Nvidia (NVDA.O) has released new data demonstrating that its latest AI servers can accelerate the performance of emerging artificial intelligence models—including China’s Moonshoot AI—by a remarkable tenfold. The announcement underscores Nvidia’s continuing dominance in AI training hardware and its growing focus on deploying AI models at scale for millions of users worldwide.

---

Mixture-of-Experts AI Models Drive Efficiency

The recent performance gains were observed in mixture-of-experts (MoE) AI models, a technique that divides complex tasks into smaller subtasks, each assigned to specialized “experts” within the model.

The MoE approach gained popularity in early 2025 after China’s DeepSeek launched a high-performing open-source model that required less training on Nvidia chips compared to competitors. Since then, major AI developers—including OpenAI, France’s Mistral, and China’s Moonshoot AI—have adopted MoE strategies to enhance model efficiency.

---

Nvidia’s Latest AI Server Technology

Nvidia highlighted that its newest AI server configuration packs 72 of its leading chips into a single machine with high-speed interconnects, enabling unprecedented performance for large-scale AI inference.

For instance, Moonshoot AI’s Kimi K2 Thinking model achieved 10x faster performance on Nvidia’s latest servers compared to the previous generation. Similar performance boosts were observed with DeepSeek’s models, demonstrating Nvidia’s advantage in chip density and high-speed connections.

Nvidia emphasized that while MoE models may require less training, its servers excel in delivering high-performance inference to end users—an area where rivals like AMD and Cerebras are still catching up. AMD is reportedly preparing a multi-chip AI server expected to enter the market next year.

---

Implications for AI Deployment

The performance gains position Nvidia to maintain a leadership role in AI infrastructure, particularly in running models efficiently for large-scale applications. As AI adoption expands, businesses and developers are increasingly focused on model serving rather than just training, creating new opportunities and competition in the high-performance AI hardware market.

Nvidia’s advancements may also influence the global AI ecosystem, encouraging developers to optimize models for its hardware while accelerating the deployment of next-generation AI applications.