Is AI Bubble Too Early? Taiwan Stocks More Stable Than US? Cheng Shih-Chia Reveals Global Computing Power Gap and Taiwan's Supply Chain Advantage

Cheng Shih-Chia, co-founder of cross-border AI data technology company iKala, believes it's too early to worry about an AI bubble. He points out that global AI computing power demand over the next three years is expected to be 24 times current supply. "Can the world have 24 TSMCs to meet this demand? Obviously not," he said.

Looking at major technological breakthroughs over the past 20 years, such as the internet, smartphones, and social media, each market cycle lasted about 10-12 years. This wave of AI, starting with ChatGPT in 2022, is expected to continue conservatively until 2032-2033.

AMD CEO Lisa Su previously said that the current AI wave is like the third inning of a baseball game. Therefore, talking about an AI bubble is indeed premature.

On May 22, 2026, Lisa Su said at an AI forum that the growth rate of the AI infrastructure market is astonishing, and the data center market size alone is expected to reach $1 trillion in the next 3-4 years.

Cheng Shih-Chia: Computing power shortage cannot be solved in the next 2-3 years, most Taiwanese manufacturers' capacity will be full until 2028

Cheng Shih-Chia published his second book on AI, "New Business Opportunities in the Era of Smart Inflation," in the past two years. The 2024 book discusses trends, while this book looks at AI from the perspective of enterprise demand, such as the shift from traffic thinking to influence thinking, from click economy to transaction economy, from multi-person enterprises to one-person companies, from division of labor to human-machine collaboration, and from execution competition to judgment competition.

He also believes that AI is rapidly devaluing intelligence, and enterprises must first add AI and then add people. Therefore, smart inflation not only accelerates the elimination of those who do not use AI, but also redefines what a "valuable person" looks like.

Recently, in a small media interview, when asked how long Taiwan's semiconductor supply chain will benefit from the AI demand explosion, Cheng Shih-Chia believes that there is at least 5 years to go. "I think at least 5 years or more. First, looking at the computing power shortage of AI data centers, the computing power shortage cannot be solved in the next 2-3 years, because most Taiwanese manufacturers' capacity will be full until 2028. You use computing power to backtrack, and we still need 24 times the computing power in the next three years, which means that the computing power we have already produced is still 24 times short."

"From the perspective of user demand, it still needs to increase 24 times the computing power. Can the world have 24 TSMCs to do these things? Obviously not. Even if TSMC expands its capacity, its annual wafer capacity is about 60,000-80,000. You need to expand to 600,000-800,000, which is impossible."

"So you know that Taiwanese manufacturers are now standing very firmly in the supply chain, and there will be at least 5 years of good times. That's why you see that Taiwanese stocks keep falling and then rising back, and everyone keeps going up and down."

Cheng Shih-Chia pointed out that TSMC's annual wafer capacity is about 60,000-80,000, but the global demand in the next three years is 24 times, far beyond what TSMC can meet.

Cheng Shih-Chia: A situation is gradually happening, that is, "Taiwan stocks are actually more stable than US stocks"

Regarding whether the AI industry has a bubble, he believes that downstream application companies are more likely to have a bubble, while upstream supply chain opportunities are smaller. "Downstream companies will have a bubble because they are targeting consumers, but consumers are now very mobile. You look at OpenAI, Anthropic, Google, they are already in vicious competition in AI. OpenAI and Anthropic are heavily subsidizing consumers. OpenAI lost nearly $400 billion last year, which is more than 1 trillion NT dollars. How can this be a (viable) business model?"

"You will find that (market) demand exists, but the loyalty of these demands is very low, and the loyalty of downstream is very low. Today, if you say ChatGPT has become stupid, I will immediately switch to Claude or Gemini. It's just the difference between canceling a credit card and adding a credit card."

Cheng Shih-Chia believes that the most special thing about this round of AI is that no matter which AI market consumer chooses, NVIDIA will eventually benefit, because all AI giants need to buy NVIDIA's AI computing chips. "So you will see a situation gradually happening, that is, Taiwan stocks are actually more stable than US stocks. In the past, everyone looked at US stocks, but now you find that Taiwan stocks don't fall much. Why? Because the US stock market is killing red-eyed, but no matter how the capital flows, it will eventually flow to the computing power manufacturers."

AI bubble observation indicators: enterprise adoption rate, not consumer market

Cheng Shih-Chia also explained that when observing this wave of AI, not only should we refer to the experience of several previous major technological trends, but we should not ignore the impact of enterprise demand relative to consumer demand on the AI industry. "If you look at the entire technology cycle in 10 years, that is, ChatGPT came out in 2022, and the development period of about 10 years is in 2032 and 2033, at that time everyone will be very familiar with what AI is. Technology has a cycle of about 10-12 years, and new things will come out, such as the internet, smartphones, social networks, and then AI, each of which is about 10-12 years."

"This round is probably in the third year, because AMD's CEO Lisa Su said that she thinks it is now in the third inning, that is, this decade is only in the third year."

"So why won't AI bubble? You are not looking at the short-term explosion of the consumer market like VR and AR, which are unreliable. The key is whether enterprises will adopt it."

He explained that when Gartner and other well-known industry think tanks analyze technology cycles, they will use whether enterprises will continue to increase the adoption rate as an indicator of whether a technology has entered a super cycle. From the current situation of AI, there is almost no doubt, but everyone is not yet sure how to introduce AI, but they all know that AI will eventually be a standard configuration in the enterprise. Even if enterprises do not regard AI as a standard configuration, employees will use their own cards to use Gemini or ChatGPT. "So, the enterprise adoption rate is a major indicator of whether a technology will bubble, and now it has been proven that the enterprise adoption rate will definitely increase with time, even if the speed is slow, because the enterprise adoption must be slow. As long as the enterprise starts to adopt, this thing will definitely not bubble."

More in-depth reports by Wind Media: · Rushing for 5% of the land in exchange for 140,000 casualties! Ukrainian lawmaker visits Taiwan: Ukraine's annual war expenses are 4 times Taiwan's defense budget, the biggest lesson is "this thing" · Exclusive interview: Chip frenzy has entered the "late Soviet period"? "Chip War" author Miller warns: China's two major economic contradictions, Taiwan's "three moves" to deter aggression · In-depth: Chinese "civilian ships become warships" targeting the Taiwan Strait? French military expert: Roll-on/roll-off ships, cargo ships, and unmanned ships hide which anti-Taiwan scripts