[Tech Talk] Tens of Trillions Invested in AI... Will the Day of Returns Come?

Big Tech Spending Tens of Trillions on AI
But AI Product Performance Remains Lackluster
Risk of a 'Third Winter' Looms

So far, global big tech companies have invested tens of trillions of won in artificial intelligence (AI) development. Most of this spending has gone toward securing specialized personnel for new AI model development, purchasing graphics processing units (GPUs), and constructing data centers.

However, the industry is gradually being troubled by a 'fundamental question.' Can the tens of trillions of won poured in so far generate profits exceeding that amount? Especially given that the previously disclosed AI business performance has not been very high, concerns are growing.

The current AI boom where only Nvidia is making money

NVIDIA logo. [Image source=AFP Yonhap News]

American big tech companies such as Microsoft, Google, and Meta, as well as Chinese IT companies like Tencent, are all scrambling to secure GPUs. Due to new data centers, power demand is skyrocketing, and some in the IT industry even suggest that AI-related investments could approach $1 trillion (about 1380 trillion won) within a decade.

The heat of the AI investment boom can be felt most directly among semiconductor manufacturers that provide chips for actual AI training and inference tasks. Nvidia, a representative AI beneficiary, is showing nearly double growth every quarter, and the performance of Samsung Electronics and SK Hynix, which supply high-speed memory chips like HBM, is also improving.

However, the bigger the spending, the greater the risk. If $1 trillion is poured into a specific technology, the products and services created by that technology must return value exceeding $1 trillion. Otherwise, a harsh correction will come. Not only companies riding the AI boom but also investors and consumers could suffer.

Planning to spend $1 trillion... what if generative AI is useless?

NVIDIA's next-generation chip architecture 'Blackwell'.
[Image source=NVIDIA]

At the end of last month, investment bank Goldman Sachs released a new report encompassing these concerns. The core of the report is that the return on AI investment could be much lower than previously expected.

Goldman Sachs is particularly worried about 'generative AI,' the main driver of the latest AI boom. When generative AI products like ChatGPT first appeared, many predicted innovation in services, information search, and other fields, but perhaps the benefits of generative AI have been overestimated.

Goldman Sachs cites Google's case as evidence. Google attempted to integrate generative AI into its search service, but during testing, the AI model began to produce unintended results due to 'hallucinations,' leading Google to ultimately abandon the effort. Attempts to introduce chatbots trained with specialized knowledge into various professional industries such as law and medicine have been ongoing, but no product has made a significant impact.

Questions about generative AI being overhyped are also raised in academia. Ted Chang, a renowned American science fiction writer and computer engineering expert, attended the 'People and Digital Forum' held in Korea last month and bluntly stated, "AI cannot be considered to possess human intelligence."

There are even concerns that premature AI adoption could weaken corporate competitiveness. Ted Chang explained that the current AI boom is "fundamentally about executives choosing AI as a way to reduce labor costs," and criticized that "even if service quality and company reputation deteriorate, only investors are cared about."

Having endured two winters already, will this time be different?

Marvin Minsky, Professor Emeritus at MIT, during his lifetime. Although he was the one who revealed the limitations of the decision-making automation machine 'Perceptron,' he was also a scholar who laid the foundation for future artificial intelligence research.
[Image source=MIT]

Above all, this is not the first time AI has collapsed under excessive expectations. Humanity has already experienced two so-called 'AI winters' from the mid-20th century to the present. Excessive investment flowed into immature computer technologies, but when they failed to generate proper returns, funding rapidly dried up, forcing a cold period of survival.

The first AI winter occurred in the 1950s, when some scientists believed that replicating human brain neurons could mimic human intelligence. However, after initial funding, research showed no progress for a decade, and brain-related research was frozen for decades thereafter.

The second AI winter appeared in the 1980s. At that time, the 'perceptron' project aimed to reproduce human decision-making with machines using advanced computer equipment. Although astronomical public and private funds were invested, in 1984, Marvin Minsky, a professor at MIT, proved fundamental flaws in the perceptron neural network, triggering another long stagnation.

In markets, the process of capital excessively flowing in and then retreating is called a 'boom-bust cycle.' The more irrationally overheated the capital markets are during the boom, the deeper and sharper the recession during the bust.

Even if currently useless, many innovative companies researching products that contribute to technological and scientific progress can be swept away during the bust. This is why the current AI boom must not become the 'third winter.'