Slop TVNewsLatest
News

Wangsu pays 300M yuan for 4.4% of video startup Sand.ai

The Shenzhen filing values Sand.ai near $1 billion, roughly 196 times what the video-model maker's first half of 2026 annualises to.

Illustration: Wangsu pays 300M yuan for 4.4% of video startup Sand.ai
Illustration: SLOP TV News. Shenzhen Stock Exchange photograph: soramimi (CC BY-SA 4.0).

Key takeaways

  • Wangsu Science and Technology will pay 300 million yuan, about $44.45 million, for 4.3963% of Sand.ai.
  • The price per share puts a pre-money valuation of $800 million on Sand.ai and about $1.01 billion after the round.
  • Sand.ai reported 366,000 yuan of revenue for 2025 and 17.4 million yuan for the first half of 2026, against a 151.6 million yuan half-year loss.
  • The money moves as a convertible note and warrant first, because the equity sits offshore and the approvals have not come through.

Wangsu Science and Technology will pay 300 million yuan, about $44.45 million, for a 4.3963% stake in the video-generation company Sand.ai, the Shenzhen-listed content-delivery company told the exchange after Tuesday's close on September 29. Its board voted seven to nothing.

The price per share is the number to start from: 2,105,352 Series A+ preferred shares at $21.1137 each. Divide one into the other and the round prices Sand.ai at about $1.01 billion after the money, on a pre-money valuation of exactly $800 million, per China AI Dispatch's reading of the filing. The newsletter shows its working: 2,105,352 shares are 4.3963% of a post-round count of 47,889,180, so the round issues just under ten million new shares and about $211 million of new money.

Then the accounts, which Wangsu has to disclose because Shenzhen makes it. Sand.ai was incorporated in the Cayman Islands on January 17, 2024. Per China AI Dispatch's reading of the filing, for the year to December 2025 it reported revenue of 366,000 yuan and a net loss of 186.8 million yuan; for the six months to June 2026, revenue of 17.4 million yuan, about $2.59 million, against a net loss of 151.6 million. Both sets are unaudited and the filing says so. Against the $1.01 billion price, the newsletter's math puts the first half of this year at roughly 196 times what it annualises to.

The products are real and public. MAGI-1, an autoregressive video model, was open-sourced in April 2025; MagiHuman does audio and video together; VidMuse is a music-video agent. MAGI-2 Preview, listed on Sand.ai's own site at 114 billion parameters on a mixture-of-experts architecture, went up in August. On GitHub, the MAGI-1 repository carries 3,792 stars on the older model (checked September 30); the newer MAGI-2-preview repository carries 634, with no commits since August 6.

Getting the money there takes three steps, because a Chinese listed company cannot simply wire 300 million yuan to a Cayman entity. Wangsu will first lend the 300 million yuan as a convertible note to Shanghai Sandai Technology and take a warrant over the A+ shares, half by October 15 and half by October 31. If the outbound approvals come through, Shanghai Sandai repays the loan, Wangsu converts the yuan and exercises the warrant. If they never come, the borrower can repay with interest at the one-year loan prime rate and take the warrant back. The agreements are not signed, and the filing says so repeatedly.

The asymmetry is in the same document: Shanghai Sandai's audited total assets were 735,200 yuan as of August 31, against the 300 million yuan loan, a ratio of about 408 to one. Wangsu can carry it. Per China AI Dispatch's reporting on Wangsu's own half-year disclosures, its revenue for the first half was 2.318 billion yuan, down 1.38% on a year earlier, with net profit down 6.61%, and it held 1.692 billion yuan of cash and 5.537 billion yuan of trading financial assets at the end of June. The stake is 4.15% of that pile.

The first half of the bridge loan is due by October 15; the equity transfers only once outbound investment approval clears.

Sources

  1. chinaaidispatch.substack.com - the filing's figures, share price, cap table and the convertible-note mechanics
  2. static.cninfo.com.cn - the Shenzhen disclosure the figures come from
  3. sand.ai - Sand.ai's own site: the MAGI model line
  4. github.com - the open-sourced MAGI-1 repository
  5. github.com - the MAGI-2 Preview repository, checked for stars and last commit
  6. panews.io - secondary report of the same investment