Invest1 distinct publisher3 min readPublished
The quant fund that seeded DeepSeek with cash and compute now farms state-favored listings, where a fivefold debut and a 460% opening day outran a trading book that lost money in eight of nine products in July, while DeepSeek raises capital elsewhere.
The Investor · Invest desk

Compiled by The InvestorSomething wrong?How this is made
Run the two placements PaiPaiWang could put a number to forward at market and the year takes shape: $26 million of CXMT pre-IPO stock [6], against a Shanghai debut that ran more than fivefold in July plus another 20% since [7], carries at roughly $156 million [2], and the $5.8 million of Unitree [8], up 460% on day one and off about 27% since [10], marks near $23.7 million [3]. That is about $180 million against $31.8 million of cost [1][4], all of it unrealised marks rather than cash [19].
The more interesting term, or rather the one that tells you who is doing what, is the lock-up. DeepSeek took 2.31% of the Unitree offering as one of nine strategic investors and agreed to sit still for 36 months, three times the 12 months most strategic backers accepted [9][6]. Hutong Research's Sigrid Wang draws the line cleanly: the quant funds traded Unitree as an investable asset, the lab used its balance sheet to buy a relationship in the future AI stack [13]. Ke Zong, a Shanghai-based portfolio manager, told CNBC that Liang's team are still traders at heart, inclined to chase maximum upside [16], and yet a trader who accepts a three-year lock has moved past trading into paying for a seat.
Nearly half of the two affiliates' allocations this year went into semiconductors and their supply chain [12], which reads less like a factor tilt than like a reading of the state catalogue. Rhodium's Ciel Qi says Beijing's push to keep strategically important tech firms listing at home has created lucrative pre-IPO opportunities, and that maximising returns in China increasingly requires investing in line with Beijing's agenda [4]. Brookings' Kyle Chan adds the reflexive part: a DeepSeek or a Liang fund on the book lends attention and legitimacy, with a potential extra lift to the valuation those same backers are then marked against [14]. Beijing's endorsement of both, in turn, improved their access to marquee listings [15].
This is probably the wrong emphasis, but the more durable fact sits on the funding side. Qi says High-Flyer's revenue has become unstable, and that as DeepSeek's capital and compute needs grow it may be unable to rely on the fund [3]; July showed the mechanism, with eight of the fund's nine products in the red as the global AI-chip rout spilled into momentum quant trades, before Chinese quants rebounded in August [11]. A patron whose monthly P&L moves with the same chip complex its lab is trying to buy from makes a wobbly treasury, which is the plainest explanation for DeepSeek going outside for money [2]. The counter-thesis is respectable: the placement book may be small next to High-Flyer's trading business, the lab's outside round may be about partners rather than a compute bill, and Unitree's 27% give-back is a reminder that a policy-lifted mark is not a price you can sell into [10][18]. I would drop this read if DeepSeek's next raise turns out modest and strategic, or if the affiliates rotate out of semiconductors once the listing queue thins. Neither High-Flyer nor DeepSeek responded to CNBC's requests for comment [17].
Ranked by verification strength, evidence, and original report placement.
DeepSeek has turned to outside investors to fund its growing ambitions, while High-Flyer has secured allocations across some of China's hottest hard-tech IPOs, from chips to robotics.
CXMT was by far the two funds' largest allocation, with a combined $26 million, or 175 million yuan, of pre-IPO shares, according to data compiled by Shenzhen PaiPaiWang Investment & Management.
A global AI-chip rout spilled into momentum-driven quant trades in July, when eight of High-Flyer's nine products recorded losses according to state-backed media, and Chinese quant fund performance rebounded in August.
Nearly half of the two High-Flyer affiliates' allocations this year went to semiconductors and their supply chain, including advanced-packaging specialist SJ Semiconductor and chip-equipment and testing companies, according to available PaiPaiWang data.
Hutong Research tech analyst Sigrid Wang said High-Flyer traded Unitree as an investable asset while DeepSeek invested as a strategic partner, describing a genuine distinction between quant funds seeking returns and DeepSeek selectively using its corporate balance sheet to build strategic relationships around the future AI stack.
Beijing's endorsement of DeepSeek and Liang has helped open doors and improved their access to marquee listings.
Distinct publishers with included, body-backed reporting in this cluster.
1 article · August 27, 2026
Follow any of these and your For You feed starts watching them — no settings page required.
product
Unitree gave back $30bn and still sits near five times its pre-float valuation1 distinct publisher
product
Unitree's $9bn Shanghai listing prices the humanoid story, not the robot-dog business1 distinct publisher
invest
Korea's venture investment jumped 54%. One robot listing is worth eight times all of it.1 distinct publisher
invest
Unitree's $905M Shanghai listing prices humanoids at 35x sales while profit halves1 distinct publisher
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Documented allocations and prices, one publisher, silent principals
The core numbers come from identifiable third parties - Shenzhen PaiPaiWang for allocation sizes, LSEG for Unitree's price path, state-backed media for July product losses - and four named analysts provide interpretation. But it is a single-publisher cluster, High-Flyer and DeepSeek did not comment, sector shares rest on 'available data' rather than a complete filing set, and CXMT's debut is described only as 'more than fivefold', leaving the largest position's mark imprecise.
Capital actually deployed across named listings
This is not software adoption but capital deployment, and it is concretely observed: two named affiliates hold disclosed dollar placements in CXMT and Unitree, DeepSeek holds 2.31% of Unitree under a 36-month lock-up, and nearly half of this year's affiliate allocations sit in semiconductors and supply-chain names. What is not observed is position persistence or exits, so deployment is documented while outcomes are not.
Paper marks presented ahead of realizable returns
The headline mark-up from $31.8 million to near $180 million is arithmetic applied to reported allocations and price moves, not disclosed proceeds. The source itself calls such placement gains paper gains, notes lock-ups of 12 to 36 months for strategic backers, records Unitree already 27% below its debut, and reports eight of nine High-Flyer products losing money in July - facts that make the multiple a snapshot rather than a return. Modestly overstated rather than unfounded, since the inputs are all sourced.
Policy-aligned allocations, silent principals, analyst-sourced motives
Incentive structure is explicitly on the record: an analyst says maximizing returns increasingly requires investing in line with Beijing's agenda, a Brookings fellow calls the allocations a financial bonus for boosting China's tech sector and notes star backers lend legitimacy that can lift valuations, and CNBC states Beijing's endorsement improved access to marquee listings. Counterweight: the principals declined comment, so motive is entirely third-party attributed, and one analyst frames the alignment as convergence rather than state direction.
Solid underlying data, thin corroboration, unverified marks
Confidence sits mid-range: the discrete facts (allocation sizes, stake percentage, lock-up terms, price moves, July losses) are attributed to identifiable data providers and are internally consistent, but everything flows through one publisher, the principals are silent, and the story's headline conclusion is derived arithmetic on unconfirmed holdings.