Free NewsletterPro Login

Free Live Investors Workshop

Seats limited

Tue, Sep 29.

The dollar is losing value.

Here’s how investors can still profit.

Hosted By

Jaspreet Singh

Founder, Briefs Finance

X
Free Live Investor Workshop

Chinese AI Startup Moonshot Aims to Acquire Additional Nvidia Blackwell Processors for Upcoming Model, Report Says

Published Jul 29, 2026
Share:
Modern data center aisle with GPU racks glowing blue
Summary:
  • Moonshot AI is seeking more Nvidia Blackwell chips to train its next Kimi K4 model.
  • US export controls restrict sales of advanced processors to China due to security concerns.
  • The debut of Moonshot's Kimi K3 triggered a selloff in AI stocks, echoing the market reaction to DeepSeek in early 2025.

What Exactly Is Moonshot After

The company's previous model, Kimi K3, was trained on Nvidia hardware as noted by Bloomberg, despite U.S. export restrictions that prohibit the sale of such cutting-edge processors to China due to national security concerns.

The startup's push for Blackwell chips underscores the critical role of hardware in the AI arms race. Nvidia's Blackwell architecture, the latest and most powerful series, is subject to strict export controls that restrict its sale to Chinese entities. While Moonshot managed to train Kimi K3 using older or alternative chips, the company appears to believe that Kimi K4 will require the full capabilities of Blackwell to maintain a competitive edge.

Broader Implications for the AI Race

The latest developments highlight ongoing tensions between US export policy and the fast-growing Chinese AI sector. Since 2022, Washington has steadily tightened restrictions on advanced semiconductors and manufacturing equipment, aiming to slow Beijing's progress in military-critical technologies. However, Chinese firms have shown remarkable ingenuity, finding ways to develop competitive models using less powerful hardware or through algorithmic efficiency gains.

Get the market news that matters in a five-minute read with Market Briefs, our free daily newsletter

DeepSeek's breakthrough in early 2025 demonstrated that Chinese AI can rival Western models even with limited access to top Nvidia accelerators. The pattern suggests that the gap between Chinese and American AI capabilities may be narrowing faster than previously assumed - a trend that could have significant implications for global tech valuations, trade policy, and national security planning.

Yet Moonshot's current pursuit of Blackwell chips reveals a nuanced reality: despite impressive results with constrained resources, leading Chinese AI labs still covet the most advanced hardware. If they cannot legally obtain Blackwell, they may resort to gray-market channels or rely on stockpiled chips, as some analysts have speculated. The outcome of this pursuit could influence not only Moonshot's next model but also the broader geopolitical narrative around technological decoupling.

A Familiar Pattern

If the story sounds familiar, it should. This trend is expected to heighten concerns across the Atlantic regarding Chinese AI systems narrowing the performance difference with OpenAI and Anthropic.

Observers viewed the K3 launch as confirmation that Chinese enterprises can develop competitive AI using fewer resources and less advanced chips than companies that have unrestricted access to Nvidia's best accelerators.

For investors, the recurring pattern raises key questions. If Chinese AI can keep pace without Blackwell, what is the long-term moat of Western AI companies? Conversely, if Moonshot and others are desperate for more powerful chips, then the true bottleneck may still be hardware - meaning export controls could remain effective. Either way, the next few months, particularly the release of Kimi K4, will provide critical data points for the market.

Join Market Briefs, our free daily newsletter, for a quick daily rundown of the markets

Disclosure

Recent News

1 2 3 78

Get Market Briefs delivered to your inbox every morning for free!

No fluff. No noise. No politics. Just finance news you can read in 5 minutes.
0 Shares
Share via
Copy link