Reflection AI, a startup founded in March 2024 by two former Google DeepMind researchers — Misha Laskin and Ioannis Antonoglou — has raised a stunning $2 billion at an $8 billion valuation, just seven months after being valued at $545 million. The company aims to become America’s open frontier AI lab, positioning itself as both an open-source alternative to closed Western AI giants like OpenAI and Anthropic, and a U.S. counterpart to China’s fast-rising frontier labs such as DeepSeek and Qwen.
From DeepMind to the Open Frontier
Laskin, who led reward modeling for DeepMind’s Gemini project, and Antonoglou, one of the creators of AlphaGo, founded Reflection AI on the belief that frontier-level intelligence systems can be built outside big tech — if the right people and infrastructure are in place.
The startup has recruited top AI researchers from DeepMind and OpenAI and built an advanced training stack that it plans to make partially open. According to Laskin, Reflection AI now has around 60 employees, most of whom are AI engineers focused on infrastructure, data, and algorithmic innovation.
The company says it has secured a large-scale compute cluster and will release its first frontier-level language model in 2026, trained on tens of trillions of tokens — a scope previously achievable only by the largest labs.
“Building what was once thought impossible”
In a statement on X, Reflection AI wrote:
“We built something once thought possible only inside the world’s top labs: a large-scale LLM and reinforcement learning platform capable of training massive Mixture-of-Experts models at frontier scale.”
The startup first applied this system to autonomous coding agents, demonstrating its potential in reasoning and decision-making tasks. With its new funding, the company now plans to bring these methods to general-purpose AI reasoning.
This move puts Reflection AI directly in competition with China’s DeepSeek — which recently achieved global recognition for scaling open Mixture-of-Experts architectures — and signals a strategic response from the U.S. AI ecosystem.
As Laskin put it:
“DeepSeek and Qwen are our wake-up call. If we don’t act, the global standard of intelligence will be defined by someone else — and it won’t be built by America.”
Open access — with commercial guardrails
Reflection AI’s philosophy of “openness” centers on accessibility, not full transparency. Similar to Meta’s Llama and Mistral, the company will openly release model weights — the critical parameters of AI models — while keeping datasets and full training pipelines proprietary.
“The model weights are what matter most,” Laskin said. “They let anyone experiment and build, while the infrastructure itself will remain in the hands of only a few capable companies.”
This approach also shapes the company’s business model:
- Researchers will be able to use the models for free.
- Enterprises will pay to deploy customized versions on their own infrastructure.
- Governments will use them to build “sovereign AI” systems — national-level models developed and controlled independently.
“Once you’re a large enterprise,” Laskin explained, “you want an open model — one you own, run, and optimize for your specific workloads. That’s the market we’re serving.”
Backed by tech’s power players
Reflection AI’s massive $2 billion round was backed by an impressive lineup of investors, including Nvidia, DST, Disruptive, B Capital, Lightspeed, GIC, Eric Schmidt, Eric Yuan, Sequoia, CRV, and Citi.
Tech leaders have publicly applauded the startup’s vision.
- David Sacks, the U.S. White House AI & Crypto Czar, wrote on X: “It’s great to see more American open-source AI models. The global market needs affordable, customizable, and controllable alternatives — and we want the U.S. to lead this category too.”
- Clem Delangue, CEO of Hugging Face, told TechCrunch: “This is great news for American open-source AI. The next challenge will be maintaining a high velocity of open model and dataset sharing.”
Reflection AI’s rise underscores a geopolitical shift in frontier AI development. As Chinese models like DeepSeek, Qwen, and Kimi push forward with massive open architectures, American AI builders are responding with renewed urgency — determined to keep the frontier of intelligence open and democratic.
In Laskin’s words:
“You can either live at a competitive disadvantage — or rise to the occasion.”
Reflection AI’s $2 billion funding round marks one of the largest investments in an AI startup in history. With a small but elite team, an open-access philosophy, and heavy-weight backers, the company aims to create America’s open-source answer to frontier AI, ensuring that the next wave of intelligence — and innovation — remains built in the West.
