供资
Fal 宣布系列 F 由贝塞默领导，用于扩展AI推论能力
在 Google 上的首选来源中添加 Unite.AI
基因媒体基础设施公司fal 宣布于2026年11月3日作为战略投资人由Bessemer Venture 与……合作 NVIDIA领导的系列F，同时也披露了由Sequoia Capital领导的系列E于2026年早些时候关闭。
这些回合将在2027年资助跨沙漠生态系统的工作，引导资本用于两个相关需求：大规模运行模型的推论能力和员工开发未来产品。 支持还旨在为研究和新形式的创造性工作创造空间。
客户管道和研究基础设施
公告中描述的客户基础跨越娱乐工作室，广告代理，电子商务品牌，游戏开发商和营销团队。 客户在生产完工的管道中运行图像、视频和音频模型，包括Fal自己的模型。 在2026年期间，Fal的技术工作还包括训练后现有的模型，以测试质量和速度能否一起提高。
它的无服务器文件描述了客户自己的型号、管道和应用的部署系统，包括飞毛腿处理供给、缩放、联网和可观察性。 运行者可以随着需求的上升而从零到千，当交通停止时又回到零。 客户保持对其代码，模型重量和容器环境的控制，而市场模型API则在同一没有服务器的基础之上运行。
对于持续的工作量，fal计算提供专用的NVIDIA H100 SXM实例，这些实例仍在客户控制下运行。 该产品设计用于培训，扩展微调，批次加工和研究，而不是点播自动升级。 八GPU实例可以放置在同一部门，并在InfiniBand上连接，用于分布式工作，使Fal计划的研究方与波动的推论流量不同。
投资者连续性和公司增长
新公告还延续了Fal2025年12月9日D系列公告中记载的顺序。 该回合融资1.4亿美元，并增加了塞夸亚，克莱纳·珀金斯和NVIDIA作为新投资者。 它宣称的目的是在全球范围扩大该平台。 Sequoia随后领导了E系列与F系列同时披露，而NVIDIA在最新一轮中作为战略投资者回归。
在D系列时，fal有70名员工，正在招聘跨工程，产品，设计，上市和运营。 最新公告再次将团队投资与基础设施联系起来，这次明确包括研究与服务客户的系统。
Burkay Gur和Gorkem Yurtseven在Coinbase和Amazon期间看到AI-基础设施挑战后，于2021年创立了Fal. 公司总部位于旧金山，拥有全球分布的团队，其核心技术焦点是使基因媒体推论更快和更符合可扩展应用的成本效益。
2027程序将两个不同的基础设施需求联系起来：即时，可变推论流量和用于研究的持续计算。 其核心挑战是立即扩大这两个领域，支持目前的生产管道，同时给予研究和产品开发足够的能力，以便同时推进。
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
Funding
Fal Announces Series F Led by Bessemer to Expand AI Inference Capacity
Add Unite.AI to your preferred sources on Google
Generative-media infrastructure company fal announced on November 3, 2026, a Series F led by Bessemer Venture Partners with NVIDIA as a strategic investor, while also disclosing a Series E led by Sequoia Capital that closed earlier in 2026.
The rounds will fund work across fal’s ecosystem in 2027, directing capital toward two connected needs: inference capacity to run models at scale and the employees developing future products. The backing is also intended to create room for research and new forms of creative work.
Infrastructure for Customer Pipelines and Research
The customer base described in the announcement spans entertainment studios, advertising agencies, e-commerce brands, game developers and marketing teams. Those customers run image, video and audio models, including fal’s own models, within pipelines that produce finished work. During 2026, fal’s technical work also included post-training existing models to test whether quality and speed could improve together.
Its Serverless documentation describes a deployment system for customers’ own models, pipelines and applications, with fal handling provisioning, scaling, networking and observability. Runners can scale from zero to thousands as demand rises and return to zero when traffic stops. Customers keep control of their code, model weights and container environment, while marketplace Model APIs operate on the same Serverless foundation.
For sustained workloads, fal Compute instead supplies dedicated NVIDIA H100 SXM instances that remain running under customer control. The product is designed for training, extended fine-tuning, batch processing and research, rather than on-demand autoscaling. Eight-GPU instances can be placed in the same sector and connected over InfiniBand for distributed jobs, giving the research side of fal’s plan a different operating model from fluctuating inference traffic.
Investor Continuity and Company Growth
The new announcement also extends a sequence documented in fal’s December 9, 2025, Series D announcement. That round raised $140 million and added Sequoia, Kleiner Perkins and NVIDIA as new investors. Its stated purpose was to scale the platform globally. Sequoia subsequently led the Series E disclosed alongside Series F, while NVIDIA returned as a strategic investor in the newest round.
At the time of Series D, fal had 70 employees and was recruiting across engineering, product, design, go-to-market and operations. The latest announcement again connects team investment with infrastructure, this time explicitly including research alongside the systems that serve customers.
Burkay Gur and Gorkem Yurtseven founded fal in 2021 after seeing AI-infrastructure challenges during their time at Coinbase and Amazon. The company is headquartered in San Francisco with a globally distributed team, and its central technical focus is making generative-media inference faster and more cost-effective for scalable applications.
The 2027 program links two different infrastructure demands: immediate, variable inference traffic and the sustained computing used for research. Its central challenge is to expand both at once, supporting current production pipelines while giving research and product development enough capacity to advance alongside them.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.