VAST · Tripo
China’s text-to-3D generation unicorn is monetizing a workflow product ahead of a nascent world-model bet — three financing tranches in six months are pricing an optionality-heavy roadmap that has yet to clear a single audited financial disclosure
VAST is a Beijing-founded AI 3D foundation-model and content-generation company established in March 2023. Its “Tripo” model family and companion SaaS workflow product, Tripo Studio, convert text, image, and sketch prompts into production-grade 3D assets. We flag upfront that the legal entity name is reported inconsistently across sources (see the data-integrity note below); we treat this as unresolved pending corporate-registry verification rather than paper over the discrepancy.
Born 1997, Song read international relations and economics at Johns Hopkins University — a non-technical academic background for a foundation-model founder — with a stated side interest in theology. He joined SenseTime’s CEO office before graduation in 2018, building the company’s AI-animation and AI-gaming businesses from zero to one; from 2020 he ran day-to-day operations for the AI gaming unit, which was co-led at the time by Yan Junjie, who later co-founded MiniMax. In 2021, Song joined MiniMax as employee No. 001 and an early co-founder, owning external business ecosystem, data, and compute partnerships. He founded VAST in March 2023 with the stated ambition of collapsing the cost of 3D content creation. He was named to the Forbes 30 Under 30 Asia list, and the company positions him as the first Chinese founder to deliver a mainstage keynote at SIGGRAPH — a claim we have not independently verified.
Tsinghua University-trained through the doctoral level. Previously led general vision and large-language-model efforts at SenseTime, with 50+ published papers, 100+ patents, and citations approaching 10,000. Owns VAST’s foundation-model architecture end to end.
Tsinghua BS and PhD (advised by Shi-Min Hu). Previously led generative-3D research at Tencent’s ARC Lab and AI Lab, and co-founded 3D holography company Owlii, later acquired by Kuaishou. Author of 70+ papers spanning Dream3D, DreamAvatar, and Triplane-Gaussian Splatting.
VAST’s revenue architecture combines its Tripo foundation-model family (Tripo 1.0 through 3.0 and the P-series) with the Tripo Studio workflow product and API-based B2B integrations. Consumer and prosumer subscriptions run $20–$140 per month with usage-based token billing layered on top. The company runs a parallel open-source track — TripoSR (co-released with Stability AI), TripoSG, and TripoSF, among 30+ published projects — as a developer-acquisition channel alongside the paid product.
Analyst take: the usage and dataset metrics below vary by as much as 5–7x depending on the reporting window and outlet (July 2026 press vs. the company’s own August 2026 disclosure vs. IPO-linked coverage in August 2026). Definitions — registered users, monthly actives, API developers — are not held constant across disclosures. We treat these as directional indicators of momentum rather than reliable point-in-time KPIs.
Converts text, image, or sketch inputs into geometry and PBR textures in roughly 10 seconds each. Iterated from Tripo 1.0 through P2.0 (unveiled September 2026) at multi-billion-parameter scale, with target-polygon-count output engineered to drop directly into Unity, Unreal, and other engines without a manual retopology pass — the company’s stated differentiator versus rougher open-source outputs.
A one-stop production layer spanning generation, texturing, rigging, and engine hand-off. Management frames this as lifting raw model output from an “80” to a “95” on a quality scale, with the explicit goal of widening the creator funnel from professional (PGC) down to casual (UGC) users.
Unveiled alongside the September 2026 Series B/B+ close. A three-layer architecture decouples state inference from visual rendering — a structured 3D state base layer, a condition-transformation middle layer, and an on-demand rendering top layer — targeting simulated training environments for embodied AI, digital twins, and autonomous-driving use cases. Management describes this as the first autonomous, deterministic maintenance of world state globally; this is a company claim at a pre-productization stage and is not independently verified.
B2B pipeline penetration: NetEase’s open-world RPG Where Winds Meet uses a Tripo-API-based feature (“Everything Taiji”) that lets players upload a photo and receive a real-time 3D avatar; NetEase’s casual title Egg Party added a Tripo-powered model-splitting feature to its creator workshop in July 2026. Coverage of art-pipeline integration at Tencent and Microsoft comes from a single Chinese outlet (163.com, September 2026); Western primary reporting (Bloomberg and peers) corroborates only the Sony relationship, so the true customer roster carries some cross-source variance. On the 3D-printing hardware side, VAST has secured coverage across major manufacturers including Bambu Lab, Creality, Anycubic-adjacent player Zongwei Cube, and iFun3D, and announced a strategic tie-up with HeyGears’ full-color print platform HeyVerse in July 2026.
Roughly three years post-founding, VAST closed at least three financing tranches between March and September 2026 totaling approximately RMB 5bn (~$744M) on a cross-sourced basis. Round labeling is inconsistent across outlets — Series A, A+, A2, A3, Pre-A+, Series B, and B+ are all used to describe overlapping tranches — which makes it difficult to reconstruct a clean cap table from public disclosures alone. We treat this labeling churn as a qualitative signal of a compressed diligence cycle rather than a mere reporting artifact.
Simon Song founded VAST in Beijing with a team drawn from SenseTime and MiniMax. Tripo 1.0-generation models began rolling out from early 2024 as product development ramped.
State-linked Beijing Artificial Intelligence Industry Investment Fund led, with Jingya Capital following on. Existing shareholders of record included Oasis Capital, Decheng Capital, Chunhua/Primavera Venture, Eminence Angel Fund, and a Tsinghua alumni seed fund. Tripo Studio launched formally around this period.
Alibaba Group and Hengxu Capital co-led, with Baidu Ventures following on. Valuation was not disclosed. This tranche is the point at which both of China’s dominant internet platforms — Alibaba and Baidu — held simultaneous positions in the company’s cap table.
Ince Capital and a China Life-affiliated venture fund co-led, with Genesis Capital, Eminence Ventures, and Primavera Venture Partners participating. Press coverage placed VAST’s valuation above $1B on the back of this tranche, though the company has never confirmed a precise figure. Gaming and industrial capital — Geely Capital, 4399, Giant Network, and Fosun — joined shortly after, marking one of the first instances of Chinese gaming-industry capital deploying into AI 3D generation.
Bloomberg reported early-stage discussions with Bank of America and CICC on a potential share sale. Deal size, timing, and whether a listing proceeds at all remain unresolved; the company and both banks declined to comment.
Round composition: Matrix Partners China led, joined by industrial investors Perfect World, BlueFocus, Core Dynamic Investment, ThunderSoft, a state radio-and-television investment vehicle, and 37 Interactive Entertainment, alongside financial investors CDH VGC, CICC Capital, CMC Capital Partners, A+H Capital, Zhongping Capital, Fujian-affiliated provincial funds, Oriza Asia, and Jiangxi Financial Holding. Existing shareholders Dachen Caizhi, Primavera, 4399, Muhua Tech, Oasis Capital, and Decheng Capital followed on.
Concurrent disclosure: VAST unveiled its flagship Tripo P2.0 model alongside the world-model roadmap Project Eden — layering optionality in embodied AI and autonomous-driving simulation on top of a core 3D-generation business whose unit economics remain largely undisclosed.
VAST’s positioning rests on four pillars: (1) an execution track record drawn from SenseTime, MiniMax, and Tsinghua; (2) a data flywheel in which user-generated content feeds back into foundation-model retraining; (3) B2B channel lock-in via broad participation from gaming and media industrial capital; and (4) option value from layering a world-model roadmap atop the core 3D-generation business. None of these constitutes a structurally durable moat that later entrants cannot replicate on a comparable timeline — they read more as a relative advantage built on capital intensity and execution speed, which should be weighed against the competitive risks in the following section.
The CEO’s track record spans building SenseTime’s AI-gaming business from zero to one and co-founding MiniMax as employee No. 001; the CTO and Chief Scientist hold Tsinghua doctorates and prior tenure at SenseTime and Tencent’s research labs respectively. We flag separately, under Risk Factors, that VAST is the CEO’s first experience as sole operating leader.
3D assets and prompts generated through Tripo Studio are intended to feed back into foundation-model retraining. Cumulative dataset figures, however, swing from 30M to nearly 100M generated assets depending on the disclosure window, which makes it premature to size the quantitative data advantage with confidence.
Live production use by NetEase (Where Winds Meet, Egg Party) and the roster of industrial investors in the September 2026 round — Perfect World, 37 Interactive, BlueFocus — signal that China’s gaming and media sector views this technology as adoption-ready. Full-line coverage of 3D-printing hardware partners reinforces B2B channel lock-in.
Conceptually adjacent to Fei-Fei Li’s World Labs (which raised $1B), Project Eden layers an option on embodied AI and autonomous-driving simulation — a materially larger TAM — on top of a core business that claims $12M in ARR. It remains pre-productization, and we would apply a meaningful discount before reflecting it in any valuation framework.
Tencent’s free, open-source Hunyuan3D can be self-hosted on commodity GPU infrastructure, exerting structural downward pressure on VAST’s subscription model. Competitor Meshy claims to outperform Tripo 3.1 on single-view geometry-alignment benchmarks — a claim from an interested party and subject to bias — while Microsoft’s open-source Trellis 2 also scores well on visual quality. In our view, VAST’s edge is better characterized as relative advantage in productization, distribution, and fundraising velocity than as durable technical superiority on any single benchmark.
Alongside genuinely strong signals — fundraising velocity and broad industrial-capital participation — VAST carries a number of open items across data integrity, geopolitics, competitive dynamics, and governance that warrant institutional-grade diligence. The items below are ordered by category, not severity.
Some outlets list the legal entity as “Shanghai Washite Technology Co., Ltd.” (Shanghai-domiciled); others list “Beijing Sanqi Wanwu Technology Co., Ltd.” (Beijing-domiciled). We do not resolve this pending cross-verification against public corporate-registry filings.
User counts (2M–20M), cumulative generated assets (30M–~100M), and ARR ($12M, single-sourced) all show material variance across reporting windows and outlets, and no audited financial statements are public. Some variance is inherent to a private-stage company, but a prospective IPO candidate should be expected to converge on consistent metric definitions ahead of any listing process.
With overseas revenue reportedly exceeding 80% of the total and the US described as the largest user base, a China-domiciled AI software company carries exposure to US data-security review, potential app-store access restrictions, and the possibility of expanded export-control scrutiny.
Tencent’s free open-source Hunyuan3D, alongside competing benchmark claims from Meshy and Microsoft’s Trellis 2, keeps sustained pressure on the subscription model’s defensibility. Technical benchmarks alone do not establish durable differentiation.
A sequence of tranches labeled A/A+/A2/A3/B/B+ within six months, with inconsistent naming across outlets, is consistent with a compressed diligence cycle. This embeds down-round risk should the next disclosed valuation fail to clear the implied step-up from the September 2026 round.
VAST is the CEO’s (b. 1997) first experience as sole operating leader. With the core 3D-generation business apparently still pre-breakeven, running Project Eden as a parallel, large-scale R&D track concurrently introduces execution risk from divided organizational focus.
Bloomberg’s early-August 2026 report on a Hong Kong listing reflects preliminary discussions with Bank of America and CICC. Deal size, timing, and whether the listing proceeds at all remain unresolved. Whether the September Series B/B+ round is running in parallel with a listing process or represents an alternative path cannot be determined from public information at this time.