NVIDIA has set its sights on the $25 billion AI unicorn

NVIDIA's AI landscape is likely to add another puzzle piece.

Citing industry sources, the Financial Times reports that NVIDIA is in acquisition talks with U.S. AI startup Reflection AI, and is also considering additional investments to further deepen cooperation between the two sides.

The transaction is still in the early stage of negotiations, and no definite plan has been reached yet.

People familiar with the matter disclosed that the transaction is expected to be finalized in the coming weeks, but the overall negotiation still has uncertainties, and the specific terms have not been made public.

From Shareholder to Buyer?

There are multiple possible forms of this transaction, including full acquisition, acqui-hiring, increased equity investment, and provision of additional chips and computing resources to Reflection AI.

The so-called "acqui-hiring" refers to NVIDIA hiring Reflection AI's employees and obtaining relevant technology licenses.

This method can quickly gather talents and technologies while avoiding the high transaction costs arising from a full acquisition. Compared with a complete merger and acquisition, acqui-hiring has lower integration difficulty and less regulatory resistance, but it still needs to undergo compliance review and cannot completely bypass supervision.

Data shows that Reflection AI was founded in 2024 by two former DeepMind researchers. It is a startup focusing on the R&D of open-weight models, with core focus on two major directions: code generation and AI intelligent agents.

Different from common closed-source models, open-weight models open their core parameters to developers. After obtaining authorization, enterprises can deploy them on their own servers independently, and perform fine-tuning and optimization combined with their business data.

This model can help enterprises master data autonomy, reduce dependence on cloud service providers, and better adapt to scenarios with high privacy and customization requirements.

This Monday, Reflection AI officially launched its first open-weight model Beam, which focuses on coding and intelligent agent tasks, directly competing with cost-effective domestic models such as DeepSeek and Kimi.

NVIDIA has previously invested 800 million U.S. dollars in Reflection AI and is one of its important shareholders. In addition, well-known institutions including HSG, DSTGlobal and Lightspeed Venture Partners have participated in the investment.

Financing data from March this year shows that Reflection AI's valuation has reached 25 billion U.S. dollars.

However, the enterprise is still in the early stage of commercialization, and its technical and market feasibility needs further verification.

NVIDIA's Calculation for Its AI Ecosystem

NVIDIA's negotiation for cooperation this time aims to improve the layout of its own AI ecosystem.

At present, AI programming and autonomous agents are the two most fiercely competitive tracks in the industry. Such businesses require continuous reasoning tasks, and have extremely high demand for GPU and data center computing power resources.

For NVIDIA, binding Reflection AI through equity investment, chip supply, computing power cooperation and other methods can convert the other party's business growth into long-term and stable demand for its own hardware products.

At the same time, relying on cooperation forms such as technology licensing and absorbing the core team, NVIDIA can deeply master the training, deployment and intelligent agent load technologies of open-weight models, feed back its own CUDA software, computing platforms and developer tools, and make up for the shortcomings of its model layer and application layer businesses.

In recent years, NVIDIA has continuously invested huge sums of money to expand the layout of the entire AI industry chain.

In addition to leading AI enterprises such as OpenAI and Anthropic, the company has also invested in emerging cloud service providers such as Nebius and CoreWeave, deeply binding core players in the upstream and downstream of the computing power industry.

Not limited to the AI track, NVIDIA's investment landscape covers multiple technology fields. It currently holds shares in optoelectronic enterprises such as Coherent and Lumentum, as well as companies such as Sound Hound, Arm Holdings and Marvell Technologies.

Multi-dimensional ecological investment has continuously expanded NVIDIA's industry influence and also provided a stable demand base for its chip business.

Jensen Huang, CEO of NVIDIA, also publicly stated in July this year that the United States needs to build a self-sustaining open-source AI ecosystem, which is crucial to creating opportunities for innovation and prosperity in the United States.

By increasing its stake in Reflection AI, NVIDIA has taken an important step to implement its open-source ecosystem layout and consolidate its leading position in the industry.

This article is from the WeChat Official Account "Gelonghui Finance Hotspot" (ID: glh_finance), written by Chashan, and published with authorization from 36Kr.