A Reuters exclusive resting on two anonymous sources describes an early procedural step, not a filing — but the structural read is genuinely significant: the public-market route for frontier AI labs is already running, and it is Chinese.
Context — Press-Reported Figures, Not IPO Valuations
~$75B
Current round implied valuation (~500bn yuan, approx. conversion; press-reported)
$7.4B
June 2026 raise at above $50B valuation (press-reported)
1,848×
MiniMax HK IPO oversubscription (Jan 2026; press-reported)
1,164×
Zhipu AI HK IPO oversubscription (Jan 2026; press-reported)
In a Reuters exclusive published September 9, 2026, DeepSeek — the Hangzhou-based AI developer founded by Liang Wenfeng — has engaged CITIC Securities to prepare for a potential initial public offering on Shanghai’s STAR Market and to conduct pre-listing tutoring before submitting an application. The sourcing requires precision: the story rests on two people with knowledge of the matter who spoke anonymously because the discussions are private. DeepSeek and CITIC did not immediately respond to Reuters’ requests for comment. No Chinese regulator has made any statement. There are no quotes from any company, executive or official in the record.
What pre-listing tutoring actually is matters here. In a mainland Chinese listing, appointing a securities firm for pre-listing counselling is a standard, early procedural step that occurs before an application is even submitted to regulators — let alone accepted or approved. It is evidence of intent and preparation. It is not a filing, a registration, or a market event. Reuters states that DeepSeek aims to begin the process this year; the Wall Street Journal has been cited for a 2027 listing. That conflict sits in the record and is worth naming rather than resolving — neither timeline has been officially confirmed, and the two may not be contradictory depending on what “begin the process” means in practice.
Critically, Reuters states explicitly that the timing of a potential offering, the amount DeepSeek could seek to raise, and its target valuation have not been determined. For context — attributed to press reporting, not company disclosure — a current funding round is reported to value DeepSeek at around 500 billion yuan, approximately $75 billion at an approximate conversion. The company raised $7.4 billion in June 2026 at a valuation above $50 billion, per press reporting. Its shareholders are reported to include Tencent, listed in Hong Kong, and battery maker CATL, listed in Shenzhen. None of those marks is an IPO valuation, and none should be read as one.
DeepSeek — What Is Confirmed and When
June 2026
$7.4B raise at above $50B valuation — press-reported, not company-confirmed. Tencent and CATL reported as shareholders.
9 September 2026
Reuters reports CITIC Securities engaged for pre-listing tutoring ahead of a potential STAR Market IPO. Timing, raise size and valuation explicitly undetermined. No application filed; no regulator comment.
The key insight: The public-market route for frontier AI labs is not a future event — it is already running, and it is Chinese. Zhipu AI and MiniMax listed in Hong Kong in January 2026 and both now report audited results on a public schedule. What a potential DeepSeek listing would add is capability — the most capable lab yet to attempt the route — and venue: the first of this cohort onto a mainland exchange rather than Hong Kong, running through a domestic, state-supervised, substantially administrative process rather than a globally priced market event.
Both figures are press-reported private-round marks, not company-confirmed and not IPO pricing. Reuters states explicitly that the timing of any offering, the amount DeepSeek could seek to raise and its target valuation have not been determined. The dollar equivalent of the ~500 billion yuan figure is an approximate conversion.
The Structural Read
The framing that matters is not “DeepSeek IPO” — it is that the two blocs are running structurally different capital plumbing for their frontier labs. American labs — OpenAI, Anthropic — are financed through enormous private rounds from sovereign wealth funds, crossover investors, and strategic corporates, with a listing held out as a future liquidity event at marks the private market has already set. Anthropic’s reported IPO preparation is the live example of that model: the valuation is established in private, the public offering is the eventual exit, and the financials remain internal until that moment arrives.
A STAR Market listing runs differently. The STAR Market process is domestic and state-supervised, moving through counselling, application, and registration in a sequence that is substantially administrative rather than a globally priced market event shaped by international institutional investors. That distinction affects who sets the price, what disclosure the process demands, and when that disclosure becomes public.
The Reuters exclusive notes only that Z.AI and MiniMax went public in Hong Kong earlier this year; the listing details that follow come from other outlets. Zhipu AI listed on January 8, 2026 under the code 02513.HK, priced at HK$116.20, with HK$4.35 billion its launch target and debut reports putting the actual raise at roughly HK$4.17 billion to HK$4.3 billion, on retail demand reported oversubscribed around 1,159 times. MiniMax followed on January 9, raising approximately HK$4.8 billion on an offering reported oversubscribed some 1,848 times, and roughly doubled on its debut. Both now publish results: MiniMax filed full-year 2025 figures on March 2, 2026. American labs are the outliers here, not the Chinese ones. OpenAI and Anthropic remain private. The Chinese labs went first.
And the precedent already shows what disclosure exposes. Per press reporting on those Hong Kong listings, Zhipu’s 2024 revenue was approximately $44.6 million against a net loss of roughly 2.96 billion yuan. MiniMax’s full-year 2025 revenue was $79.0 million, per the results it published on March 2, 2026, against roughly $1.3 billion in accumulated losses (the $53.4 million figure that circulated covers only the first nine months of 2025). Tens of millions of revenue against billions of losses — audited, public, on the record. That is the real asymmetry in this industry: the Chinese labs’ numbers can be read; the American ones can only be inferred. If DeepSeek were to list — and the conditional is doing real work, because nothing has been filed — that readable record would extend to a genuinely leading-capability lab, which is precisely why this procedural step is worth taking seriously at all.
Disclosure Asymmetry — Author’s Analysis
The American Labs’ Numbers Are a Black Box. The Chinese Labs’ Are Not.
Zhipu and MiniMax publish audited revenue and loss figures on a public schedule. OpenAI and Anthropic do not. A conditional DeepSeek listing would extend that readable record to the lab with arguably the strongest demonstrated efficiency-per-parameter results in the field. Investors, competitors, and governments would all be reading the same document. That is a different kind of exposure than a private fundraise — and a different kind of information advantage for whoever reads it carefully.
Three Analyst Observations
Both offered as the author’s structural analysis, not as reported motives or stated intentions of any party.
OBSERVATION 1 — COMPUTE UNDER CONSTRAINT
DeepSeek builds under US export controls. The day after the listing story, on September 10, a separate Reuters exclusive reported that Chinese accelerator prices had been rising on the back of a global high-bandwidth-memory shortage: Huawei’s Ascend 950DT reportedly up between 20 and 50 percent, Cambricon’s 690 up between 20 and 30 percent. A Chinese frontier lab is therefore paying structurally more than a US peer for equivalent compute and has fewer options for sourcing the best of it. Domestic public capital is one of the few levers still fully available when the hardware lever is constrained. You cannot buy your way around an export control, but you can broaden who funds the attempt — and a public listing is a mechanism for doing exactly that at scale, on domestic terms.
Tencent and CATL sitting alongside each other on DeepSeek’s reported shareholder list is not a conventional financial syndicate. It pairs consumer distribution and cloud reach — Tencent — with one of the world’s largest battery and energy-systems manufacturers — CATL — in a single cap table. The combination is telling for a business whose binding inputs at scale are distribution and power. Consumer reach gets models in front of users; energy infrastructure underwrites the compute capacity to serve them. That pairing looks less like opportunistic investment and more like a structural read on what frontier AI actually requires to operate at scale — though the author notes that motive is inferred from structure, not reported.
OBSERVATION 3 — THE INFORMATION ASYMMETRY INVERTS
If a STAR Market listing proceeds — conditional in every sense, nothing filed, no approval granted — the standard assumption that US labs operate with a structural intelligence advantage over their Chinese counterparts acquires a complication. A prospectus lodged with a mainland regulator and eventually made public would contain audited cost structures, compute expenditure, and revenue breakdowns from a lab whose reported ability to reach strong results at comparatively low training cost has been widely discussed and never independently audited. That document would be read by US labs, US investors, and US policymakers simultaneously. The disclosure asymmetry would run in the opposite direction from what most Western observers assume.
Business Engineer Framework
The Map of AI — Where DeepSeek Sits in the Stack, and What a Listing Changes
The Map of AI traces the nine layers of the AI stack — from raw compute and foundation models through distribution, application, and capital structure. DeepSeek sits at the foundation-model layer, constrained at the compute layer by export controls, and potentially opening the capital layer via a domestic public market. Understanding which layer a company controls — and which layers it depends on others for — is the cleanest way to read what this pre-listing step actually changes structurally, and what it does not.
Any formal application to the CSRC. Pre-listing tutoring precedes an application; an application precedes acceptance; acceptance precedes approval. None of those steps has occurred. The next meaningful event in the administrative sequence would be a filing — which would be public. Until then, nothing in the regulatory record exists.
This account rests on a Reuters exclusive sourced to two people who spoke anonymously; it is not official and not company-confirmed. DeepSeek and CITIC Securities did not immediately respond to requests for comment and no Chinese regulator has commented. No application has been filed or accepted, no approval has been granted, and regulatory acceptance is not assured — engaging a securities firm for pre-listing counselling is a standard early procedural step, not a filing. Reuters states that the timing of any offering, the amount DeepSeek could seek to raise and its target valuation have not been determined; outside reporting differs on whether a listing would come in 2026 or 2027. Valuation figures cited here are press-reported private-round marks rather than IPO pricing, and yuan-to-dollar equivalents are approximate conversions. Any geopolitical or policy framing is the author’s analysis, not something Reuters reported. DeepSeek is a private company; Tencent is listed in Hong Kong and CATL in Shenzhen. This is businessanalysis, not investment advice, and no view is expressed on any security. No quotes were available and none are invented here.
Gennaro is the creator of FourWeekMBA, which reached about four million business people, comprising C-level executives, investors, analysts, product managers, and aspiring digital entrepreneurs in 2022 alone | He is also Director of Sales for a high-tech scaleup in the AI Industry | In 2012, Gennaro earned an International MBA with emphasis on Corporate Finance and Business Strategy.
Scroll to Top
Discover more from FourWeekMBA
Subscribe now to keep reading and get access to the full archive.
