U.S. artificial intelligence model developers are fighting China on two fronts. On one side, they are releasing cheaper open models to challenge Chinese AI developers that have expanded their footing through low-cost, open strategies. On the other, they are raising large sums to build financial muscle. The push also aligns with the U.S. government's drive to export American AI.
Reflection AI, founded by two former Google DeepMind researchers, unveiled a new open-weight model called Beam aimed at China, Bloomberg reported on the 5th, local time. Designed to perform in coding and AI agent tasks for corporate clients and developers, the model is priced below many competing Chinese models. The company said Beam scored similarly to GLM-5.2, the flagship open model from China's Z.AI, on advanced reasoning benchmarks.
Major U.S. AI model companies, including OpenAI and Anthropic, have focused on closed frontier models. Chinese developers such as DeepSeek and Moonshot AI, by contrast, have gained ground by releasing cheaper open-weight models that can be customized.
Beam is a large mixture-of-experts coding, reasoning and agent model that operates in a manner similar to K3, a version of Moonshot's Kimi model. Mixture of experts, or MoE, splits a model with a huge number of parameters into smaller modules and activates only the portions needed. That improves efficiency and optimizes the model for professional-grade reasoning and coding work.
Reflection AI's new model is in step with Washington's efforts to export U.S. artificial intelligence overseas. Earlier this year, Reflection AI joined a U.S.-brokered project to build sovereign AI in South Korea. The company agreed to partner with Shinsegae Group to pursue construction of what would be the country's largest AI data center.
Reflection AI, which raised $2 billion from Nvidia and other investors in October last year, has sharply increased its valuation this year. The Wall Street Journal reported in March that the company was discussing a $2.5 billion raise at a pre-money valuation of $25 billion, and the company said the fundraising succeeded. To secure the computing resources needed to develop and run its models, Reflection AI signed a contract worth about $6.3 billion with SpaceX in June and another worth more than $1 billion with neocloud provider Nebius Group in July.
OpenAI Valued at $1.4 Trillion, Up 60% in Seven Months
Fundraising is also accelerating at established AI model developers that have stuck with closed models. OpenAI is pursuing a $30 billion round, holding talks with several United Arab Emirates investment funds including Abu Dhabi-based MGX. Those funds are expected to commit up to $10 billion. BlackRock is also in discussions about co-investing.
OpenAI is currently valued at about $1.4 trillion. In March, the company raised $122 billion at a valuation of $852 billion, meaning its valuation has grown more than 60% in seven months. OpenAI has also been in talks with existing investors, including the University of California endowment, Thrive Capital and Andreessen Horowitz, about joining the round. Citing a focus on ensuring AI safety, the company has pushed back any plans for an initial public offering to next year at the earliest.
