Less hype, more power: SenseTime leverages multimodal AI to reclaim its lead

Daftar Isi

The Evolution of SenseTime: From Vision to Embodied Intelligence

SenseTime, once a prominent name among China’s "AI dragons," has been quietly redefining its role in the evolving landscape of artificial intelligence. While the company may have stepped back from the spotlight in recent years, its strategic shift towards multimodal AI and embodied intelligence signals a potential resurgence. This move aligns with broader industry trends that emphasize the integration of multiple AI modalities—such as vision, language, and robotics—to create more sophisticated and practical AI systems.

Lin Dahua, co-founder and chief scientist at SenseTime, highlighted the company's long-standing expertise in computer vision as a key asset in this transition. He emphasized that SenseTime's foundational strengths in visual recognition place it in a unique position to lead in the development of embodied intelligence, which refers to AI systems that interact with the physical world through robots or other intelligent agents.

Lin drew parallels between SenseTime's approach and that of Google, noting that both companies focus on multimodal AI. He mentioned Google's recent advancements in multimodal systems, such as the Nano Banana Pro, as an example of how vision capabilities can be integrated with language models to create more robust AI solutions.

Building a Strong Infrastructure for Future Growth

One of the critical factors contributing to SenseTime's current position is its early investment in large-scale data centers. Since 2018, the company has been expanding its computing infrastructure, which now boasts a total computing power of approximately 25,000 petaflops. This capacity has seen significant growth, increasing by 8.7% since the start of the year and surging by 92% throughout 2024.

A petaflop represents one quadrillion calculations per second, making it a crucial benchmark for AI infrastructure. According to Yang Fan, president of SenseCore, SenseTime's AI infrastructure unit, the company plans to continue expanding its computing power at a "high double-digit to triple-digit" pace over the next two years.

This expansion is not just about raw computing power; it also reflects a broader strategy to support the development of advanced AI models and applications. SenseTime has been proactive in adopting domestically produced chips, with the share of its computing power provided by these chips now reaching "double digits."

Navigating Challenges and Rebuilding Momentum

Despite its technological advancements, SenseTime has faced challenges, including being placed on a U.S. blacklist. However, the company has continued to innovate and adapt. Lin acknowledged that while SenseTime's flagship model, SenseNova, may not have the most advanced language capabilities, its strengths in multimodal AI are particularly valuable for embodied intelligence and robotics.

These areas have received significant government support in China, giving SenseTime a competitive edge. Additionally, the company's long-term relationships with major enterprise clients provide access to valuable data for training the next generation of AI agents and robots.

SenseTime has also been supporting robotics startups, such as Ace Robotics, led by co-founder Wang Xiaogang. The startup is set to launch its first robot dog, showcasing the company's commitment to advancing real-world AI applications.

A Strategic Shift for Sustainable Growth

In response to market dynamics, SenseTime has undergone a restructuring to refocus on core AI activities. This strategic move has yielded positive results, with the company narrowing its adjusted first-half loss by 50% compared to the previous year. Revenue from its generative AI division has surged by 72.7%, contributing significantly to overall revenue.

The company's shares have gained roughly 32.5% over the past 12 months, reflecting investor confidence in its long-term vision. Lin emphasized that while some companies may release models frequently to generate media buzz, sustained success requires deeper technological expertise, data accumulation, and strong infrastructure.

As the AI landscape continues to evolve, SenseTime's journey from a leading facial recognition provider to a pioneer in multimodal and embodied intelligence highlights the importance of adaptability and long-term strategic planning. With its focus on real-world applications and continuous innovation, the company is well-positioned to play a significant role in shaping the future of AI.