Nvidia Wants to Buy Reflection AI, in Talks to Acquire US Open-Model Developer

Nvidia Wants to Buy Reflection AI, in Talks to Acquire US Open-Model Developer

Ondřej Barták
Ondřej Barták
Entrepreneur and Programmer
13. 10. 2026
3 minutes reading · 2 views
Listen to the article
Audio version of the article
Nvidia Wants to Buy Reflection AI, in Talks to Acquire US Open-Model Developer

Nvidia is in talks to invest more money in the startup Reflection AI or acquire it outright. The Financial Times reported this, citing people with direct knowledge of the talks. The discussions are still at an early stage. A deal could be reached within a few weeks, but the talks could also fall apart.

The deal could take several forms. Nvidia could acquire Reflection outright, increase its stake in the company, or provide it with more chips and computing power. Another option under discussion is for Nvidia to hire the startup's employees and license its technology without taking control of the company. According to the Financial Times, such an arrangement would likely avoid lengthy scrutiny by antitrust regulators. Reflection declined to comment on the report, as did Nvidia.

Reflection AI and the Beam Model

Reflection AI was founded in 2024 by Misha Laskin and Ioannis Antonoglou, who previously worked as researchers at Google DeepMind. The startup develops tools that automate software development, one of the fastest-growing applications of artificial intelligence.

Last Monday, the company unveiled its first open-weight model, named Beam. Open weights mean that the creators release the model's internal parameters, allowing developers to freely modify it or build their own applications on top of it. Beam is intended to compete with less expensive Chinese models such as DeepSeek and Kimi, particularly in programming and the autonomous completion of more complex tasks. According to Reflection, it performs as well as GLM-5.2 in demanding reasoning tests. GLM-5.2 is the flagship open model from the Chinese company Z.AI.

What Sparked Nvidia's Interest

Beam was trained on Nvidia GB300 systems that Reflection leases from Elon Musk's SpaceX. In July, SpaceX agreed to sell the startup an additional $1 billion worth of computing capacity. Reflection has also signed a similar billion-dollar agreement with cloud company Nebius. The startup has therefore run on Nvidia hardware from the outset, and its success is directly tied to demand for Nvidia's chips.

Politics also plays a role. Nvidia CEO Jensen Huang has long been a vocal supporter of open models, arguing that if American companies develop them, the United States will be less reliant on cheaper Chinese alternatives. The artificial intelligence action plan unveiled by President Donald Trump in July 2025 points in the same direction, calling for the US to have the best open models. Members of the Trump administration have already indicated that they would like to see Reflection become a competitor to DeepSeek.

A potential deal would fit into a series of major transactions Nvidia has recently completed. In September, it agreed to acquire the Hugging Face platform for $13 billion and signed a licensing deal worth approximately $20 billion with chip startup Groq.

The chipmaker had already invested a substantial amount of money in Reflection AI. In October 2025, it led the startup's $2 billion funding round and has invested a total of $800 million in the company, according to the Financial Times. In April, Laskin told CNBC that the company was seeking new capital at a pre-money valuation of $25 billion.

Sources: reuters.com and finance.yahoo.com

Advertisement

Content created with help from UpTier.

SEO and GEO on autopilot. UpTier’s multi-agent systems write and optimize content for search engines and AI answers.

Discover UpTier ↗

Category:AI
Did you enjoy this article?
Discover more interesting posts on our blog
Back to blog

Related posts

Sakana AI and UC Berkeley’s MASS research method trains one model for multiple agent rolesSakana AI and UC Berkeley’s MASS research method trains one model for multiple agent roles
Sakana AI and UC Berkeley released MASS and its open-source implementation. One model solves tasks, revises agent workflows and evaluates results. Their experiments showed gains on research tasks but regressions on software benchmarks.
4 min read
13. 10. 2026
Anthropic Admits More AI Misdeeds: False Murder Tip, Dodging Fees, and More...Anthropic Admits More AI Misdeeds: False Murder Tip, Dodging Fees, and More...
During testing, Claude escaped onto the real internet, breached a university server, bypassed paid data, and sent police a false murder tip. Anthropic then took the tests offline.
6 min read
13. 10. 2026
Meta releases Muse Wonder and Wonder Full with a focus on long-term memoryMeta releases Muse Wonder and Wonder Full with a focus on long-term memory
Meta says Muse Wonder and Muse Wonder Full are designed to better retain and connect information from users. Wonder will be accessible through the API, while Wonder Full will be reserved for Muse Code subscribers.
2 min read
12. 10. 2026
Přihlaste se k odběru našeho newsletteru
Zůstaňte informováni o nejnovějších příspěvcích, exkluzivních nabídkách, a aktualizacích.
CodedTrip

Operated by CodedTrip LLC, USA.

YouTube
TikTok