Valuation Soared From 500 Million to 25 Billion in One Year: Why Are Nvidia and SpaceX Jointly Betting on Reflection AI, Founded by AlphaGo’s Core Developer?

TradingKey - In the AI sector of 2026, the wildest valuation stories come not only from Anthropic and OpenAI, but also from Reflection AI, an open-source AI startup founded just two years ago.

In just one short year, its valuation soared from $545 million to $25 billion. Even more surprising is that this company has not yet officially released its flagship model, yet it secured a funding round led by Nvidia ( NVDA ), signed a major computing power contract with SpaceX ( SPCX ), and was even added to the partnership list of the US Department of Defense.

What kind of company is Reflection AI? On what grounds does it command simultaneous bets from Silicon Valley's top capital and industry giants?

What kind of company is Reflection AI?

Reflection AI was founded in March 2024 and is headquartered in New York. Both founders are veterans of Google DeepMind. Misha Laskin previously served as the reward modeling lead for DeepMind's Gemini project, while Ioannis Antonoglou is one of the co-creators of AlphaGo, AlphaZero, and MuZero.

In 2016, AlphaGo defeated Go world champion Lee Sedol, triggering the first wave of global public awareness of AI. Ten years later, Antonoglou chose to leave DeepMind to found Reflection AI. His core belief is that frontier models should have open weights rather than being black boxes hidden behind APIs.

The company currently has about 200 employees, and its core product is an autonomous coding agent named Asimov. Unlike code completion tools like GitHub Copilot, Asimov is designed to understand entire codebases, documentation, design specifications, and team communications, allowing it to autonomously complete the entire process of planning, writing, testing, and optimizing code. In the words of co-founder Laskin, the company has been answering one question from the very beginning: "How does AI become a true software engineer?"

Silicon Valley's Most Insane Valuation Leap

The fundraising pace of Reflection AI is almost unprecedented in Silicon Valley.

In March 2025, the company completed a $130 million Series A funding round at a valuation of approximately $545 million. Just seven months later, in October 2025, Nvidia led a $2 billion Series B round, causing the valuation to jump directly to $8 billion. The investor lineup was exceptionally prestigious, featuring Nvidia, Eric Schmidt, Citigroup ( C ), Lightspeed Venture Partners, Sequoia Capital, and 1789 Capital, where Donald Trump Jr. serves as a partner.

By March 2026, market rumors indicated that Reflection AI was in talks for a new $2.5 billion funding round, targeting a valuation of $25 billion.

From 545 million to 25 billion represents nearly a 46-fold increase in just one year, yet the company has not even released its flagship model.

Why is the market willing to grant such a high valuation to a company with "no product"? An analysis by 36Kr provides the answer: what the market truly values is not whether it has a blockbuster product today, but whether it has the potential to become an "infrastructure company" in the future AI world.

Why is Reflection AI dubbed the "American version of DeepSeek"?

The core strategic positioning of Reflection AI has been summarized by the market as the "American version of DeepSeek."

This positioning is not just a random label. DeepSeek's rise in 2025 was driven by a core logic: using open-source models to break the monopoly of closed-source models, allowing enterprises to truly own and control their own AI systems instead of relying on API calls forever. What Reflection AI is doing almost mirrors this path.

Over the past few years, the way most enterprises accessed AI was by calling OpenAI's APIs. This model was highly effective in the early stages, but as AI penetrated core corporate operations, several issues became increasingly prominent: data did not truly belong to the enterprises themselves, inference costs kept rising, and models could not be deeply optimized for specific scenarios. More importantly, institutions in sectors such as government, finance, and defense simply could not accept having their core AI systems run entirely on external platforms.

Reflection AI's strategy is a hybrid model of "open models + private training stacks." The model weights are open, allowing enterprises to deploy, own, and customize them; however, the training systems, data frameworks, and underlying infrastructure remain private, preserving the scalability of an open ecosystem while maintaining core technological barriers.

CTO Antonoglou explained the logic behind this choice in an interview: open models accelerate research and gain broader external validation, and security performance is enhanced through ecosystem stress testing. For enterprises and governments, "sovereignty" is becoming increasingly important, and they need to be able to fully control their own AI stacks.

This strategy has already been validated. In May 2026, the U.S. Department of Defense announced agreements with seven AI companies to deploy advanced technologies into the Department's classified networks, with Reflection AI listed alongside SpaceX, OpenAI, Google ( GOOGL ), Nvidia, Microsoft ( MSFT ), and Amazon ( AMZN ). In addition, Reflection has entered into a multi-billion-dollar partnership with South Korea's Shinsegae Group to customize Korean-language large language models for them.

A $6.3 Billion Computing Power Deal: Why Is SpaceX Betting on Reflection AI?

In June 2026, Reflection AI signed a computing power cooperation agreement with SpaceX.

Under the agreement, from July 1, 2026, through 2029, Reflection will pay SpaceX $150 million monthly in exchange for the right to use Nvidia's GB300 AI chips at SpaceX's Colossus 2 data center near Memphis, Tennessee. If the agreement is executed until expiration, the total payment will reach approximately $6.3 billion. After the first three months, either party can terminate the contract with 90 days' advance notice.

This deal holds strategic significance for both parties.

For Reflection, computing power is the most scarce resource for training frontier large models. The GB300 is one of Nvidia's most advanced AI chips, and direct access to it means training progress will no longer be bottlenecked by computing power. A company spokesperson stated that this agreement will provide additional computing resources to accelerate the goal of achieving "American Open Intelligence."

For SpaceX, this is yet another major deal opening up its Colossus infrastructure to external parties. Previously, SpaceX had reached computing power partnerships with Anthropic, Google, and Cursor. By opening Colossus to external clients, SpaceX is positioning itself as a cloud service provider and an AI infrastructure enterprise.

More importantly, this transaction comes after Anthropic was forced to shut down Fable 5 and Mythos 5 due to export controls. This event has prompted global enterprises and governments to re-evaluate the risks of relying on closed AI systems—if key business operations are completely tied to a single closed-source model provider, the business could grind to a halt immediately upon changes in policy or security reviews. Against this backdrop, Reflection's open-source model route has gained additional strategic value.

Open-Source Models Reach an Inflection Point: What is Reflection AI’s 25-Billion Valuation Betting On?

The rise of Reflection AI is, in essence, an early bet on the shifting underlying logic of the AI industry.

Over the past two years, closed-source models have proven the immense value of "frontier models." However, the upcoming questions are: Who owns the model? Who controls the cost? Who owns the data? And who can truly run AI on their own infrastructure?

As more enterprises and governments realize that AI is not just a SaaS tool but the core production system of the future, "open-source models" cease to be merely a technological ideal and become a hard requirement on both commercial and geopolitical levels. Reflection AI happens to stand at this inflection point. The technical endorsement from AlphaGo's core developer, computing power and capital backing from Nvidia, computing infrastructure supply from SpaceX, and the trust endorsement from the U.S. Department of Defense together form a complete closed loop.

The $25 billion valuation is a bet that open-source models will become the core infrastructure for the second half of the AI race. The stakes are high, but the cards in Reflection AI's hand are indeed not bad.