2024-07-31
Taiwanese companies are currently profiting from the first wave of the AI revolution, but where will the next source of growth come from? Even Nvidia CEO Jensen Huang is exhorting the virtues of AI foundries—a system under which the creation of AI models is outsourced to contractors. This may give Taiwan’s AI supply chain a substantial competitive edge. In an exclusive interview with CommonWealth Magazine, Dr. Liang-Gee Chen (陳良基), Professor Emeritus of the Department of Electrical Engineering at National Taiwan University, posits that if Taiwan can leverage its advantages in hardware and software integration, it can become the island kingdom of the world’s most trusted AI foundries.
Artificial intelligence (AI) is the latest iteration of the industrial revolution. At this moment, tech giants like Microsoft, Meta, and Google are competing over what is the most profitable way to work with hardware suppliers. In the semiconductor industry, on one end of the spectrum are “integrated device manufacturers”, or “IDMs”, represented by Intel. On the other end are companies like TSMC, which exemplify a more open approach to operating wafer foundries, one that revolves around specialization and division of labor.
With regard to the tech industry, Taiwan’s decision to divide the work among different companies on the supply chain birthed an incredibly strong semiconductor sector. It also nurtured our hardware integration capabilities, which is why so many functions are now integrated into the microchips we make. In the age of AI, when everyone is using AI to do more than was previously possible, the question we should ask is: Who will be the supplier of AI? Current AI suppliers that appear to be far ahead of the competition are like chip suppliers in the early days of the semiconductor boom, such as Intel 40 years ago.
What really infuses AI with new abilities is the data on which it is trained. If all our data is in the hands of the aforementioned tech leaders, like Microsoft, Google, and Amazon Web Services (AWS), then they themselves will become the new generation of IDMs, as users will rely on the AI models that they provide. Two years ago, I was already promoting the concept of the “AI Foundry”. I use the word “foundry” to convey the meaning that, no matter where the user is located, they can trust the AI that Taiwan provides. This is because in the age of AI, people are not as worried about computing power as they are about data. If we’ve already done the computing for you and given you a fine-tuned AI model, computing power becomes a less critical part of the equation.
If this is where the AI industry is headed, then whoever becomes a trusted contractor that can take on outsourced AI workloads will become the most important supplier in the entire AI industry.
To use the semiconductor industry as a familiar example, microchips make up the core of servers, while cloud service providers (CSPs) and computing providers form the second link in the supply chain. The final link comes from service providers who create AI models, such as OpenAI.
Taiwan’s chance lies in this final link of AI model services. Taiwan can offer the world an open environment where users of AI models don’t need to surrender their data to a monolithic IDM. Instead, they can outsource the training and fine-tuning of their AI models, just as they did with microchips. It goes without saying that the contractor must be able to be trusted not to leak precious and sensitive data.
Taiwanese companies have made a name for themselves on the world stage because they are unrivaled when it comes to OEM and ODM—outsourced contracting work. The world trusts that the job will get done if it is in the hands of a Taiwanese company. Therefore, it won’t take long for international clients to get used to a new model in which data is delivered to Taiwanese vendors for AI development. The clients control their own data and can make adjustments as they see fit. They are not restricted by the rules of existing IDMs, i.e. “If GPT4.0 is what’s available, that’s what you're using; if there’s an update, then you’d better get on-board.”
Nowadays, if a company wishes to use AI, it must hand over its data to OpenAI for training. In the end, OpenAI controls all the information. But is OpenAI’s business model trustworthy?
TSMC is trusted because it doesn’t compete with its customers. Will OpenAI compete with its customers? Right now, there’s no way to know. How OpenAI fine-tunes its data is also a huge unknown. The whole thing is very different from the semiconductor industry.
With semiconductors, when you outsource to TSMC, not only are you certain that it will not compete with you, you can also tell TSMC what design you want as well as what you want to adjust. Everything that is discussed is confidential and will not be released to the public. But when you give your data to OpenAI, you are in fact helping them to train their AI models. OpenAI owns all of your data. For this reason, many big companies prohibit their employees from using AI services from OpenAI.
Hence, the common pain point of enterprises is this: Is there an unrestricted AI development tool that can be controlled by the user? This is a chance for Taiwan to demonstrate its strengths. We already have the hardware, and the quality of our work is universally trusted. We can leverage that trust to invite enterprises to let us work with their data and develop the perfect AI model for them to use.
In a future where every profession is using AI, it is imperative for AI suppliers to allow their users to exert tighter control over their data and enjoy absolute peace of mind. OpenAI is a closed system; users may only need a smaller model to get the job done. For example, a user who requires art design services does not need the AI to also know how to write poems. If we can get a clearer picture ahead of time of what sort of know-how and knowledge are required, we will be able to tell when a smaller AI model will suffice, which cuts down on the amount of computing resources that are needed. In many ways, Apple is already headed in this direction. Taiwan controls the microchips at the heart of AI hardware. Therefore, if we know what customers are looking for, we can simplify the workload for them, to the point that we may even be able to recommend more cost-effective hardware configurations.
In summary, the “AI foundry” service needs both hardware and software integration. In all the world, no one does this better than Taiwan.
A time will come when every industry needs more independent AI models. The IDMs of the current AI wave will face many challenges in the future, just as Intel had to learn how to outsource its manufacturing. This is because the current AI models are trained in a grab bag, one-size-fits-all fashion. It will not be easy for AI developers who built their success on this foundation to adapt to training smaller AI models.
In the end, it all comes down to who controls the data, because that's where the money is.
For example, TSMC has remarkable production prowess mainly because it adopted smart manufacturing early on. Smart manufacturing has already become mainstream; there are probably over 30 fully automated factories in Taiwan. This was made possible by collecting data and then using it to manage the factories. All the manufacturing know-how is in the data. That is why it is so important to control the data: It is a competitive advantage.
Taiwan can help train the world’s open-source AI models. That is why we are currently working with Lama3. When it comes to training AI models, Taiwan can work with anyone on Earth who is willing to provide open-source AI models. We can elevate this market; with Taiwan’s help, this can become our edge when competing with the IDMs of the AI world.
The current IDMs have everything going for them, and if they invest in things that solve their customers’ pain points, they may continue to dominate the market. If Taiwan can put its full strength behind the plan outlined above, just as we did for the semiconductor industry, and if our policymakers can take into account Taiwan’s role in the supply chain of the world’s AI foundries—then Taiwan will have a greater chance to thrive in the age of AI.
Read the full article in CommonWealth Magazine.
Source： CommonWealth Magazine web-only article, 2024-06-04
About Us
InvesTaiwan InvesTaiwan Service Center Department of Investment Promotion, MOEA Department of Investment Review, MOEA Contact TaiwanWhy Taiwan
Reasons to Invest Overview Key Industries for Investment Promotion Incentives Important Policies Success Stories
MOEA
MOEA DOIP
MOEA YouTube
Let's Juiker Together
