NeuroAI NEUROAINEUROAI.SITE
China AI, reported clearly
1,300+ verified stories on models, robotics, BCI and the business of AI.
Browse stories
ESC

After 27 Years at Microsoft Research Asia, Xin Tong Takes the Chief Scientist Seat at Meshy

Xin Tong (童欣), a computer-graphics researcher who spent 27 years at Microsoft Research Asia, became chief scientist at AI 3D startup Meshy in September 2026 — and is now pushing it toward interactive 'world models.

2026-11-19 · 1063 words · NeuroAI
After 27 Years at Microsoft Research Asia, Xin Tong Takes the Chief Scientist Seat at Meshy

A designer types a sentence, and a character appears in three dimensions, ready to drop into a game engine. Ten years ago, that was a research demo. Today it is a product with more than 12 million registered users, by the company's own count.

The researcher who helped build the math behind it has changed jobs. In early September 2026, Xin Tong (童欣) became chief scientist at Meshy, a Chinese-founded company that turns text and images into 3D assets.

It is a quiet appointment by the standards of the AI industry — no product launch, no funding headline. But it says something about where computer graphics is heading, and about how Chinese labs are competing for the people who can build what comes next.

Key takeaways

  • Now: Xin Tong is chief scientist at Meshy, responsible for long-term research strategy. Previously: he spent 27 years at Microsoft Research Asia (MSRA), leading its Internet Graphics Group and serving as a Partner Research Manager (时任全球研究合伙人、网络图形组负责人).
  • He earned a PhD from Tsinghua University in 1999 and joined MSRA the same year.
  • The company's announcement cites more than 190 papers, over 60 in ACM SIGGRAPH or ACM Transactions on Graphics, and more than 21,000 citations with an h-index of 80, as of September 2026.
  • Meshy says it serves more than 12 million registered users and raised nearly US$400 million in a July 2026 Series B at a US$1.5 billion valuation. Both figures are company-stated.

Who Tong is, in one paragraph

Tong's field is computer graphics — the science of how computers represent, reconstruct, generate and render three-dimensional scenes. His better-known work includes O-CNN, an octree-based architecture that let deep learning process high-resolution 3D shapes, published at SIGGRAPH in 2017. More recent projects include TRELLIS, a structured 3D generative model; MoGe, which estimates geometry from a single image; and RenderFormer, which renders triangle meshes with global illumination.

Research from teams he led was deployed in Microsoft products including Xbox development tools, Windows 3D printing features and Direct3D. That blend of published theory and shipped engineering is exactly what an AI 3D company wants when it tries to move beyond making isolated objects.

What he will work on

Meshy's current business is generating a single 3D asset from a prompt. Tong's mandate is broader: next-generation 3D foundation models (基础模型), multimodal world models (世界模型) and real-time interactive systems (实时交互系统).

The distinction is worth spelling out. A useful 3D asset is like a prop. A world model is the entire set — geometry, appearance, structure, behaviour — that lets someone walk around inside a generated environment and change it. Meshy is already testing this through a project called Mora, short for Multimodal Open-world Real-time Architecture. Mora 1 opened to the public at mora.fun, combining coding agents, Meshy's own 3D generation and real-time video models.

In the company's announcement, Tong said that "computer graphics is entering a new chapter," with foundation models beginning to unify geometry, appearance, structure and semantics. He pointed to the harder challenge of keeping those elements consistent as things move and respond. That is a research agenda, not a shipped feature — fair to remember when reading the launch language.

Meshy's founder and chief executive, Ethan Hu (胡渊鸣), who holds a PhD from MIT, framed the hire as a step toward models that generate complete interactive environments rather than individual assets. Again: that is the company's stated ambition, not an independent assessment.

Why this hire is worth noticing

Two trends meet here.

The first is that 3D generation has stopped being a niche. Games, film, product design and 3D printing all need assets, and robots need spatial understanding to navigate the physical world. A researcher who spent 27 years making graphics both rigorous and usable is a natural fit for that convergence.

The second is a talent pattern across Chinese AI. Companies are pulling senior researchers out of established labs and giving them authority over long-term strategy, rather than hiring them as figureheads. Tong's remit — research direction, team building, plus moving findings into models and products — is an operating role, not an honorary one.

There is a caution attached. Announcing a chief scientist is cheap; building a world model that holds together under interaction is not. The Mora demo is an early validation, in the company's own words, which is another way of saying it is unfinished.

Honest limitations

Nearly all public coverage of this appointment traces to Meshy's own announcement, distributed through PR Newswire Asia and reprinted by China Daily Hong Kong, TMCnet and Chinese technology outlets. I have not interviewed Tong or Meshy, and there is no independent source confirming the appointment beyond the company's statement. Where I quote the announcement, I have said so.

The citation and publication counts are attributed to Google Scholar by the company; I have not checked them directly. The 12-million-user and US$400-million Series B figures come from the press-release boilerplate carried by TMCnet and other reprints, not from an audited filing. Meshy is a private company with limited public financial disclosure, and the valuation is a reported financing term rather than a market price.

I have not verified the exact date Tong left MSRA, so I have used "early September 2026" — the Chinese report says September 7 and the English release was published September 11. English renderings of Chinese titles and research-project names vary; originals are kept where they carry specific meaning.

What readers can do now

  • If you build with 3D tools, try the public demo before the press release. Mora 1 is open at mora.fun; judge the consistency of a generated world for yourself rather than from a description.
  • If you follow AI talent, watch the mandate, not the title. A chief scientist who controls research direction, hiring and product translation has a very different job from one who only lends a name.
  • If you are early in a graphics or spatial-computing career, note the skill mix. Tong's value comes from pairing deep theory with research that shipped in real products — a useful model for anyone entering the field.

The prop on the screen is already convincing. The room around it is still being built, and that is the part Tong has been hired to figure out.


Sources: Meshy announcement via PR Newswire Asia / China Daily Hong Kong, Sept 11, 2026; TMCnet reprint, Sept 8, 2026; China.com technology channel, Sept 8, 2026.

Related coverage

Cite this article

NeuroAI (脑机网), "After 27 Years at Microsoft Research Asia, Xin Tong Takes the Chief Scientist Seat at Meshy," 19 November 2026, https://neuroai.site/a/na-people-xin-tong-meshy-chief-scientist

Machine-readable: BibTeX · RIS · JSON · Markdown

More in “Builders” → · Back to home · Markdown version

China AI, reported clearly
1,300+ verified stories on models, robotics, BCI and the business of AI.
Browse stories