Reflection AI, a New York-based startup founded by former Google DeepMind researchers, is in talks to raise over US$1 billion to develop open-source large language models.

The company aims to compete with Meta, DeepSeek, and Mistral in the open-source AI space.

One source said Reflection AI has already secured most of its funding target.

The startup has spent the past year building a coding agent called Asimov, which analyzes corporate data to generate code.

It has generated a small amount of revenue from a preview with corporate clients. Investors include Lightspeed Venture Partners, Sequoia Capital, and CRV.

PitchBook previously valued the company at US$545 million.

The funding push comes amid rising demand from US companies for open-source AI models due to data security concerns with Chinese providers and the need for more customizable solutions.

.source-ref{font-size:0.85em;color:#666;display:block;margin-top:1em;}a.ask-tia-citation-link:hover{color:#11628d !important;background:#e9f6f5 !important;border-color:#11628d !important;text-decoration:none !important;}@media only screen and (min-width:768px){a.ask-tia-citation-link{font-size:11px !important;}}

🔗 Source: The Information

🧠 Food for thought

1️⃣ Cost efficiency gap drives urgent competitive response from US AI companies

DeepSeek’s breakthrough demonstrates how dramatically lower development costs can disrupt established players in AI.

The Chinese company trained its R1 model for less than $6 million while achieving performance competitive with models that cost hundreds of millions to develop1.

This represents a cost advantage compared to traditional approaches, with OpenAI expecting to spend over $7 billion on training models this year alone.

DeepSeek’s efficiency comes from its Mixture-of-Experts architecture, which activates only a subset of its 671 billion parameters during operation, and innovative techniques like FP8 mixed precision training that reduces energy and memory consumption2.

The rapid user adoption—10 million users within weeks of launch—shows that cost-efficient models can quickly gain market traction when offered for free3.

This explains why Reflection AI’s founders see an urgent opportunity to establish a US-based alternative, as the current cost structure puts American companies at a competitive disadvantage in the open-source market.

2️⃣ Data security concerns create forced market segmentation in AI

The inability of US companies to use Chinese AI models due to data security risks has created a protected market opportunity for domestic providers.

Multiple cybersecurity analyses highlight specific risks with AI systems, including adversarial attacks, data leakage, and the potential for models to inadvertently expose sensitive information through their outputs4.

This security barrier explains why Reflection AI can raise $1 billion despite entering a market with strong existing players—US enterprises need alternatives they can trust with confidential data.

The market segmentation is reinforced by regulatory frameworks and corporate policies that prohibit or restrict the use of foreign AI systems for sensitive applications5.

This creates a dual market structure where Chinese models can compete on cost and performance in general applications, while US companies maintain advantages in enterprise and security-sensitive use cases.

The $1 billion funding round reflects investor recognition that this security moat creates sustainable competitive advantages, even if the underlying technology may not initially match the cost efficiency of international competitors.

……