--- title: "NVIDIA bets on Reflection to layout open-source artificial intelligence, and now this startup faces increasing pressure to catch up" type: "News" locale: "en" url: "https://longbridge.com/en/news/294338062.md" description: "NVIDIA previously invested $800 million to support Reflection AI's transformation into open-source AI, helping it attract significant capital. However, nearly a year later, due to the lack of a self-developed large model, Reflection has been significantly outpaced by intense competition from DeepSeek, Moonshot AI, and local competitors, facing severe pressure to catch up" datetime: "2026-07-30T09:46:22.000Z" locales: - [zh-CN](https://longbridge.com/zh-CN/news/294338062.md) - [en](https://longbridge.com/en/news/294338062.md) - [zh-HK](https://longbridge.com/zh-HK/news/294338062.md) --- # NVIDIA bets on Reflection to layout open-source artificial intelligence, and now this startup faces increasing pressure to catch up Yannis Antonoglou (left), co-founder of Reflection AI, and Misha Laskin. According to informed sources, in August last year, a co-founder of Reflection AI flew to California to meet with NVIDIA CEO Jensen Huang. At that time, the startup was originally developing AI programming tools, but Huang suggested they change direction and explore a more novel path at the time: creating open-source artificial intelligence models that developers could download and modify independently. With Huang's strong support, along with NVIDIA's initial investment of $800 million, Reflection CEO Misha Laskin and co-founder and CTO Yannis Antonoglou positioned the company as a core force in the U.S. open-source AI sector to compete with Chinese firms. This development vision attracted billions of dollars from various investors, including JP Morgan, Sequoia Capital, and 1789 Capital (the venture capital firm where Donald Trump Jr. serves as a partner). The U.S. Department of Defense also included it in the military AI supplier list alongside the seven major tech giants in the U.S. However, nearly a year has passed since that meeting, and with the global open-source AI boom fully erupting, Reflection has yet to officially launch its self-developed large model, falling further behind its competitors. Since the launch of the R1 model by DeepSeek in January 2025, which shook the U.S. tech industry, several increasingly powerful Chinese open-source models have gained a large number of enterprise clients. This month, Moonshot AI released Kimi K3, which performs on par with the top closed-source models from Anthropic and OpenAI. This has prompted intense discussions in U.S. political circles about the regulatory methods and implementation details for open-source AI. Domestic competition has also become fierce, with well-funded startups Thinking Machines Lab and Poolside consecutively launching open-source models, and NVIDIA itself has also entered the fray by releasing its self-developed open-source large model. According to sources close to the company, the New York-based Reflection plans to launch its first large model this year, with a larger parameter version scheduled to go live in early 2027, followed by the release of lightweight small models. The source stated that Reflection's first-generation product performance will be on par with the top Western models released at the same time, but still lag behind the leading Chinese open-source models; the company hopes to achieve a lead with subsequent iterative versions. Anastasios Angelopoulos, co-founder and CEO of AI evaluation startup Arena, stated: "All open-source vendors are in a fiercely competitive track." Arena's evaluations show that after the release of Kimi K3, it topped the leaderboard in UI interface generation tasks. He believes that Reflection still has a chance to enter the market as long as it delivers a quality model, but the earlier it releases its open-source product, the easier it will be to establish a foothold in the crowded market Open-source large models can hedge against the monopoly of OpenAI and Anthropic's closed-source models. The pricing of leading closed-source models continues to rise, prompting more companies to turn to open-source solutions — not only are they more cost-effective and predictable in expenses, but they also allow for deep customization. Overseas companies and multiple governments are increasingly adopting open-source models. Reflection bets that many institutions will prioritize American domestic models over the currently dominant Chinese open-source solutions. In an interview with TITV, a program under Information Network, last October, Laskin candidly stated: "DeepSeek R1 has sounded a huge alarm for the entire United States. We need to create top-tier open-source intelligent models that can compete with the strongest products globally, allowing the U.S. and the world to build various intelligent applications based on this model." Reflection's first product may find it difficult to achieve this grand goal, but the entry qualifications for the field remain. The company is attracting a large number of top AI researchers, and its valuation reached $25 billion in this year's new round of financing, with investors recognizing its model training progress. Even if the initial model cannot reach industry-leading status, Reflection can still provide customized services to attract clients. **Grand Layout Behind** 36-year-old Laskin was born in Russia, moved to Israel in his childhood, and later settled in the United States. He graduated from Yale University and obtained a Ph.D. in physics from the University of Chicago. He joined Google DeepMind in 2022, participating in the early development of Gemini; he left in 2024 to found Reflection. 38-year-old co-founder Antonoglou was born in Greece and earned a Ph.D. in artificial intelligence from University College London. He has been deeply involved in DeepMind for over a decade, leading the development of the milestone AI — AlphaGo, which defeated Go champion Lee Sedol in 2016. In the early stages of their startup, the two focused on AI programming tools, during which similar startups like Cursor were gaining popularity. However, after meeting Jensen Huang, Reflection almost overnight abandoned its original path. Insiders revealed: "The company completely transformed from 'dual-track development of applications and models' to a pure AI laboratory, marking a fundamental strategic shift. The two founders completely set aside their initial plans and steered the company entirely according to Jensen Huang's direction." Reflection completed a $2 billion financing round in early October last year, with 40% coming from NVIDIA; this year, it completed another round of fundraising totaling $2.5 billion. Before and after the last round of financing, Laskin gathered all employees at a hotel in Hampton to call for everyone to work towards the new goal of becoming "the top open-source AI service provider in the U.S." An attending employee stated that everyone was greatly encouraged and generally believed that with NVIDIA's computational power and financial support, Reflection has the capability to grow into a heavyweight player in the industry For Jensen Huang, supporting Reflection can build the core foundation of the American open-source AI industry, with related models deeply adapted to NVIDIA chips and hardware devices. NVIDIA's hardware sales to Chinese AI companies are restricted, while its traditional customers (major AI giants in the U.S.) are developing their own chips, creating competitive pressure. Supporting the local open-source ecosystem has become a breakthrough strategy. **Building an Open-Source AI Commercialization System** According to sources familiar with the details of NVIDIA's meetings, after transforming into a professional AI laboratory, Reflection set a recruitment goal to ultimately form a research team of 150 people. Currently, the company has over 230 employees, with more than 100 researchers. Although no formal models have been released yet, the company has already begun recruiting a sales team to connect with governments around the world and dispatching frontline engineers to customize models for large enterprises. Reflection has reached a cooperation agreement with Dell, where Dell will promote Reflection models to its enterprise customers using its own software and hardware, replicating the cooperation model between Dell and the French open-source AI company Mistral AI. Reflection continues to expand the computing power needed for model training, having recently finalized agreements to lease NVIDIA AI servers from SpaceX and Nebius, with total leasing costs potentially reaching billions of dollars over the next few years. **High Influence in U.S. Politics** Reflection is making significant inroads into the Washington policy circle, hiring Rachel Appleton, the first policy lobbying head of Anthropic. In terms of corporate size, its influence in the U.S. federal AI planning far exceeds that of conventional peers. In May of this year, eight companies signed contracts with the U.S. Department of Defense, allowing the military to use their AI technology on classified internal networks, with Reflection being one of them; the other seven are SpaceX, OpenAI, Google, NVIDIA, Microsoft, Amazon Web Services, and Oracle. The U.S. Department of Energy has also included Reflection's technology in the "Genesis Project," relying on AI to accelerate scientific breakthroughs. According to reports from Information Network, Reflection is actively cooperating with White House industry discussions to explore how the federal review rules for cutting-edge AI in the June executive order will impact open-source models. **Two Cutting-Edge Racing Tracks** Reflection is diligently refining its models, while competitors are continuously pushing performance limits. Before the launch of Kimi K3 this month, the Chinese open-source model Z.ai GLM-5.2 had already received widespread acclaim from developers. The UK government's cybersecurity tests show that GLM-5.2 performs on par with mainstream large models released in the U.S. just four months earlier. Thinking Machines Lab, founded by former OpenAI Chief Technology Officer Mira Murati in February 2025, released its first open-source model Inkling this month, with a parameter count close to 1 trillion, less than half the parameter scale of Kimi K3. In the intelligent scoring system of the professional evaluation agency Artificial Analysis, Inkling scored 41 points, while Kimi scored 57 points, and Anthropic's latest model scored 61 points NVIDIA updated its own open-source product line in June, launching Nemotron 3 Ultra, which has parameters about half the size of Inkling, with a score of 38, lower than Inkling and leading Chinese open-source models. To achieve the goal of matching top Western open-source models, the performance of Reflection's first product needs to at least catch up with Inkling. Sources reveal that the Reflection model has parameter counts in the hundreds of billions, positioned between the open-source model Laguna S 2.1 from programming AI company Poolside and Inkling, focusing on robust coding capabilities, specifically optimized for AI agents. In an interview in April this year, Laskin stated, "The difficulty of developing large models is extremely high, comparable to building rockets. Small rockets are not difficult to make, but it's hard to form a competitive edge; creating a large rocket that can lead the industry will inevitably require a long cycle." ### Related Stocks - [NVDA.US](https://longbridge.com/en/quote/NVDA.US.md) - [NVDD.US](https://longbridge.com/en/quote/NVDD.US.md) - [NVD.US](https://longbridge.com/en/quote/NVD.US.md) - [NVDY.US](https://longbridge.com/en/quote/NVDY.US.md) - [NVDU.US](https://longbridge.com/en/quote/NVDU.US.md) - [NVDL.US](https://longbridge.com/en/quote/NVDL.US.md) - [NVDQ.US](https://longbridge.com/en/quote/NVDQ.US.md) - [NVDX.US](https://longbridge.com/en/quote/NVDX.US.md) - [NVDS.US](https://longbridge.com/en/quote/NVDS.US.md) - [NVDB.US](https://longbridge.com/en/quote/NVDB.US.md) - [NVDG.US](https://longbridge.com/en/quote/NVDG.US.md) - [NVDO.US](https://longbridge.com/en/quote/NVDO.US.md) - [NVDW.US](https://longbridge.com/en/quote/NVDW.US.md) - [NVYY.US](https://longbridge.com/en/quote/NVYY.US.md) - [NYYY.US](https://longbridge.com/en/quote/NYYY.US.md) - [DIPS.US](https://longbridge.com/en/quote/DIPS.US.md) - [MAGX.US](https://longbridge.com/en/quote/MAGX.US.md) - [SMH.US](https://longbridge.com/en/quote/SMH.US.md) - [SOXX.US](https://longbridge.com/en/quote/SOXX.US.md) - [SOXL.US](https://longbridge.com/en/quote/SOXL.US.md) - [SOXQ.US](https://longbridge.com/en/quote/SOXQ.US.md) - [XSD.US](https://longbridge.com/en/quote/XSD.US.md) - [PSI.US](https://longbridge.com/en/quote/PSI.US.md) - [FTXL.US](https://longbridge.com/en/quote/FTXL.US.md) - [07388.HK](https://longbridge.com/en/quote/07388.HK.md) - [07788.HK](https://longbridge.com/en/quote/07788.HK.md) - [09388.HK](https://longbridge.com/en/quote/09388.HK.md) - [DPSK.NA](https://longbridge.com/en/quote/DPSK.NA.md) - [JPM.US](https://longbridge.com/en/quote/JPM.US.md) - [OpenAI.NA](https://longbridge.com/en/quote/OpenAI.NA.md) - [GOOGL.US](https://longbridge.com/en/quote/GOOGL.US.md) - [GOOG.US](https://longbridge.com/en/quote/GOOG.US.md) - [DELL.US](https://longbridge.com/en/quote/DELL.US.md) - [NBIS.US](https://longbridge.com/en/quote/NBIS.US.md) - [MSFT.US](https://longbridge.com/en/quote/MSFT.US.md) - [AMZN.US](https://longbridge.com/en/quote/AMZN.US.md) - [ORCL.US](https://longbridge.com/en/quote/ORCL.US.md) - [NVD.DE](https://longbridge.com/en/quote/NVD.DE.md) - [JPM-M.US](https://longbridge.com/en/quote/JPM-M.US.md) - [JPM-C.US](https://longbridge.com/en/quote/JPM-C.US.md) - [JPM-D.US](https://longbridge.com/en/quote/JPM-D.US.md) - [JPM-L.US](https://longbridge.com/en/quote/JPM-L.US.md) - [8634.JP](https://longbridge.com/en/quote/8634.JP.md) - [JPM-K.US](https://longbridge.com/en/quote/JPM-K.US.md) - [JPM-J.US](https://longbridge.com/en/quote/JPM-J.US.md) - [ORCL-D.US](https://longbridge.com/en/quote/ORCL-D.US.md) ## Related News & Research - [Mark Cuban and Michal Burry warn the AI boom is dangerously reliant on Nvidia](https://longbridge.com/en/news/294232031.md) - [DeepSeek’s $71 Billion Boom Collides With an Awkward Nvidia Question](https://longbridge.com/en/news/293972328.md) - [Nvidia's Potential $250 Billion OpenAI Financing Deal Is Reviving a 1990s Tech Bubble Habit, Analyst Warns: 'Usually Don't End Well'](https://longbridge.com/en/news/294077412.md) - [Tech coalition warns against sweeping artificial intelligence limits](https://longbridge.com/en/news/293799635.md) - [Nvidia forms industry alliance for open AI security after Hugging Face hack](https://longbridge.com/en/news/293916175.md)