Illustration by Megaton
Image: Illustration by Megaton

Technology

Singapore startup bets world models can outrun video generation giants

By Julius RobertThursday, June 25th 20263-minute read

A former Tencent AI chief raised $80M to build real-time 3D worlds, with AMD and Hyundai backing the thesis

Share

A former Tencent AI chief raised $80M to build real-time 3D worlds, with AMD and Hyundai backing the thesis

Liu Wei left one of China's largest technology companies to found Video Rebirth in Singapore, and the startup's first major product is not a video generator in any conventional sense. Olympus, the system Video Rebirth is building, is designed to produce real-time, interactive 3D worlds complete with sound, a different category of output than the clip-length video tools most competitors are shipping.

That distinction matters in a market where the funding numbers are getting hard to ignore. PixVerse, a regional rival, closed a Series C extension in July 2026 at a $2 billion valuation after raising $439 million. Video Rebirth's $80 million raise, first reported by Forbes in June 2026, is smaller by an order of magnitude. But the investor composition, with AMD and Hyundai among the backers, suggests the startup is pitching Olympus as infrastructure for industries that need simulated spaces rather than content pipelines.

What world models actually promise

In Video Rebirth's framing, Olympus is meant to model physical spaces well enough to generate them interactively in real time, rather than rendering a fixed video sequence. The practical gap between that ambition and current video generation tools is large. Most AI video systems produce deterministic outputs from a prompt, while a world model would need to respond coherently to inputs mid-generation.

Whether Olympus can deliver that at launch is unresolved. The company is targeting a late 2026 release, which gives it roughly two product cycles to close the gap with better-capitalized competitors.

The Singapore angle

Video Rebirth's Singapore base is not incidental. The city-state has positioned itself as neutral ground for AI ventures that want access to both Western capital markets and Asian manufacturing and distribution networks. Liu Wei's background at Tencent gives the company credibility in the latter, while backers like AMD connect it to the semiconductor supply chains that determine who can train and run large models at scale.

Editorial illustration for Singapore startup bets world models can outrun video generation giants
th AMD and Hyundai backing the thesis Liu Wei left one of China's largest technology companies to found Video Rebirth in Singapore, and the startup's first major product is not a video generator in any conventional sense.

ByteDance, which operates its own AI video tools, represents the most direct competitive pressure in the region. PixVerse's $2 billion valuation suggests investors believe there is room for multiple well-funded players, but the market is compressing quickly as the largest technology companies integrate video generation into existing platforms.

The Hyundai angle

AMD's involvement is legible: the chipmaker has a direct interest in backing startups that will consume GPU compute. Hyundai's participation is less obvious, and no reported source explains the investment rationale. One plausible read is that an automotive company sees world models as relevant to simulation platforms for autonomous systems or manufacturing design, though that remains inference. What is documented is that Video Rebirth has commitments from two large industrial companies alongside its AI funding, which gives it a different capital structure than a pure software startup.

Olympus's late 2026 launch window will be the first concrete test of whether the world model framing holds up against products already in users' hands.

Weekly AI Video Rankings

Get weekly leaderboard changes and AI video model launches delivered to your inbox.

Related Articles