智东西（公众号：zhidxcom）
编译 | 孙小雯
编辑 | 程茜
智东西11月4日消息，今天凌晨，美国云推理平台Fal称已完成13亿美元（约合人民币87.33亿元）F轮融资，投后估值达160亿美元（约合人民币1074.88亿元），该轮融资于今年10月完成。今年4月，其E轮估值为80亿美元（约合人民币537.44亿元），本轮融资后估值翻倍。
据彭博社报道，Fal计划将新资金用于采购更多算力，并扩充目前185人的团队。公司今年夏天一度因无法满足客户需求，不得不限制客户请求量。
本轮融资由美国知名投资机构Bessemer Venture Partners领投，英伟达继续参投。
Fal成立于2021年，总部位于旧金山，核心业务是为创作者和开发者提供图像、视频及音频模型托管与推理服务，帮助客户运行模型、生成内容。
其两位联合创始人为Burkay Gur和Gorkem Yurtseven。首席执行官Gur此前曾负责Coinbase的机器学习业务，联合创始人兼首席技术官Yurtseven曾为亚马逊开发者。
▲Fal联合创始人Burkay Gur与Gorkem Yurtseven合影（图源：First Round Capital）
据彭博社援引知情人士消息，Fal今年早些时候的年化营收已达到8亿美元（约合人民币53.74亿元）。
据Bessemer发布的投资公告，Fal为开发者提供统一的模型调用API，为模型开发商提供分发服务，并为企业部署自有模型提供推理基础设施。
据其本轮领投方Bessemer披露，Adobe、Canva和Shopify等公司已在实际业务中使用Fal的服务。Fal官网称，其客户覆盖娱乐工作室、广告公司、电商品牌、游戏开发商和营销团队，相关模型已接入客户的内容生产流程。
Fal透露，其后训练合作方包括中国大模型独角兽MiniMax和德国视觉模型开发商Black Forest Labs。今年8月，Fal推出基于MiniMax H3后训练的H3 Max，优化提示词遵循能力、视觉质量和生成速度；10月，该公司又与Black Forest Labs合作推出FLUX 3 Max，继续优化视频生成质量与推理效率。
Fal称，H3 Max的训练和推理均使用英伟达GB200 NVL72系统，生成一段5秒视频约需3秒。
据彭博社报道，Gur称，客户需求激增是Fal此次大额融资的重要原因。今年夏天，公司一度无法满足客户需求，不得不限制客户请求量。Fal在官网透露，这两轮融资将支持其2027年的业务发展，并为团队开展新研究和内容创作探索提供资金。
结语：需求超出承载能力，推理平台通过融资支持扩张
今年7月，AI推理服务商Fireworks完成15.05亿美元（约合人民币101.11亿元）D轮融资，英伟达同样参投，资金将用于扩大工程团队和全球算力规模。客户需求增长，但承接这些业务需要先花钱采购算力、扩充团队，两家公司都选择通过融资支持扩张。
对Fal而言，融资要解决眼前的服务容量问题，与MiniMax、Black Forest Labs合作则着眼于提高模型生成速度和质量。广告、电商和游戏客户已将AI模型接入日常内容生产，平台限流、生成耗时和素材效果，都会影响客户的使用选择。
接下来，Fal需要把新增算力和模型优化转化为更稳定、高效的服务，争取这些客户持续投入的模型调用预算。
来源：彭博社、Fal、TechCrunch
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。
Intellectual (WeChat Official Account:zhidxcom)
Compiled by Sun Xiaoxuan
Editor, Concubine.
According to the source on 4 November, in the early morning hours of this morning, the American cloud-drive platform Fal claimed to have completed the $1.3 billion (approximately $8.733 billion RMB) F round of financing, valued at $16 billion (approximately $10,488 million RMB), which was completed in October this year. In April this year, its E round was valued at $8 billion (approximately 53,744 million yuan) and doubled after the current round of financing.
According to Bloomberg, Fal plans to use the new funds to purchase more money and to expand the current team of 185 people. At one point this summer, the company was unable to meet its clients ' needs and had to limit its requests.
The current round of financing was led by Bessemer Venture Partners, a leading United States investment agency, and continued with the participation of Weda.
Founded in 2021 and based in San Francisco, Fal ' s core business is an image, video and audio model hosting and reasoning service for creators and developers to help clients run models and generate content.
Two of its co-founders were Burkay Gur and Gorkem Yurtseven. Chief Executive Officer Gur was previously in charge of machine learning in Coinbase, and the co-founder and chief technical officer Yutseven was an Amazon developer.
Co-founder of Fal, Burkay Gur and Gorkem Yurtseven (source: First Round Capital)
According to information sources cited by Bloomberg, the annualization of Fal earlier this year amounted to $800 million (approximately RMB 5,374 million).
According to the investment announcement of Bessemerpublish, Fal provides a uniform model for developers to call API, a distribution service for model developers, and a reasoned infrastructure for enterprises to deploy their own models.
According to Bessemer, its own lead investor, Adobe, Canva and Shopify have already used Fal services in their operations. Fal official network claims that its customers cover entertainment studios, advertising companies, electric boards, game developers and marketing teams, and that the relevant models are connected to the customer ' s content production process.
Fal disclosed that subsequent training partners included Mini Max, a Chinese LLM unicorn, and Black Forest Labs, a German visual model developer. In August this year, Fal launched H3 Max, based on post-Mini Max H3, optimizing prompt compliance, visual quality and production speed; and in October, the company launched FLUX 3 Max in partnership with Black Forest Labs, continuing to optimize video production quality and reasoning efficiency.
Fal states that H3 Max ' s training and reasoning used the British GB200 NVL72 system to generate a five-second video for approximately three seconds.
According to Bloomberg, Gur claims that the surge in client demand is an important reason for Fal ' s massive financing. This summer, companies were once unable to meet their clients ' needs and had to limit their requests. Fal revealed on the Web that these two rounds of financing will support its business development in 2027 and provide funding for teams to undertake new research and content creation exploration.
Conclusion: Demand exceeds carrying capacity and the reasoning platform supports expansion through financing
In July of this year, the AI reasoning servicer, Fireworks, completed the $1.505 billion (approximately RMB 10.11 billion) D-round financing, with the same participation of Weida, which will be used to expand engineering teams and global computing. Client demand has increased, but taking over these operations requires first-time purchasing power and expanding teams, with both companies choosing to support expansion through financing.
For Fal, financing addresses the immediate problem of service capacity, while working with MiniMax, Black Forest Labs aims to increase the speed and quality of model generation. Advertising, electricians and game clients have access to the AI model for daily content production, platform rate limit, time-consuming generation and asset effects affect client choice of use.
Next, Fal needs to translate the added computing power and model optimization into a more stable and efficient service, and to obtain the constant input of these clients to leverage the budget.
Source: Bloomberg, Fal, TechCrunch
Auto-fetched and translated by AI Atlas. All rights belong to the original author.