A game studio needs a prop by tomorrow — a broken lantern, a stone goblin, a chair that doesn't exist yet. A decade ago that meant a 3D artist and a week of modelling. A Beijing startup now says a single sentence is enough.
For everyone outside game and film production, the interesting part isn't the demo. It's that one slice of Chinese generative AI has stopped chasing chatbots and started building the layer underneath physical worlds.
The model that builds objects, not pixels
VAST (三启万物), the Beijing company behind Tripo, was founded by Simon Song (宋亚宸), a former co-founder of MiniMax. Its first Tripo model shipped in early 2024 and did something narrow but useful: turn a text prompt or a 2D image into an editable 3D mesh in seconds.
The lineage since then reads like a product ladder:
- Tripo 1.0 (2024) — text/image to editable 3D mesh.
- TripoSR — an open-source image-to-3D model built with Stability AI, pushing generation to about 0.5 seconds.
- Tripo 2.0 (September 2024) — the team's argument that 3D follows the same scaling law as language models.
- Tripo 3.0 (August 2025) — a larger model the company says moved 3D generation from "toy" to usable in real pipelines.
- Tripo P1.0 (March 2026) — native 3D-space probabilistic generation, a finished mesh in about two seconds.
- Tripo P2.0 (announced 1 September 2026) — native quad-face topology, the format game engines actually expect.
Tripo ships two ways: Tripo Studio, a web workbench for creators, and a Tripo API for studios that want to plug generation into their own pipelines.
The money behind the mesh
On 1 September 2026, VAST closed its Series B and B+ rounds totalling about 3 billion yuan (≈ US$446 million / about HK$3.3 billion), led by Matrix Partners China (经纬创投). The roster reads like a customer list: game publishers Perfect World (完美世界), 37 Interactive Entertainment (三七互娱) and Yanqu Games (燕趣), plus financial backers CMC Capital, CICC and CDH.
That round capped a compressed fundraising run. Across the first nine months of 2026 VAST raised roughly 5 billion yuan (≈ US$744 million), and Bloomberg reported its valuation crossed US$1 billion by June. The proceeds are earmarked for 3D-native foundation models, training and inference compute, and commercialisation.
The signal is less the size than the speed: Chinese investors are funding 3D-native labs on a cadence closer to frontier model shops than to typical vertical-AI startups.
Why 3D is the quieter AI race
Text, image and video models get the headlines. But games, 3D printing, film, XR and robot training all need geometry — meshes, topology, materials — not pixels. That is exactly what Tripo outputs.
Real deployments already exist:
- NetEase's Eggy Party (蛋仔派对) uses Tripo for model-splitting in its creation workshop.
- Where Winds Meet (燕云十六声) ships Tripo-powered building blocks.
- Bambu Lab (拓竹) integrated Tripo into its MakerWorld 3D-printing community.
The hard problem Tripo P2.0 is aimed at is the "pretty but useless" trap. AI-generated meshes have long arrived with chaotic wireframes and face counts too high to drop into a game engine without hours of manual cleanup. Quad-face topology is the company's answer: assets that can be bound, animated and edited without a human rebuilding them first.
What it still can't do
Two caveats are worth saying plainly.
First, the "industry-first native quad topology" claim is VAST's own, and no independent benchmark of P2.0's output quality has been published yet. The models are clearly being used in production, but the headline differentiators are vendor assertions.
Second, user and asset counts vary by source. VAST has cited figures ranging from millions of creators to more than 700 enterprise customers; treat those as directional, not audited. Tripo also competes with Meshy and Tencent's Hunyuan3D (混元3D), so "the leader" is a moving claim.
What readers can do now
- If you build games or XR, run a Tripo P2.0 asset through your normal import-and-cleanup step and measure the hours saved. The quad-topology pitch only matters if it shows up in your pipeline.
- If you invest or track AI, read the funding cadence as a category signal: 3D-native foundation models are now a capitalised lane, not a research curiosity.
- If you follow Chinese AI strategy, see Tripo as specialisation — labs branching from general chatbots into vertical generative niches with paying customers.
Honest limitations
Core facts (Tripo founding/products and timeline; 1 September 2026 Series B/B+ of ~3 billion yuan / ≈US$446M led by Matrix Partners China; ~5 billion yuan / ≈US$744M raised in six months; US$1B valuation by June 2026 per Bloomberg; Tripo P2.0 quad-topology claim; customers NetEase Eggy Party, Where Winds Meet, Bambu Lab) come from DealStreetAsia, InvestGame, PANews, The SaaS News, VentureCapitalTracker and the company's statements via 36Kr. The lead investor (Matrix Partners China) was confirmed across multiple independent outlets after one wire report misnamed it. The "industry-first" topology claim and all user/asset counts are VAST's own and are not independently audited. No RMB-amount conversion issues beyond those shown. Analysis is current to 28 September 2026 and is not investment advice.