In the main jump for AI Edge processing, NTT Corporation He announced a groundbreaking AI inference system, which can process 4K video in real time at a speed of 30 frames per second-due to less than 20 watts of power. This new one Large -scale integration (LSI) The chip is the first in the world that has reached such a high -performance AI video in which it is concluded in limited energy environments, which makes it a breakthrough for the computing applications of the edges.
Revealed during NTT Update 2025 peak In San Francisco, the system has been specially designed for implementation in edge devices – software for use physically close to the data source, such as drones, intelligent cameras and sensors. Unlike traditional AI systems, which are based on cloud processing to inference, this system brings strong AI capabilities directly to the edge, drastically reducing the delay and eliminating the need to transmit films about ultra-high resolution to centralized cloud servers for analysis.
Calculation of the edge vs. cloud processing: why does it matter
In traditional cloud processing, data from devices such as drones or cameras are sent to distant data centers – often located hundreds or thousands of miles – where they are processed and analyzed. Although this approach offers virtually unlimited computing energy, it introduces delays as a result of data transmission, which is problematic for real -time applications, such as autonomous navigation, security monitoring and live decisions.
On the other hand, the processing of the edges processes data locally, on the device itself or near it. This reduces delay, retains capacity and enables real -time observations even in environments with limited or intermittent internet communication. It also increases the privacy and security of data, minimizing the need to send confidential data to the public network.
The new AI NTT system fully includes this philosophy of the first edge-getting 4K video analysis in real time directly in the device, without relying on the cloud.
A new era for artificial intelligence in real time on drones and devices
After installing this system, the drone can detect persons or objects from up to 150 meters (492 feet) – a legal limit of height for drones in Japan. This is a dramatic improvement compared to traditional AI systems in real time, which are generally limited to a 30-meter range due to lower resolution or speed of processing.
This progress enables many new cases of use, including:
-
Infrastructure inspections in difficult places available
-
Reaction to a disaster in areas with limited connectivity
-
Agricultural monitoring in wide fields
-
Safety and supervision Without continuous connections in terms of clouds
All this is achieved thanks to a system that consumes less than 20 watts-digestively lower than hundreds of watts required by AI servers powered by GPU, which are impractical for mobile systems or battery powered.
Inside the system: own AI NTT application engine
The LSI efficiency depends on the AI NTT non -standard application, which provides quick, accurate results when minimizing power consumption. Key innovations include:
-
Correlation: When comparing sequential video frames, the system reduces redundant calculations, improving performance.
-
Dynamic bit-consumption control: This technique adapts the numerical precision required in flight, using less bits for simpler tasks, saving energy without prejudice to accuracy.
-
Native Yolov3 Execution: The system supports direct performance You only look once V3One of the fastest algorithms of detecting facilities in real -time machine learning.
These combined functions allow the arrangement to ensure solid performance of artificial intelligence in environments previously considered too much or bandwidth for advanced application.
Path for commercialization and vision IOWN
NTT plans to commercialize the chip in the 2025 tax year through an operational company, Ntt Innovative Devices Corporation.
Scientists are already studying his integration with Innovative optical and wireless network (IOWN)-The new generation of NTTT infrastructure was aimed at reviewing the digital spine of modern society. In Iown's Infrastructure focused on data (DCI)The chip would use the ALL-DHOTONICS network to very low delay, fast communication, supplementing the local processing power, which it introduces to edge devices.
NTT also cooperates NTT Data, Inc.. To combine the possibilities of the system with his Attribute -based encryption (ABE) Technology that allows a safe, fine -grained access control over confidential data. Together, these technologies will support AI applications that require both speed and safety – such as in healthcare, intelligent cities and autonomous systems.
Innovation heritage and vision of the future
This AI inference system is the latest demonstration of the NTT mission involving the strengthening of a sustainable, intelligent society through deep technological innovations. As a global leader with over $ 92 billion in revenues, 330,000 employees and $ 3.6 billion annual research and development, NTT serves over 75% of Fortune Global 100 companies and millions of consumers in 190 countries.
Regardless of whether drones flying outside the visual vision line, cameras detect real-time events without depending on the cloud, or protecting data flows using the attribute-based encryption, the new NTT system prepares the stage on the next limit and on the edge-on the edge-where Intelligence meets directness.
