内容来源：https://aibusiness.com/generative-ai/ai-startup-runway-raises-315m-for-world-models

内容总结：

AI初创企业Runway转型世界模型，估值激增至53亿美元

近日，AI视频生成领域的初创公司Runway宣布完成新一轮融资，公司估值跃升至53亿美元，较去年增长逾20亿美元。本轮融资由General Atlantic领投，英伟达、Adobe Ventures等多家知名投资机构跟投。

Runway成立于2018年，最初以视频生成技术闻名，并于2023年发布了Gen-1视频生成模型。然而，该公司正将战略重心转向当下企业级AI市场备受关注的“世界模型”。这类基础模型通过学习物理世界的运作规律，能够预测因果关系、物理效应及空间动态，从而在虚拟环境中高精度模拟现实。

公司计划将新获资金用于预训练新一代世界模型。去年12月，Runway已推出面向交互环境、机器人及数字人的通用世界模型GWM-1。这一转型反映了企业客户对能够提供高度预测性和安全模拟的AI技术的强烈需求。

市场分析机构Omdia的分析师苏联杰指出：“随着世界模型精度的提升，基于其训练的应用几乎可以确保在现实世界中安全运行。这对于功能实现、安全合规具有重要意义。”他认为，能够提供如此精准的物理世界表征，对企业而言价值重大。

目前，世界模型赛道竞争日趋激烈。除Runway外，科技巨头如英伟达已于2025年1月发布Cosmos世界模型平台，谷歌DeepMind也在去年夏季推出了通用世界模型Genie 3。该技术因其卓越的真实性与准确性，正被广泛应用于医疗健康、自动驾驶、机器人等多个行业。

分析师进一步表示，随着AI技术不断进步和企业应用门槛降低，未来物理世界与虚拟世界的融合交互将更加深入，空间计算等技术有望嵌入更多应用场景，从而为各行业开启新的可能性。Runway的此次战略转向，正是顺应了这一市场趋势。

中文翻译：

由谷歌云赞助
选择您的首个生成式AI应用场景
要开始应用生成式AI，首先应关注能够改善人类信息交互体验的领域。

自2023年起，该供应商便专注于视频生成领域。向世界模型的转型反映出企业对这类先进物理AI模型日益增长的兴趣。
AI视频初创公司Runway的新一轮融资标志着其战略重心转向一类基础模型——这类模型因其高度拟真性与精确性，在物理AI市场中正迅速走红。

此次融资使这家AI初创公司的估值跃升至53亿美元，较不到一年前增长逾20亿美元。
资金将助力这家成立于2018年的企业聚焦世界模型——这类基础模型因其能在多元场景中提供可预测性，正日益受到企业青睐。世界模型通过训练学习物理世界的运行规律，从而预测物理现象、因果关系与空间动态。

这家总部位于纽约的供应商计划将资金用于预训练新一代世界模型，这标志着其自2023年发布Gen-1视频转换模型以来，从专注视频模型向世界模型的战略转移。
Runway的战略转向凸显了世界模型在企业级AI市场中日益重要的地位。由于世界模型能在虚拟环境中模拟物理世界，其应用已覆盖医疗健康、自动驾驶汽车和机器人等多个行业。

该领域增长显著，Runway正与同样布局世界模型的谷歌和英伟达展开竞争。2025年1月，英伟达推出Cosmos世界基础模型系列，并发布Cosmos WFM平台，专为物理AI和机器人训练生成"具备物理感知"的视频。谷歌DeepMind则于去年夏季发布了通用世界模型Genie 3。Runway于12月推出面向交互环境、机器人与数字形象的通用世界模型GWM-1。

世界模型应用的增长源于其能更精准预测物理世界的发展轨迹。
"随着世界模型精度的提升，几乎可以确信基于该模型训练的内容能在现实世界中安全运行，"Informa TechTarget旗下研究机构Omdia分析师苏连杰表示，"从功能、安全与合规角度来看，这具有深远意义。"

"能够提供如此精准的物理世界模拟对企业至关重要，"他补充道。
苏连杰指出，尽管Runway此前一直是专注于企业的视频生成供应商，但其进军世界模型领域意味着客户存在相关需求。

"未来我们或将见证物理与虚拟世界更深入的融合，"他举例说，"例如将空间计算嵌入其他技术领域的可能性。"
"得益于AI技术的进步与企业应用门槛的降低，将有更多企业尝试这项技术，"苏连杰强调，"这将为企业领域开辟新的可能性。"

本轮融资由泛大西洋投资集团领投，英伟达、Adobe风投、联博基金、AMD风投以及富达管理与研究公司参与跟投。

英文来源：

Sponsored by Google Cloud
Choosing Your First Generative AI Use Cases
To get started with generative AI, first focus on areas that can improve human experiences with information.
The vendor has focused on video generation since 2023. The transition to world models reflects enterprises' growing interest in these advanced types of physical AI models.
AI video startup Runway's new fundraising round signals its shift in orientation to a type of foundation model that has been growing in popularity in the physical AI market for its realism and accuracy.
The new funding bolstered the AI startup’s valuation to $5.3 billion, $2B more than it was less than a year ago.
The funding will help the 2018 startup focus on world models, a type of foundation model that is proving popular with enterprises for the predictability it provides across different settings. World models are trained to learn how the physical world works and to predict physics, cause-and-effect, and spatial dynamics.
The New York-based vendor plans to use the capital to pretrain the next generation of world models, a shift from its focus on video models since it released the Gen-1 video-to-video model in 2023.
Runway’s redirection highlights the growing significance of world models in the enterprise AI market. Because world models enable the simulation of the physical world in a virtual environment, they are used across industries such as healthcare, autonomous vehicles and robotics.
The area has seen significant growth, and Runway competes with Google and Nvidia, which also have world models. In January 2025, Nvidia introduced Cosmos World Foundation models and launched the Cosmos WFM platform to generate “physics-aware” videos for training physical AI and robots. Google DeepMind released its general-purpose world model, Genie 3, last summer. Runway introduced GWM-1, a general-world model for interactive environments, robotics, and digital avatars, in December.
The growth in the use of world models stems from their ability to provide greater predictability about what will happen in the physical world.
“As the level of accuracy increases with the world model, you can almost be sure that whatever has been trained with that model can work in a safe manner in the real world,” said Lian Jye Su, an analyst at Omdia, a division of Informa TechTarget. “That has a lot of implications from a functional, safety and compliance standpoint.”
“Being able to offer that level of accurate representation of the physical world is quite significant to enterprises,” Su continued.
He added that although Runway has been up to now a video generation vendor focused on enterprises, the fact that it sees an opportunity in world models means that its clients demand it.
“As we move forward, we’re probably going to see a lot more cross-pollination between the physical and virtual world,” Su said. For example, there might be opportunities to embed spatial computing into other types of technology, he said.
“Because of the advancement in AI, because of the low barrier to entry for enterprise, we're going to see a lot more enterprises trying this technology,” Su added. “That opens up new possibilities in the enterprise sector.”
General Atlantic led the funding round. Nvidia, Adobe Ventures, AllianceBernstein, AMD Ventures and Fidelity Management and Research Company participated.