Back

Taiwan ICT Industry Builds AI DX Supply Chains for Various Industries, GenAI Becomes an Opportunity to Get Ahead


 Publish Date :2024/07/29

The rapid growth of generative AI (GenAI), large language models (LLM), and the need for digital transformation; this year’s COMPUTEX has become the focus of the global ICT industry. At the same time, the event also promoted Taiwan’s ICT industry which gained a lot of attention from the international media.

To continue the energy of COMPUTEX and actively share the opportunities and challenges that Taiwan’s ICT industry will face in the new industrial AI era; Taipei Computer Association (TCA) and Monte Jade Science and Technology Association of Taiwan hosted a conference titled Embracing The New AI Era - Looking At The Next Steps Of Taiwan's ICT Industry from COMPUTEX on July 23. Despite being held at the same time as the annual Wan An air raid drill and right before the landfall of typhoon Gaemi, the event still attracted many domestic and overseas media and VIPs.

The conference was led by Chris Hung, Market Intelligence & Consulting Institute (MIC) Deputy General Director who delivered a keynote speech titled “Looking at the Next Steps of Taiwan’s ICT Industry from COMPUTEX” and moderated the panel discussion. The keynote speech analyzed the future advantages of Taiwan’s ICT industry and discussed the trend of GenAI as well as the evolution of the ICT industry ecosystem. The panel discussion featured executives from various leading global level Taiwanese ICT companies including:

  • Honorary Chairman of TCA and Chairman of the Monte Jade Science & Technology Association of Taiwan, TH Tung
  • Acer Chief Operating Officer, Jerry Kao
  • Director of MiTAC Holdings, Liang Su
  • Taiwan Web Service Corporation Chief Strategy Officer, Kevin Lee
  • Chairman of SOLOMON Technology Corporation, Johnny Chen
  • CEO & Founder of Kneron, Dr. Albert Liu

Paul SL Peng: Computing Power, Electricity, and Human Talent can Enhance Taiwan’s Industrial Competitiveness in the new AI Era

The Chairman of TCA, Paul SL Peng stated in his opening speech that this year’s COMPUTEX was very successful as it gathered so many AI industry leaders in Taiwan. As the focus of the ICT industry shifted towards AI, COMPUTEX has also become the place for AI digital transformation products and solutions. COMPUTEX 2024 also gathered a truly unprecedented number of VIPs, buyers, and professional visitors from all over the world as well as hundreds of matchmaking sessions during the event.

The products and solutions showcased during COMPUTEX showed how computing power is national power and computing power requires electricity. To address this, President Lai promised during the opening of COMPUTEX that Taiwan will have enough electricity. Of course, enough electricity is not the end goal as enough green energy and better energy ratio are expected. The government also aimed to provide more low-carbon energy in the future. The other focus is on human talent cultivation.

Mr. Peng stated that the 3 factors of computing power, electricity, and human talent can promote Taiwan’s ICT industry to new heights in the new AI era. By discussing the next development of Taiwan’s ICT industry, it is possible to enhance the industry together. The 3 factors will further improve Taiwan's status and industrial competitiveness while continuing the momentum of COMPUTEX. The hope is that it can reflect the momentum of Taiwan’s industry.

Chris Hung: AI Drives the Evolution of the ICT Industry Ecosystem. Taiwanese Industries Must Leverage AI to Get Ahead of the Competition

Market Intelligence & Consulting Institute (MIC) Deputy General Director, Chris Hung delivered a keynote speech titled “Looking at the Next Steps of Taiwan’s ICT Industry from COMPUTEX”. In the speech, he stated that this year’s COMPUTEX had strongly promoted AI PCs, AI servers, and various AI-related products from Taiwanese companies in addition to gathering technology giants in Taipei. From the perspective of MIC, many of them are cooperative endeavors.

Chris Hung stated that the AI trends right now, especially GenAI, are not in a bubble. Data, computing power, and algorithms are the 3 main cornerstones of GenAI applications. GenAI has developed towards multi-modal applications, including texts, pictures, audios, videos, and other formats. GenAI has become capable of performing a variety of tasks such as answering questions, generating texts, generating images, and data extraction. In the future, AI will develop towards becoming AI agents, transforming AI from a tool to a working agent. These changes in the AI industry are directions that industries related to Taiwan’s AI ecosystem should pay attention to.

AI as a technology drives the evolution of the ICT industry ecosystem. It not only allows the industrial ecosystem, but also it help realize the industrial ecosystem built by users, model developers, and cloud operators; it also creates new patterns. The large number of training models and computing power requirements have also created many opportunities as well as challenges. The development of GenAI will bring many development opportunities to various industries through cross-domain cooperation and innovative collaboration. From the perspective of the AI ecosystem, Taiwan’s infrastructure such as the AI computing power plays an important global role. In addition, Taiwan also has organizations that build AI models.

In terms of computing power, Chris Hung believes that AI computing power is not only national power, but the source of national competitiveness because AI will have a ubiquitous presence. In the future, AI will be applied in various settings, including: food, clothing, housing, transportation, etc. which means when there is an abnormality in the AI computing, the error will affect organizational operations. The recently growing trend of AI PC supports the demand for hybrid AI and collaboration. In the future, whether the AI will be on the cloud or on-premise, the demand for computing integration will only increase.

He further stated that the use of GenAI to improve efficiency is an ongoing trend in all walks of life, including in smart factories, smart medical, and other similar fields. The main challenge is in choosing the right model and confirming the results from said models, which will require the domain know-how of experts in the relevant fields. In addition, the lack of talent and budget are going to be the biggest challenges for introducing AI in the industry. After all, companies will have to make cost considerations and talent allocations. Another key factor is that the effectiveness and efficacy of introducing AI is not easy to measure. This will greatly affect the willingness of organizations to introduce AI to their operations.

Chris Hung believes that Taiwan’s ICT industry must leverage AI to grasp digital transformation opportunities and the next wave of development trends. It is important to utilize emerging technologies to create advantages, identify the different changes made possible by digital optimization and digital transformation, and consider the effects of the transformation beforehand to avoid the high risks caused by changes when the market turns. At the end of the day, it is important to remember that time is the most expensive resource for companies.

TH Tung: Taiwan has a Good Position in AI Hardware Manufacturing, but AI Applications Require Fine Tuning of AI Models

The Honorary Chairman of TCA and Chairman of the Monte Jade Science & Technology Association of Taiwan, TH Tung stated in the panel discussion that AI is not a singular monolithic product. AI will be similar to the internet and become an important technology trend that will be integrated with and affect every hardware. From sensor chips, photographic lenses, and even rackmount servers will have various degrees of AI integration. AI applications will not only happen on the hardware side, but also software; including built-in AI software in smartphones or office programs, all of which have started to have AI integration in them. The next development will see AI integration in service platforms, introducing the idea of AI services.

AI models launched by various manufacturers, such as the GPT-4o are easy to use in certain languages, especially English; however, the actual use in other languages such as Traditional Chinese in Taiwan requires fine tuning. It is important to remember that specific languages have different accents and linguistic nuances which may cause some difficulties in usage.

So, if Taiwan aims to lead in the AI field, there are 2 aspects that it can succeed in. The first one is in production; from chip manufacturing to product production capabilities such as in rackmount servers/chassis, PCBs, systems, AI PCs, AI smartphones, etc. Taiwan currently has a strong position in hardware manufacturing. The other aspect is in application. Taiwan has always emphasized the concept of “hard over soft”, meaning that Taiwan has focused on production over application for a long time. If Taiwan wants to apply AI in productivity, manufacturing, daily life, etc., fine tuning the model will be the focus because the quality and speed will affect the AI productivity and quality.

In the face of future AI applications, Mr. Tung believes that moving AI from the cloud to device around its users to enable edge computing in the next few years will be important. First of all, edge computing can respect personal privacy, democratize AI, and provide AI computing power in easy reach of everyone; which will greatly help AI users.

In the field of professional AI, many AI models are updated faster; making them better assistants in the laboratory or in the office and human-controlled. For example, AlphaFold 3 which is the latest version launched by Google can simulate more protein folding interactions with the human body, medications, and diseases. The previous version - AlphaFold 2 has simulated over 200 million protein molecules, which is very important in developing drugs for terminal illnesses. AI can analyze and process huge amounts of image data far better than humans and excels in it. Just like the invention of the Watt steam engine, AI brings a lot of motivation for people.

Regarding the issue of electricity, TH Tung pointed out that due to the industrial structure, the only advanced country with similar per capita electricity consumption or larger than Taiwan is only the US. Other countries such as Japan or Germany have a large proportion of electricity consumption in the service industry and their per capita electricity consumption is about one-half or less than in Taiwan. Due to Taiwan’s work patterns, Taiwan’s domestic electricity consumption actually accounts for only 18% of the total electricity consumption. Conversely, the industrial electricity consumption accounts for 56%, while the service industry makes up for 19%, and the rest is consumed for agriculture and other industries.

Taiwan’s electricity consumption in 2023 reached 282.14 TWh. Following a growth rate of 2% to 2.5%, it will exceed 300 TWh in 3 to 4 years. By that time, if the cost of electricity generation decreases by NTD 0.5 (±USD 0.015) annually, there will be a difference of NTD 150 billion (±USD 4.6 Billion) and if the difference is NTD 1 (USD ±0.03), there will be a difference of NTD 300 billion (±USD 9.3 Billion). Many segments in Taiwan require significant budgets such as higher education, AI research, building sovereign AI, building sufficient AI computer rooms with sufficient computing power, etc. If the energy aspect can be matched with a combination of high-quality energy and high economic efficiency, the funds can be reallocated for health insurance, higher education, and research needs which is more meaningful.

Regarding Taiwan’s energy policy, TH Tung pointed out during a joint media interview that Taiwan should use a high-quality and economically efficient power generation mix to improve power supply stability and energy resilience. This will also enable Taiwan to respond rapidly to international tensions and consider the impact of various energy sources on the land, agriculture, environment, ecology, and landscape. He referred to the case of Italy announcing the reactivation of a nuclear power plant that had been inactive for 35 years because the Italian government believed that there is insufficient land for solar panels and both Italian citizens and tourists do not want solar panels marring scenery. However, they still need to achieve carbon reduction and reduce their over-reliance on solar energy, which is why the Italian government plans to return to nuclear power.

On the progress of nuclear energy technology and nuclear waste treatment, TH Tung shared that from a scientific and commercial perspective, nuclear waste is radioactive material that can no longer be used before it is fully used up. Just like how barbecue restaurants will remove the charcoal in their customers’ stove when only about 20% remain. Currently, the concentration of Uranium-235 in natural form is only 0.7%. After refining it to a concentration of 3%, the Uranium-235 fuel rods can be used as a low enriched uranium (LEU) for nuclear power plants. The conventional design of nuclear power plants are unable to use up 100% of the fuel rods, but there is a scientific opportunity to achieve it. In reality, the current result is due to business decisions. It is not because scientists are no longer researching nuclear energy generation and nuclear waste disposal methods.

To address the doubts of nuclear-skeptics about the storage of nuclear waste, TH Tung shared that in his observation, Taiwan’s nuclear waste is currently placed in a controlled area with limited access and it is carefully managed in a confined space, posing minimum hazard. In addition, the advancement of new technologies has made it possible for Generation IV (Gen IV) reactors that can utilize nuclear waste. Technologies such as fast-neutron reactors and traveling wave reactors are new nuclear technologies under development including in China, Canada, the UK, and the US are all competing to invest in their R&D. If the technology is successfully developed, the current issues of nuclear waste and depleted uranium which are the remnants of the uranium enrichment process can be converted into nuclear fuel for a new generation of reactors. This is the ideal solution for the current nuclear waste problem. He also pointed out that if the amount of green energy installations in Taiwan were to be increased by tenfold, it is possible that in the next 20 years, the amount abandoned green energy devices and systems will be equivalent to the size of several Taipei cities, which will have a significant impact on Taiwan’s environment.

Jerry Kao: AI Killer Applications in the Commercial Market are Necessary to Create a Sustainable AI Competitive Advantage

Acer Chief Operating Officer, Jerry Kao stated that the next step that will really put Taiwan in a good position in the AI trend is to apply it in the commercial market. Quoting the Maslow Hierarchy of Needs, he pointed out how AI holds different places for enterprises and users. In the consumer market, AI is something “nice to have”. Having AI is better, but if it’s not there, it will not make much of a difference. In the commercial market, AI is a game changer. If a company does not have AI, it might be outperformed by its competitors and lose its market share.

Jerry Kao mentioned that most Taiwanese companies are AI hardware manufacturers and their main income comes from hardware. However, it is important to consider how Taiwan's ICT industry can not only earn short-term income from hardware, but also gain real and long-term income from applications, software, and other AI killer applications. One hope he has from the government is that they will confirm the killer application of enterprise AI applications and integrate relevant resources for their developments. This way, the benefits will go to not only the software/ solutions companies, but also the related hardware manufacturers to create a win-win situation.

Regarding this year’s core topic of AI PC, Jerry Kao believes that the situation of AI PC is wholly unique. Compared to the launch of the smartphone in 2009, it is capable of replacing products that are smaller or about the same size such as voice recorders, cameras, and MP3 players; but not to replace PCs. A lot of work still has to be done on PCs, as seen in the rebound in PC demand during the pandemic. The emergence of AI PCs is more similar to the launch of the internet or WiFi. More new AI applications will emerge, allowing people who are interested in AI to use them. Once a new killer application for AI PC appears, demand for the technology will certainly grow rapidly.

Johnny Chen: AI Effectively Improves the Accuracy of Image Recognition as Software Platforms Reduce the Required Number of Samples to Build Customized AI Models

Chairman of SOLOMON Technology Corporation, Johnny Chen stated that after working on robotic applications for about 10 years, his biggest impression is that human beings are powerful because they are born with great dexterity and versatility. In the early days, the tasks for robot arms were very simple, mostly processing a fixed task from point A to point B. With the coming of image processing, they can do a wider range of tasks. However, the traditional rule-based vision system has many limitations, including the effect of changes in lighting, complex object shapes, or other factors that will affect the accuracy of image recognition. In recent years, AI has begun to be used in visual applications which can be used to solve problems that were difficult to solve in the past. In the technology industry, AOI (automated optical inspection) has been used in the past, but this often resulted in the problem of the overly high overkill/ false positive rate. The introduction of AI might address this issue to make a much more effective image processing system.

Johnny Chen said that when SOLOMON Technology customers ask about using 3D vision for AOI or monitoring, they all ask how many pictures are needed to generate an AI model. When they realized that they need to provide several thousand data points, many clients become discouraged. So, the SOLOMON team thought about how to use the software platform to allow customers to build customized AI models with the minimum number of samples to increase the customers’ willingness to adopt the AI. Currently SOLOMON can connect 20 to 30 robots of various brands, including a variety of 2D and 3D image formats and IP cameras of various brands, and optimized AI models which will be placed on the platform for actual use cases.

Johnny Chen stated that the main reason why customers are currently reluctant to use AI robots is that a sufficient number of databases are still needed to train AI models. In terms of quality inspection, the theoretical yield rate of the production line is close to or even exceeds 90%. On the contrary, if the number of defective samples is too small, it may not be possible to train the AI model. However, this has been greatly improved from the traditional AI model that requires thousands of samples for training purposes.

To solve the issue of staff shortage with the most popular AI robot, Johnny Chen believes that it is necessary to be almost as smart and versatile as a human. NVIDIA is currently developing some basic AI models such as the jointly developed Motion Panning, which can be incorporated into the platform for use. If future machines are able to perform complex tasks such as assisting in hair cutting and people feel confident it its performance, this will be the dawn of an era where robots are everywhere.

Regarding the problem of talent shortage, Johnny Chen said that SOLOMON Technology has relied on international talents for a long time. In the visual department alone, there are employees from 15 countries. Some came to Taiwan to study for their master’s or doctorate, some were introduced by their friends from countries such as Vietnam or India. This is a path we need to take, after all it is not easy to compete with major tech companies.

Liang Su: Common Data Platform with No Code AI Tool can Accelerate the Collaborative Development of Cross-Domain AI Applications

Director of MiTAC Holdings, Liang Su said that AI should be everywhere, but during the application it seems that there is something wrong or missing with it. Previously, there were 2 key points when promoting various AI applications. The first is to have a sufficient database. This database should not only include general data, but also data obtained from IoT devices, big data cloud servers, open data, etc. The goal of MiTAC is to help industry players create an common data platform. Only with such a platform will the data be usable for AI analysis.

Liang Su pointed out that after having the common data platform, another problem was discovered: the need for multi-domain collaboration and platform innovations. AI applications are all cross-domain and some span multiple fields. So multi-domain collaboration and platform innovations are necessary to create new AI applications. Now there are many AI tools which may not be difficult for tech-savvy users, but when it comes to cross-domain applications, it is still difficult to use. So MiTAC created a No Code AI tool that does not require any programming. This will help people from all walks of life to develop more innovative applications through No Code tools and the data on the Common Data platform.

In regards to smart cities, the most important thing for the government is to create real impact and changes. The easiest way to achieve this is by developing tools that can improve the UI and applications that use AI for UI design as well as improving efficiency and citizen safety. In other words, AI is a very versatile tool that can achieve various purposes.

Speaking on the Taiwan talent cultivation, Liang Su mentioned the launch of Intel microprocessors in the 1970s. By promoting education, it is possible to teach everyone how to use microprocessors. However, the main training will be in developing talents in professional fields. In terms of training AI talents, because AI is cross-domain, many people do not need to know AI. As long as they have innovative ideas and suitable tools, they can make it into reality, which is different from the past. MiTAC Holdings is currently developing internal plans and provides AI laboratories to senior high schools. The aim is that by training senior high school students, they will have innovative ideas and can succeed in any field.

Kevin Lee: Implementation of Enterprise AI Brain Relies on Utilizing the Relevant Datasets

Taiwan Web Service Corporation Chief Strategy Officer, Kevin Lee stated that GenAI has opened up new possibilities in many industries; including the financial industry where it is primarily used to automate front-end businesses through LLM, handling B2C businesses, and conducting financial analysis services. In the medical industry, one of the main applications is through traditional Speech to Text, which can convert a physician’s instructions to text during the consultation and use LLMs to summarize the instructions. The summary can even be processed by specialty and in the long run, a large amount of medical data can be analyzed through GenAI.

Taiwan has mostly introduced GenAI in the manufacturing industry, including by analyzing data through LLMs, issuing instructions, and even scheduling traditional machine learning (ML) models. These steps have returned good results in production parameters and quality assurance predictions. For traditional Chinese users, Taiwan Web Service Corporation has also added traditional Chinese training to the Llama 3 model for fine tuning. They have also done fine tuning in many professional fields in the manufacturing industry, including for the 70B or 8B models. Enterprises use fine tuning to solve specific tasks and make production process predictions.

Kevin Lee believes that from a commercial perspective, AI will have an increasing impact on enterprises and GPU computing power will be cheaper and cheaper. As a result, enterprises can have Enterprise AI Brains that not only understand traditional Chinese data, but also understand vertical data within the enterprise. If enterprises aim to build their own Enterprise AI Brain, they need to look at the data sets they own. Only with a combination of useful data and sufficient computing power and appropriate models, settings to solve the problems, and input from external consultants; will it be possible to introduce AI in the enterprise. It is important to not integrate AI simply for the sake of integrating AI.

Kevin Lee also mentioned that Taiwan Web Service Corporation has helped many companies implement 70B models in the financial, manufacturing, and medical industries; as well as IC design, heavy industry, and more with good results. The latest Llama 3.1 from Meta includes the 405B model, as well as the 70B and 8B models. Since the 405B model is benchmarked against GPT-4o, it seems that there is a chance that it can be implemented.

Albert Liu: Utilizing NPU for On-Premise Enterprise GPT Solutions can Ensure the Privacy of Enterprise Data and Operate with Low Energy Consumption

CEO & Founder of Kneron, Albert Liu stated that due to the cost factors of training models and GPU servers, many GenAI applications, including GPT (Generative Pre-Trained Transformer) are currently executed in the cloud. In addition to the security issues of uploading data to the cloud, users will also encounter different answer during every inquiry, which creates problems for corporations aiming to introduce AI to their operations.

Taiwanese manufacturers have great opportunities in To B GenAI applications while Kneron is the developer and patent/ trademark holder for NPU and owns key technologies such as NPU accelerators, AI chips, and AI models. So there are ways to embed NPU into devices to make the device AI-capable.

Albert Liu stated that because most of the opportunities for Taiwanese companies are in the To-B GenAI applications, Kneron has launched an on-premise AI software and hardware integration solution based on NPU with a built-in Local RAG (Retrieval-Augmented Generation) function. The lightweight LLM can even be deployed directly with fast calculation speed, low energy consumption, and can be applied to various enterprise GPT solutions. By utilizing this solution, enterprises can train and conduct inference with their own data locally without having to go to the cloud to ensure data security and prevent data contamination. There are already use cases in education, manufacturing, medical, and legal industries.

On the AI trend, Albert Liu believes that the complete system should include CPU+GPU+NPU, including current AI PCs and AI smartphones. If the goal is to introduce GenAI applications in a low-power environment, NPU is the optimum solution.

In terms of cultivating AI talents in Taiwan, Albert Liu stated that the talent gap can be filled from 2 approaches. The first is education. Kneron has published several AI textbooks for various levels of readers, including an university of science and technology level, junior high school level, and even elementary school level. The second approach is by directly introducing AI technology to the industry. For example, Kneron has introduced AI into the backend and routing of IC design at very early stages, so they can mass-produce chips using the 12 nm process.

Watch the full Embracing The New AI Era - Looking At The Next Steps Of Taiwan's ICT Industry Conference here: (English Audio by interpreter)
>>>
https://www.youtube.com/watch?v=BfzpPEjmpaU

■《About COMPUTEX TAIPEI》

COMPUTEX TAIPEI was founded and named by the then Chairman of Taipei Computer Association (TCA), Stan Shih. In 1985, TCA invited TAITRA to be a co-organizer of COMPUTEX TAIPEI. In 2016, the startup focused event, InnoVEX was introduced.

COMPUTEX Official Website: https://www.computex.biz/
Event information and pre-registration website: https://show.computex.biz/
COMPUTEX CYBERWORLD website: https://show.computex.biz/online.aspx
Facebook: https://www.facebook.com/ComputexTaipei
YouTube Channel: https://www.youtube.com/user/COMPUTEXTAIPEIshow/
LinkedIn: https://www.linkedin.com/company/computex-taipei/

Back