The startup AI, which received an investment of $800 million from NVIDIA, Reflection, is preparing to release its first open-weight model.

According to a report by Axios on October 4th, the first model of Reflection is expected to lag behind the most advanced models in the United States in its initial stages, but it has the potential to compete directly with China's leading open-weight models.

According to reports, sources stated that this model possesses sufficiently powerful intelligent capabilities to help companies build proprietary AI systems at low costs. In some cases, combining a slightly less capable open-source model with the company's own high-quality data can result in outcomes that are even comparable to those of the most expensive, cutting-edge AI systems.

Open-weight model: China takes the lead, while the West is striving to catch up

The core difference between the open-weight model and closed-source models such as ChatGPT and Claude is that developers can freely download and modify the model itself within the scope of the licensing terms. This feature has led to its rapid adoption among individual developers and small and medium-sized enterprises.

However, reports indicate that currently, the most powerful open-weight models in the world almost all come from China. On platforms that offer a large number of such models, the combined market share of open models sometimes accounts for the majority of usage. Nevertheless, in the larger-scale corporate market, the proportion of open-weight models is still relatively limited. Several executives and analysts from corporate companies pointed out that companies typically pay to use services through corporate accounts, and the penetration rate of open models in this market is still low.

The entry of Reflection, as well as the open-weight models expected to be released by other Western manufacturers this month, will bring new variables to this landscape. According to a report by Axios, sources have revealed that after the release of the Reflection model, other Western manufacturers will also follow suit and launch open-weight models this month.

Reflection CEO Misha Laskin previously worked at Google. In an interview with CNBC, he admitted that it still takes time to bring the company's AI model to the most advanced level, comparing it to a "rocket": "It's a bit like building a rocket. It takes time to construct a large rocket."

On the eve of the model's release, Reflection has been in communication with relevant parties in Washington and other regions, introducing the upcoming model as well as the operational logic of the “AI factory” to all stakeholders. Reflection has not yet announced the specific release date or performance data of the model.

Analysis suggests that the entry of Reflection marks the beginning of a new front in the industry competition between AI. In the past, the market's focus was on the rivalry among OpenAI, Anthropic, and Google regarding their proprietary closed-source models; today, another question is becoming increasingly important: who can create the most powerful open-weight model?

"AI Factory": The Core Ambition of Reflection

The real goal of Reflection is far more than just releasing a model. The company's core vision is to create what they call the “AI Factory” ( AI Factory ) – allowing enterprises or government agencies to combine their proprietary data, Reflection models with their own computing power to build highly customized, lower-cost localized AI systems.

Hedge funds, trading companies, and other institutions that handle highly confidential data have shown strong interest in this model. Reflection has begun to put this concept into practice: in March of this year, the company signed a memorandum of understanding with South Korea's Shinsegae Group ( Shinsegae Group ), planning to build a 250-megawatt AI factory in South Korea to operate the Reflection model and equip it with NVIDIA chips.

It is worth noting that the concept of “AI Factory” is not an original idea of Reflection; rather, it is precisely the core vision that NVIDIA CEO Jensen Huang has repeatedly emphasized over the years. Jensen Huang has described such a future on multiple occasions: combining AI computing hardware with open models, allowing companies to establish their own AI systems while retaining ownership and control over their data. According to previous reports by The Wall Street Journal and The Information, Jensen Huang has long been a strong advocate for an open AI ecosystem and has given strong support to the transformation of Reflection towards top-tier open AI systems.

NVIDIA bets $800 million to accelerate the expansion of its computing power territory

NVIDIA plays a key role in the strategic layout of Reflection. NVIDIA has invested $800 million in Reflection and is one of its most important financial supporters.

To support model training and future scale expansion, Reflection has recently been intensively locking up computing resources:

• SpaceX Agreement: In June this year, Reflection and SpaceX ( SPCX ) signed an agreement to pay SpaceX $150 million per month from July 1, 2026, until 2029, after the initial ramp-up period has ended, in order to obtain the right to use NVIDIA GB300 chips within Colossus's 2 data centers. The agreement stipulates that either party may terminate the contract 90 days in advance after the first three months.
• Nebius Agreement: In July this year, cloud computing company Nebius ( NBIS ) agreed to sell over $1 billion in computing power capacity to Reflection, with the contract extended until 2029. On the day of the agreement announcement, the stock price of Nebius immediately fell.

Through the aforementioned SpaceX protocol, Reflection has also obtained a rental channel for NVIDIA hardware, further deepening its binding relationship with NVIDIA.