NVIDIA in talks to acquire Reflection AI: Is it building an American version of DeepSeek?

[Financial News]  

The Financial Times (FT) reported on the 10th (local time), citing sources, that NVIDIA is in talks to acquire U.S. open-source artificial intelligence (AI) startup Reflection AI. If an acquisition proves difficult, the company may also consider deepening its investment.

The Trump administration sees Reflection as a potential rival to Chinese companies that make low-cost AI models such as DeepSeek. The company works with the U.S. Department of Defense and Department of Energy, and is also developing AI models with U.S. allies, including South Korea.
According to sources, discussions are still at an early stage, and several options are being considered. One would have NVIDIA hire Reflection’s employees and license its technology instead of acquiring the company outright and undergoing regulatory review. This is known as an “acqui-hire.”
NVIDIA is already one of Reflection’s largest shareholders and could strengthen ties between the two companies through an additional equity investment or by providing more chips and computing capacity.
Sources expect an agreement within the coming weeks. Reflection was valued at $25 billion (about 33.6 trillion won) during a fundraising round last March.

The acqui-hire approach NVIDIA is considering has become popular among Big Tech companies because it can avoid regulatory scrutiny. NVIDIA used the same approach last December when it acquired semiconductor design company Groq for $20 billion.
NVIDIA, the world’s largest company by market capitalization, has made massive investments over the past two years to expand the AI ecosystem and ensure it remains at its center. It already has its own open-source AI model, NVIDIA Nemotron, but does not have its own frontier AI model. Reflection could fill that gap.

NVIDIA has been fixated on open-source AI models in recent months. Open-source AI models allow users to modify and run prototype models on their own hardware. They enable companies to run AI at far lower cost than with models such as OpenAI’s ChatGPT and Anthropic’s Claude.
In July, NVIDIA CEO Jensen Huang stressed that the United States needed to build a homegrown open-source “ecosystem,” saying it was essential to “create opportunities for innovation and prosperity across the United States.” Last September, NVIDIA agreed to acquire Hugging Face for $13 billion. The platform serves as a repository for millions of models and datasets.
NVIDIA has so far poured tens of billions of dollars into large AI startups such as OpenAI, neocloud providers such as Nebius, and a range of smaller AI companies. Its strategy is to expand the AI ecosystem and generate demand for its chips on its own.

Reflection, which is at the center of the current acquisition speculation, is seen as a key strategic asset in the United States’ effort to compete with low-cost AI models such as DeepSeek. Last week, it unveiled its first AI model, which individual users can customize to their needs. The model runs on NVIDIA chips. Reflection claims it performs on par with GLM-5.2 from China’s Z.ai, a leader in the field.
Reflection founders Misha Laskin and Ioannis Antonoglou played key roles in developing Gemini, Google’s AI model.
NVIDIA has already invested $800 million in the startup. Venture capital firms including Sequoia Capital, Disruptive, and 1799 Capital have also invested in Reflection, and Trump Jr., the son of President Donald Trump, is a partner at the company.

[email protected] Song Kyung-jae Reporter