months away at minimum.
The IPO preparation runs in parallel with a private fundraising round that has drawn significant market attention. In June 2026, DeepSeek closed its first external financing round, raising approximately $7.4 billion at a post-money valuation exceeding $50 billion. Founder Liang Wenfeng personally contributed RMB 20 billion (approximately $2.78 billion), while Tencent and Contemporary Amperex Technology (CATL) invested RMB 10 billion ($1.39 billion) and RMB 5 billion ($694 million), respectively, becoming the company's two largest external shareholders.
Valuation Gap Reveals the Scale of Capital Market Ambitions
Reuters reported in July 2026 that DeepSeek was exploring a subsequent private round targeting a pre-money valuation of approximately RMB 500 billion ($69.4 billion) — a figure that the same outlet reiterated on September 9. The Financial Times, in a concurrent report, cited a slightly lower figure of approximately $71 billion. The divergence in reported figures reflects the round's still-fluid status; South China Morning Post had reported in late August that the round was expected to close before month-end, though no formal closing has been publicly confirmed.
At a $69–71 billion target valuation, the implied revenue multiple is stark. The Information reported that DeepSeek generated approximately RMB 475 million ($65.9 million) in revenue in the first seven months of 2026, representing roughly tenfold growth over full-year 2025. Annualizing that trajectory yields approximately $112 million in recognized revenue — a fraction of the implied valuation. The Financial Times separately reported that DeepSeek's annualized revenue run rate has approached $500 million, a figure that still implies a revenue multiple exceeding 140x at the upper end of the valuation range.
What partially justifies that premium, at least on paper, is unit economics: The Information disclosed that DeepSeek's API business carried a gross margin of 82.9% in the same period — a figure competitive with the best-in-class software businesses globally. That margin reflects the company's core thesis: that superior inference efficiency, the technical hallmark that made DeepSeek-V3 and R1 globally disruptive in early 2025, can be monetized at scale. The question public markets will ultimately force the company to answer is whether that efficiency edge is durable as rivals close the gap.
Infrastructure Buildout Erodes the Low-Cost Narrative
The IPO rationale is inseparable from a structural shift in DeepSeek's cost profile. The company's widely-cited training cost of approximately $5.576 million for DeepSeek-V3 — derived from its own technical report using an estimated $2 per H800 GPU-hour — was always a narrow metric. The figure captured only the compute cost of a single formal training run, explicitly excluding prior research, ablation experiments, and ongoing inference infrastructure.
Bloomberg previously reported that DeepSeek is planning a gigawatt-scale AI data center in Ulanqab, Inner Mongolia, with a planned capacity of at least 1 GW. For a company once celebrated for doing more with less, a 1 GW facility represents a categorical shift in capital intensity. Beyond chip procurement, such a facility requires land, grid interconnection, energy storage, cooling systems, network infrastructure, and sustained depreciation — none of which is a one-time expenditure.
CATL's investment is a direct signal of where that infrastructure logic leads. The Ningde-based battery giant has been actively developing power supply and energy storage solutions for AI data centers; its stake in DeepSeek does not imply a finalized supply contract, but it does suggest that the competitive perimeter of large language model development has expanded well beyond algorithms and parameter counts into chips, power, storage, and physical infrastructure.
On September 8, 2026, DeepSeek posted approximately 150 new job openings concentrated in server-side development and agent elastic compute research — roles explicitly designed to support growing data volumes, training evaluation tasks, and user request loads. According to a company spokesperson cited in local media, existing systems require upgrading, maintenance, and in some cases full rewrites to handle expanded scale. Engineering headcount of this type compounds annually with the service it supports.
Pricing Signals a Monetization Pivot
DeepSeek's commercial strategy is also visibly evolving. Effective August 16, 2026, the company revised pricing for its V4 series API and introduced a peak-and-off-peak tiering mechanism, with off-peak rates set at 50% of peak pricing. The move marks a departure from the aggressive penetration pricing that rapidly built DeepSeek's developer base — and reflects an attempt to optimize simultaneously for call volume, compute utilization, and revenue yield. It is, in effect, the company's first public acknowledgment that growth and margin must now be managed in tandem rather than traded off.
Reuters also cited a source indicating that founder Liang Wenfeng views IPO proceeds as a mechanism to strengthen equity incentives for core employees and researchers. For a company whose competitive advantage is concentrated in a relatively small group of researchers, the ability to offer liquid equity compensation is a material retention and recruitment tool — one that private company structures constrain.
STAR Market Rules Create a Tailored On-Ramp
The choice of the STAR Market is not incidental. On June 17, 2026, the Shanghai Stock Exchange published dedicated guidelines permitting large AI model companies to list under the bourse's Fifth Set of Listing Standards — a pathway that allows companies without established revenue to qualify based on demonstrated technical leadership, product deployment at scale, and addressable market size. The guidelines explicitly recognize that R&D investment may precede commercialization in frontier AI, and treat scaled product deployment as a qualifying milestone.
DeepSeek's API business, its open-source model ecosystem, and its documented developer adoption would likely feature prominently in any application under this framework. Which specific standard DeepSeek ultimately selects, and how it structures the supporting documentation, will only become clear once formal guidance filings are submitted to regulators.
Sector-Wide Capital Pressure Accelerates the IPO Wave
DeepSeek is not navigating this transition alone. China's frontier AI sector is experiencing a synchronized push toward public markets, driven by the same compute cost dynamics. Zhipu AI and MiniMax have both completed Hong Kong listings in 2026. Reuters reported that Moonshot AI, developer of the Kimi assistant, has filed confidentially for a Hong Kong IPO. Moonshot's most recent private valuation stood at approximately $50 billion — below DeepSeek's $69–71 billion target — while Zhipu currently trades at a market capitalization of approximately $54 billion, also modestly below DeepSeek's fundraising benchmark.
The convergence of these timelines reflects a structural reality: the capital requirements for competitive frontier AI development have outgrown what founder balance sheets and private markets alone can sustain. An IPO resolves the funding source problem. It does not, by itself, validate the business model.
For DeepSeek, the prospectus — when it arrives — will be the first document to systematically address the central question that a $70 billion valuation demands: whether efficiency-driven technical superiority can be converted into a revenue base large enough to service the heavy-asset infrastructure it is building.
Related Coverage:
DeepSeek Eyes 2027 IPO, Targets RMB 10B Pre-Listing Raise in China AI Race