Kyocera’s licensing of Quadric’s GPNPU could kick-start on-device print AI

May 21, 2025
Artificial Intelligence, Artificial Intelligence, Article, Trends

Kyocera’s licensing agreement with Quadric for its Chimera General-Purpose Neural Processing Unit (GPNPU) technology signals a potential acceleration of AI integration within the print industry. Quadric’s GPNPU is designed for efficient on-device AI processing, offering a blend of neural network acceleration and general-purpose computing capabilities. This licensing agreement suggests Kyocera is strategically investing in incorporating advanced AI capabilities into its future printing solutions.

What is Quadric’s Chimera GPNPU?

Quadric’s Chimera GPNPU is a licensable processor that scales from 1 to 864 tera operations per second (TOPS). Chimera GPNPUs run all machine learning (ML) networks – including classical backbones, vision transformers, and large language models (LLMs). Quadric’s Chimera architecture represents a novel approach to on-device AI processing, uniquely integrating the high-throughput parallel processing capabilities essential for neural network acceleration with the versatility of a general-purpose processor within a unified silicon architecture. This design allows the Chimera GPNPU to efficiently handle a diverse range of AI inference tasks and traditional computing functions on the device itself, without constant reliance on cloud connectivity.

The potential of integrating such on-device AI capabilities, particularly within the system-on-chip (SoC) of multifunction printers (MFPs), could be transformative. By processing AI tasks locally, MFPs could achieve:

• Enhanced speed and responsiveness. Real-time analysis and decision-making for tasks such as image optimisation or document classification can be carried out within the device itself, without the latency associated with cloud communication.
• Improved security and privacy. Sensitive document content and user data can be processed and analysed locally, reducing the need to transmit information to external servers and mitigating potential security risks. Decisions can be made on the device in areas such as data classification, data leak prevention (DLP), and information redaction without any information leaving the enhanced security of the devices itself.
• Greater reliability and autonomy. As well as the simple print, scan, and finishing capabilities of the MFP itself, the advanced functionality brought via AI can remain available even without a stable internet connection, enhancing the reliability of intelligent features.
• Reduced operational costs. Although it is likely to have fewer cost savings, minimising any reliance on cloud processing can lead to lower data transmission and cloud service fees.
• New intelligent features. Embedded AI can help by enabling more complex on-device functionalities such as advanced document understanding, proactive security threat detection within print streams, personalised user interfaces driven by usage patterns, and intelligent workflow automation embedded directly within the MFP.

Is Kyocera ahead of the game?

While the specific implementation and timelines for Kyocera’s integration of the Chimera GPNPU remain to be seen – with the actual SoC and silicon expected to take approximately two years to materialise – this licensing agreement signals a clear intent to leverage the power of on-device AI. By indicating plans to embed Quadric’s innovative GPNPU technology directly into their MFPs, Kyocera is positioning itself to potentially lead the charge in developing a new generation of intelligent AI-enabled MFPs that offer enhanced performance, security, and document automation.

This move could be the catalyst that kick-starts broader adoption of on-device AI across the print industry, paving the way for smarter, more efficient, and more secure print and document capture experiences.

However, although the Chimera GPNPU brings a lot of promise to the use of embedded AI, it will need an ecosystem of developers behind it to make it a success. Quadric is making the architecture as open as possible so developers will not need to learn new languages, and the use of AI itself in development (for example, through GitHub Copilot) could lead to fast delivery of AI functionality that could accelerate the adoption of AI-embedded SoC MFPs. Kyocera must also play its part, using its own reach into the developer community and the capabilities of its channel to help create a vibrant set of AI models that buyers will see as disruptive in the market. These must also provide distinct value to buyers’ document management and workflow needs.

Market implications

Traditionally, AI in print has been largely confined to predictive maintenance and security enhancements, but Quadric’s GPNPU technology could expand its reach into real-time document analysis and intelligent data processing. With businesses handling ever-growing volumes of structured and unstructured data, AI-powered print solutions could streamline document classification, enhance compliance automation, and improve security through automatic fault resolution and intelligent redaction and anomaly detection.

The integration of a powerful and efficient AI processor such as Quadric’s GPNPU could unlock a range of more sophisticated applications, including:

• Advanced image analysis and enhancement. AI can provide a large improvement in the real-time optimisation of print quality based on image content, potentially leading to reduced waste and improved output.
• Intelligent document processing. The use of AI around automated document classification, data extraction, and workflow optimisation directly within the printing device can provide distinct business value to the user.
• Enhanced security features. AI-powered threat detection and prevention within the printing environment, protecting sensitive information, can help provide the additional security that print devices have historically lacked.
• Personalised printing experiences. Print output and workflows can be tailored to individual user, group, and organisation preferences and needs.
• Predictive maintenance and resource management. Sophisticated AI algorithms can aid in better predicting maintenance needs, optimising energy consumption, and managing supplies. This can help reduce the need for on-site engineer visits, while just-in-time (JiT) consumables management can reduce OEM/channel inventories and customer storage needs.

Kyocera’s move, even with the anticipated development timeline, could encourage other print industry players to explore similar AI integration strategies. The ability to perform complex AI tasks on-device, rather than relying on separate graphics processing units (GPUs) and natural language processing (NLP) within the device, or on completely off-device cloud processing, offers potential advantages in terms of speed, security, and reduced latency. This could lead to a new generation of printing devices that are not just output devices but also intelligent processing hubs within the office environment.

While the specific applications and timelines for Kyocera’s implementation remain to be seen over the next two years, their licensing of Quadric’s GPNPU technology represents a potentially pivotal moment, suggesting a future where AI plays a far more central role in the print industry. It will be important to monitor Kyocera’s development progress and whether this strategic adoption indeed catalyses broader AI integration across the sector in the coming years.

SoC architectures in MFPs

Other OEMs have been using an SoC approach for certain functionality in the past, and are now building on this to create on-device or hybrid AI platforms.

As a basic idea, a printer SoC is an integrated circuit that combines all the necessary components for a printer’s functionality, or a specific area of its functionality, into a single microchip. This includes elements such as the CPU, GPU, digital signal processing (DSP), memory, power management unit (PMU), and I/O ports required by the SoC to interact with the rest of the print device all on one chip. Printer SoCs are designed to provide optimised space, power consumption, and performance in printing devices.

The use of SoCs has history the printer arena. In 2021, Fujifilm Business Innovation started using Qbit’s SoC in some of its Apeos line of MFPs. Qbit was born out of a cooperation between Fuji and Xerox. As Xerox still uses Fuji’s print engines (although Xerox’s acquisition of Lexmark gives it access to a level of in-house designed and built engines), its SoC capabilities are likely to be similar to Fujifilm Business Innovation’s for a while.

Also in 2021, Kyocera Document Solutions signed an agreement with Synopsys to use its DesignWare ARC EV6x Embedded Vision Processor IP with its convolutional neural network (CNN) engine and ARC MetaWare EV development toolkit. As part of this, Kyocera also put in place the tools to create an AI ecosystem using Synopsys’ HAPS FPGA-based prototyping system.

In 2022, Konica Minolta progressed its dual chip CPU/ASIC design to being a single SoC approach, providing an upgradable platform.

With Kyocera’s adoption of GPNPU technology, print vendors now have the potential to reimagine their embedded AI strategies. If successful, this could accelerate the industry’s shift towards smarter, self-optimising print ecosystems. However, with each print OEM seemingly going its own way, creating the necessary vibrant ecosystem around print device AI may be difficult – the same problem that the tech industry has faced repeatedly. Without some standardised way of creating viable solutions that are portable across devices, users may find sticking with device-independent capabilities in the cloud more appealing.

Quocirca opinion

Kyocera’s GPNPU licensing as an important milestone. Its move towards GPNPU-based AI is a key indicator that print manufacturers are recognising the strategic importance of AI-driven workflow automation. The ability to process documents more intelligently – whether through automated classification, metadata extraction, or security enhancements – could reshape enterprise print strategies, as long as buyers can see the value in such an approach. While AI adoption in the print sector has historically been cautious, the integration of GPNPU technology could push competitors to accelerate innovation in embedded AI solutions.

Quocirca’s research suggests that print industry AI adoption will gain momentum if early implementations demonstrate tangible business value, efficiency gains, and cost benefits. While it remains to be seen whether this and other OEM-led AI development will catalyse broader AI adoption across the industry, it undoubtedly raises the bar for intelligent print solutions. If Kyocera can demonstrate clear efficiency gains, competitors may be compelled to fast-track their own AI-driven initiatives, moving away from multi-chip and cloud-based solutions to a more centralised GPNPU SoC approach, accelerating the evolution of the print sector into a more dynamic, data-driven ecosystem.

To find out more about how print vendors are applying AI to their products, read Quocirca’s AI Vendor Landscape 2025 Study

To find out more about becoming a Quocirca client, click here.