日本語 
In Japan, the field of artificial intelligence is making great strides thanks to the development of PHOTON (Parallel Hierarchical Operation for Top-down Networks) by Fujitsu, which is an advanced AI technology meant to solve the shortcomings of current Transformer-based large language models (LLMs). It is claimed by the company that PHOTON offers up to 475 times more throughput for each GPU unit compared to traditional Transformer technology under specific tasks.
This advancement comes as an attempt to compete with the widely used Transformer technology, which has been fueling most generative AI systems since 2017. As the popularity of AI grows rapidly, developments that increase efficiency and not just size of AI systems become more and more significant.
Rethinking the Foundation of Large Language Models
The transformer architecture is responsible for bringing about a revolution in the field of AI because of their ability to learn and interpret human language accurately. However, the downside is that transformers are computationally expensive, especially when working with large datasets or multiple users at once.
Also Read: Japan’s Preferred Networks Challenges AI Pricing with Cost-Effective Large Language Models
This has been one of the major obstacles to the adoption of AI in business environments.
Fujitsu’s PHOTON architecture approaches the problem differently.
Rather than process text in terms of single tokens and computing associations between all of them, the information in PHOTON is processed in terms of semantic units which are hierarchically arranged. Through data processing according to meaning and performing several tasks simultaneously, the framework makes great strides towards reducing memory usage and computational cost. The framework also includes multi-query integration, which uses several outputs to integrate responses to boost quality and performance efficiency. As per Fujitsu’s tests, this results in huge throughput gains with much lesser GPU usage than conventional Transformers.
Lower AI Costs Could Accelerate Enterprise Adoption
One of the most notable benefits of PHOTON is that it may be used to bring down the cost of implementing large language models.
Several companies would be interested in using generative AI but incur huge costs in terms of GPU infrastructure, cloud computing, and inference loads. Having an architecture that provides increased performance per GPU may help firms implement AI applications more cheaply and also support more users and loads.
Such a development becomes especially relevant since more and more businesses tend to use AI to enhance their customer service, software development, manufacturing, financial services, healthcare, and business analytics.
If PHOTON lives up to expectations in practice, it may become possible for small and medium-sized firms to implement advanced AI despite its high cost.
Strengthening Japan’s Position in Global AI Innovation
This is a very significant step towards the development of the AI framework in Japan.
Despite the long history of Japanese companies’ achievements in robotics, manufacturing, and industrial automation, the global market of generative AI has been mainly characterized by technology startups from the USA and China.
The decision of Fujitsu to create a brand new AI framework instead of creating models on pre-existing architecture reflects the desire of Japan to become a key player in artificial intelligence in the future.
The company will unveil more information about the architecture during the Association for Computational Linguistics (ACL) 2026 conference that is one of the leading international events in the sphere of natural language processing.
Opportunities Across Japan’s Technology Industry
The potential effects of PHOTON go far beyond AI.
Improvements in language models could lower the need for costly computing capacity as well as present chances for Japanese cloud providers, semiconductors firms, software providers, and system integrators.
Such industries as manufacturing, logistics, financial services, healthcare, and telecommunications could gain from AI solutions with higher workload processing capabilities and lower costs of operations.
This development is consistent with the Japanese government’s strategy for digital transformation that includes more integration of AI in industrial automation, cybersecurity, smart cities, healthcare, and business modernization projects.
Domestic development of AI models, investments in AI infrastructure, and partnerships in enterprise AI projects show that Japan is creating a more holistic AI ecosystem involving software, hardware, cloud, and computing capacities.
Supporting Sustainable AI Growth
Moreover, one can consider the ability of PHOTON to contribute to sustainable development of AI.
Training and operating any AI infrastructure requires electricity; however, training large language models implies usage of very powerful GPU clusters for training and inference. Improvement of the computational efficiency decreases the energy usage, costs, and negative impact on the environment.
With such a situation in which governments and companies seek sustainability together with development of AI, it is likely that architectures maximizing performance with minimum usage of hardware will become increasingly valued.
In addition, efficiency might be useful in solving the problem of shortage of GPUs.
A New Direction for Enterprise AI
The new PHOTON architecture of Fujitsu embodies the bigger trend in the field of AI towards building efficiency into the process of building models.
Further breakthroughs in the area of artificial intelligence may require more than just increased dataset sizes and advanced computing capabilities – a completely new architecture can be essential in making artificial intelligence much more efficient and sustainable.
PHOTON is the chance for Japan to reinforce itself in one of the most fiercely contested areas of modern technology. The development of PHOTON architecture allows Fujitsu to take part in creating the future infrastructure of AI while offering a whole new set of opportunities to companies looking for cheap yet effective AI solution.
With the growing demand of enterprises for generative AI, PHOTON technology is capable of changing the very way AI solutions are implemented.
English