Reflection AI, a one-year-old artificial intelligence startup founded by former DeepMind researchers Misha Laskin and Ioannis Antonoglou, has raised $2 billion in fresh funding at an $8 billion valuation — a staggering 15-fold jump from its valuation just seven months ago.

The company is positioning itself as both an open-source alternative to closed labs like OpenAI and Anthropic, and a Western counterweight to China’s fast-advancing AI ecosystem led by DeepSeek, Qwen, and Kimi.

According to CNBC, the funding marks one of the largest-ever rounds for an AI startup, signaling strong investor belief that Reflection AI could become a central player in shaping the global balance of AI power.

Launched in March 2024, Reflection AI is the brainchild of two DeepMind veterans with a track record of high-stakes AI breakthroughs. Laskin, who led reward modeling for DeepMind’s Gemini project, and Antonoglou, who co-created AlphaGo, the system that famously defeated the world champion in Go, are betting that elite AI researchers can now build frontier-scale models outside of Big Tech.

The company, based in the U.S., began by developing autonomous coding agents before broadening its focus to general-purpose, agentic AI systems. According to Laskin, Reflection AI has already “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 (MoEs) models at frontier scale.”

MoE architectures are critical to today’s largest models — used by OpenAI’s GPT-4, Google’s Gemini, and China’s DeepSeek — and represent the foundation for Reflection AI’s upcoming release. The startup’s first frontier language model, trained on tens of trillions of tokens, is expected in early 2026.

A Mission with Geopolitical Undertones

As China accelerates state-backed model development and restricts U.S. companies’ access to its AI infrastructure, American startups like Reflection AI are rallying private capital to build what Laskin calls “open intelligence for the West.”

“DeepSeek and Qwen are our wake-up call,” Laskin said in an interview. “If we don’t do anything about it, then effectively, the global standard of intelligence will be built by someone else. It won’t be built by America.”

He argued that the U.S. and its allies risk losing control of critical AI standards — both technologically and politically — as enterprises and governments shy away from Chinese models due to security and legal concerns.

Reflection AI has drawn attention for its hybrid approach to openness. Unlike fully open research collectives such as Hugging Face, the company plans to release its model weights publicly — enabling anyone to use and modify its systems — while keeping datasets and training pipelines proprietary.

“The most impactful thing is the model weights,” Laskin explained. “Anyone can use and build on them. But the full infrastructure stack — that’s a domain only a few companies can leverage.”

This mirrors strategies used by Meta’s Llama and Mistral, which have embraced partial openness as a competitive advantage while retaining control over proprietary infrastructure and commercial licensing.

Building the Business of Open Intelligence

Reflection AI’s 60-person team — made up largely of ex-Google, OpenAI, and DeepMind researchers — is now working on scaling its AI training stack and deploying compute clusters to prepare for model training.

Its commercial model is equally ambitious: while researchers will access its models freely, the company plans to earn revenue from enterprises and governments using Reflection AI for “sovereign AI” — models controlled by national institutions rather than foreign corporations.

“Large enterprises by default want open models,” Laskin said. “They want ownership, control, and the ability to optimize costs. You’re paying some ungodly amount of money for AI — you want to be able to run it on your own infrastructure.”

This positioning places Reflection AI at the intersection of open innovation, economic independence, and digital sovereignty, a theme increasingly resonant among Western policymakers.

Backed by AI Heavyweights

The funding round drew participation from some of the biggest names in venture and AI investment, including Nvidia, DST, Sequoia Capital, B Capital, GIC, Lightspeed, Disruptive, 1789 Capital, CRV, and Citi. Tech leaders such as Eric Schmidt, Zoom CEO Eric Yuan, and David Sacks, who also serves as the White House AI and Crypto Czar, joined the round.

Sacks hailed Reflection AI’s mission on X, writing: “It’s great to see more American open-source AI models. A meaningful segment of the global market will prefer the cost, customizability, and control that open source offers. We want the U.S. to win this category too.”

Hugging Face co-founder Clem Delangue also welcomed the move, calling it “great news for American open-source AI,” while cautioning that the challenge ahead is to demonstrate consistent model-sharing velocity comparable to leading open labs in China and Europe.

Open Source as a Strategic Asset

Reflection AI’s meteoric rise reflects a growing belief that open-source AI could become the next strategic battleground — not only in the tech industry but in national innovation policy. While China invests heavily in state-supported model development and closes its platforms to U.S. firms, the U.S. private sector is responding by empowering independent AI ventures like Reflection AI to lead in model transparency and accessibility.

In many ways, Reflection AI’s rapid valuation surge, from $545 million to $8 billion in just seven months, captures a broader shift in investor sentiment: the realization that frontier AI no longer belongs solely to Big Tech.

Microsoft has unveiled its first large-scale Nvidia-powered artificial intelligence “factory”, a massive computing cluster built to run OpenAI workloads and other frontier AI models across its Azure global cloud network.

The factory points to the company’s ambition to dominate the infrastructure layer of the AI revolution as OpenAI’s user base and service offerings continue to expand at an unprecedented pace.

Microsoft CEO Satya Nadella shared a video of the new system on Thursday on X, calling it the “first of many” such AI factories the company plans to deploy worldwide. The system — a cluster of more than 4,600 Nvidia GB300 rack computers — is powered by Nvidia’s new Blackwell Ultra GPUs and linked through InfiniBand, Nvidia’s ultra-fast networking technology. Each of these massive systems can run large-scale AI workloads, including next-generation models with “hundreds of trillions of parameters,” according to Microsoft.

The company said the new AI factory represents a “new class of supercomputing infrastructure” designed to handle both AI training and inference at a global scale. Nadella described the deployment as part of Microsoft’s long-term plan to build out “hundreds” of such systems across Azure’s 300 data centers in 34 countries.

Nadella explained that it is the first of many Nvidia AI factories we’re deploying across Microsoft Azure data centers worldwide to power frontier AI models. He added that Microsoft’s infrastructure “will meet the demands of frontier AI today — and tomorrow.”

The launch comes at a time when infrastructure has become the backbone of OpenAI’s expansion, as the company diversifies its offerings beyond ChatGPT and scales its computing capacity to meet soaring demand. OpenAI’s flagship product, ChatGPT, has now surpassed 800 million active users per week, a milestone that cements its position as the world’s most widely used AI platform.

The surge in user activity has placed enormous strain on computing resources, prompting OpenAI to invest heavily in global data centers and specialized hardware to maintain uptime and support more sophisticated models. The company’s recent $1 trillion data center commitments — spread across partnerships with Nvidia, AMD, and major cloud providers — reflect the scale of its ambition to operate as a global AI utility.

Microsoft’s new AI factories are designed to serve as dedicated infrastructure for OpenAI, as well as for enterprise clients adopting AI tools through Azure. Microsoft emphasized that the systems are optimized not only for ChatGPT and GPT-5-class models but also for multimodal AI systems like OpenAI’s video model Sora and code-generation platforms that demand extreme processing power.

The collaboration between Microsoft and Nvidia marks another milestone in a partnership that has reshaped the AI industry. Nvidia’s dominance in both GPU chips and data-center networking, strengthened by its 2019 acquisition of Mellanox Technologies for $6.9 billion, has made it the undisputed supplier of AI infrastructure. Microsoft’s alignment with Nvidia ensures early access to the Blackwell Ultra GPU, which has become the industry’s most sought-after AI processor.

The timing of Microsoft’s announcement follows OpenAI’s own major data center agreements with Nvidia and AMD, which some analysts saw as a sign that OpenAI was expanding beyond Microsoft’s Azure ecosystem. Microsoft is signaling that it remains the primary infrastructure partner with the capacity and geographic reach to support OpenAI’s exponential growth by unveiling its AI factory now.

Microsoft said its AI systems are built with “sustainability and scalability” in mind, integrating renewable energy sources and advanced cooling systems to offset the enormous energy consumption associated with training large AI models. Each AI factory, according to internal estimates, consumes as much electricity as a small city.

Beyond OpenAI, Microsoft intends to use the same infrastructure to power enterprise AI products like Copilot, its suite of AI assistants embedded in Windows, Office, and GitHub. The company said the infrastructure will also support “custom frontier models” for corporate clients developing proprietary AI systems.

Microsoft CTO Kevin Scott is expected to provide more details about the company’s AI infrastructure roadmap at TechCrunch Disrupt in San Francisco later this month (October 27–29). Scott will outline how Azure’s data centers are evolving to support the next generation of AI workloads and to remain competitive with Amazon Web Services and Google Cloud, both of which are also scaling their own AI supercomputing capabilities.

OpenAI’s growth from a research lab to a global AI platform has transformed it into one of the world’s largest consumers of computing power. However, its continued success depends on the availability of scalable infrastructure capable of handling billions of user queries, video generations, and code executions daily.