NVIDIA has invested $800 million in the startup Reflection as the company prepares to release its first open-weight model. The incoming model is designed to let businesses build proprietary AI systems at a low cost, matching the performance of select top-tier closed-source models while utilizing local proprietary data.
Reflection Prepares First Open-Weight Model Release
Reflection is positioning its upcoming software to compete directly with China’s leading open-weight models, according to an Axios report from October 4. While the startup’s initial release will trail the most advanced cutting-edge U.S. systems, industry sources indicate it possesses sufficient intelligence for enterprises to construct cost-effective, custom AI architectures.
Chief Executive Officer Misha Laskin, a former Google employee, acknowledged the development timeline in an interview with CNBC. “They’re a bit like rockets,” Laskin said regarding the pace of bringing advanced AI models to market. “Building a large rocket takes time.” Reflection has not yet disclosed a specific launch date or performance metrics for the software.
Enterprises Demand Secure Local AI Factories
The core business model centers on constructing what Reflection terms an “AI Factory.” This infrastructure allows government agencies and commercial enterprises to combine local proprietary data, Reflection models, and dedicated computing resources to build secure, on-premises systems.
Hedge funds and trading firms handling sensitive information have driven early interest in the concept. In March, Reflection signed a memorandum of understanding with South Korea’s Shinsegae Group to construct a 250-megawatt AI facility running Reflection models on NVIDIA hardware.
NVIDIA Chief Executive Officer Jensen Huang has consistently championed this open AI ecosystem. According to reports from The Wall Street Journal and The Information, Huang has strongly supported Reflection’s transition toward building open AI systems that let enterprises retain ownership and control over their data.
SpaceX and Nebius Secure Massive Compute Orders
To support model training and infrastructure scaling, Reflection has committed over $1 billion to secure massive computing power through multi-year agreements.
In June, Reflection signed an agreement with SpaceX (SPCX) to lease NVIDIA GB300 chips in the Colossus 2 data center. Starting July 1, 2026, and running through 2029, the contract requires monthly payments of $150 million after an initial ramp-up period, with provisions allowing either party to terminate with 90 days’ notice after the first three months.
Cloud computing firm Nebius (NBIS) agreed in July to sell over $1 billion in computing capacity to Reflection through 2029. Nebius stock declined on the day the agreement was made public.
Frequently Asked Questions About Reflection and NVIDIA
What is the core difference between Reflection’s model and ChatGPT?
Developers can freely download, modify, and run open-weight models like Reflection’s locally within licensing terms, whereas closed-source systems like ChatGPT and Claude are accessed exclusively via proprietary APIs or cloud accounts.
How much has NVIDIA invested in Reflection?
NVIDIA has committed $800 million to Reflection, establishing itself as one of the startup’s primary financial backers alongside its hardware supply agreements.
What infrastructure partnerships has Reflection established?
Reflection signed a multi-year hardware agreement with SpaceX starting in 2026 for NVIDIA GB300 chips and secured a $1 billion compute capacity contract with Nebius running through 2029.
Related reading