Meshy raises nearly $400 million to build AI-generated interactive worlds

Why it matters

Meshy is using a proven 3D creation product to enter runtime game generation, where the potential market is larger and the technical risk rises sharply.

Ethan Yuanming Hu, founder and CEO of Meshy, has raised nearly $400 million in Series B funding to push the 3D generation platform beyond downloadable assets and into software that creates game mechanics and interactive worlds.

Meshy disclosed the financing on July 20th. Meshy said the round valued it at more than 10 billion yuan after the investment. The size and valuation were also reported by 36Kr. Meshy did not identify a lead investor or disclose ownership terms.

Hu brings an unusually relevant background to that expansion. Before building Meshy, he created Taichi, an open-source programming language for high-performance graphics, physical simulation and GPU computing. Hu studied computer science in Tsinghua University's Yao Class before earning a doctorate from MIT's Computer Science and Artificial Intelligence Laboratory. His doctoral work covered the same graphics, compiler and simulation problems that Meshy must solve if generated worlds are to run interactively on consumer hardware.

Money backing a larger bet

New investors in the round include IDG Capital, Matrix Partners China and Monolith Capital, alongside strategic investors that Meshy did not name. Existing backers Granite Asia, HongShan, BAI Capital and Source Code Capital invested again.

Meshy said the capital will fund multimodal model research and international expansion. The investor list gives Hu substantial resources for a technically expensive move into real-time generation, while the missing lead and undisclosed strategic investors leave the round's internal structure unclear.

Meshy's established product turns text, images and conversational instructions into textured 3D models that can be exported into game engines, design software and 3D-printing workflows. Meshy's website says its community includes more than 10 million creators. In March, Meshy said users had generated more than 100 million models and that annual recurring revenue had reached $30 million. Subsequent company-supplied figures put ARR above $40 million in April and registered users above 12 million.

Those figures have not been supported by audited financial statements. Meshy also claims revenue rose twelvefold over the previous year and monthly growth remained between 20% and 30% for two years. The figures help explain the size of the round, but they reveal little about customer concentration, retention, inference costs or how much revenue comes from individual subscriptions versus enterprise contracts.

From assets to mechanics

Hu has already started testing the larger thesis. At the Game Developers Conference in March, Meshy introduced Meshy Labs and demonstrated Black Box: Infinite Arsenal, a survivor-style game in which an AI system assembles weapons and behaviors from smaller mechanics during play.

The game's Steam listing says Black Box connects to an external AI service while running to generate weapon mechanisms and scripts. Players combine components such as missiles, storm clouds and black-hole generators, and the system translates those choices into new weapon behavior. The game remains listed as coming soon and has no public user reviews.

That project moves Meshy into a harder category than producing a static mesh. A generated asset can be inspected, edited and exported after a delay. A generated game system must preserve state, obey rules, respond quickly and remain playable across a long session. It also has to operate at a computing cost that makes commercial distribution viable.

Meshy's proposed architecture combines an AI model for mechanics with conventional graphics and AI video rendering. The mechanics model determines how the world behaves, while a mixed rendering stack produces the visual result. Hu has compared the approach to AI coding systems that generate instructions for conventional computers to execute. In Meshy's version, the output is a running world rather than a software application.

Hu's Taichi work matters here. Meshy must train multimodal models and make their outputs execute efficiently on GPUs, two specialties that usually sit in separate organizations. Hu spent his academic career working on the execution side before turning Meshy into a commercial generation product.

What the round has to prove

Meshy has a distribution advantage that most experimental world-model projects lack: millions of users already arrive to create 3D content. Hu can use that product as an entry point for tools that generate environments, behaviors and playable systems, then sell those capabilities through subscriptions, enterprise integrations or game products.

The Series B funds the jump from a creation tool into a runtime platform. That jump carries much greater technical and commercial risk. Black Box has demonstrated dynamic weapon generation, but Meshy has yet to show that its architecture can sustain coherent environments, balanced mechanics and persistent state at the quality and cost required for a durable commercial game.

Nearly $400 million gives Hu time to attack those problems. It also sets a high threshold: Meshy's asset business now has to finance and distribute a second product category whose economics, computing requirements and audience remain unproven.