DeepMind alumni behind a $25bn threat to OpenAI

By the Editorial Team · Monday, October 19, 2026

Reflection AI, a U.S. start-up led by former Google DeepMind researchers, launched Beam, a downloadable open-weight AI model designed for business and government use that competes with major AI providers like OpenAI and top Chinese firms. By allowing users to run the model locally, Beam addresses data privacy concerns and aims to capture growing demand for affordable, customizable AI solutions amid geopolitical tensions and regulatory challenges.

An American artificial intelligence start-up founded by former Google DeepMind researchers is challenging leading AI companies with the release of a new open-weight model designed for business and government applications. Reflection AI, established two and a half years ago, introduced its model, Beam, last Friday, aiming to offer an efficient and cost-effective alternative to offerings from major U.S. firms like OpenAI and Anthropic, as well as Chinese competitors.

Unlike many current AI models that operate as closed systems—requiring users to access them through cloud-based services managed by companies such as OpenAI—Beam is an open-weight model. This means that users can download, customize, and run the model locally on private hardware, addressing concerns around data privacy and security associated with sending sensitive information to third-party servers. Reflection AI generates revenue by charging for the infrastructure setup necessary to operate Beam and offers support services for AI training.

China has so far led the market in open-weight models, but Reflection AI is part of an emerging push in the West to develop competitive alternatives amid geopolitical and regulatory challenges related to Chinese technology systems. Testing by Reflection AI indicates Beam performs on par with GLM-5.2, a top Chinese model developed by Z.AI, particularly in advanced reasoning capabilities.

Ioannis Antonoglou, 38, Reflection AI’s co-founder and chief technology officer, speaking from the company’s Manhattan office, stated that the goal is for Beam to match the performance of closed models within the next year. He noted a recent shift among businesses moving workloads toward more affordable open models and predicted that once open models achieve parity with closed ones, widespread adoption will follow.

Since its launch, Beam has secured more than 40 distribution partners, including major cloud service providers such as Amazon Web Services, Google Cloud, Dell, and CoreWeave. At an event in London last week, Nvidia CEO Jensen Huang encouraged the United Kingdom to leverage open AI models, highlighting their potential for national data sovereignty.

Nvidia, which backs Reflection AI, is reportedly in discussions to broaden its investment or to acquire the company, which was valued at $25 billion during an April funding round. Misha Laskin, Reflection AI’s co-founder and chief executive, declined to comment directly on talks with Nvidia but emphasized the company’s focus on developing robust open models and building a long-term business serving enterprise and sovereign customers.

Laskin has urged businesses and governments to consider the risks of relying on closed AI models, comparing it to renting from a landlord, and suggested that owning and controlling open AI infrastructure will become the preferred approach as the technology advances.

This article was reported by synthesizing coverage from multiple independent news outlets to present a balanced, unbiased account. Read about our editorial standards. Spotted an error? Let us know.