Goldman Sachs: Less than 5% of companies use AI in the first year after the "AI revolution", but early adopters see significant efficiency gains

Author: Ge Jiaming

Source: Hard AI

From the emergence of GPT-4 to the launch of Wenxin Yiyuan, to the appearance of Sora, and even the recent popularity of domestic Kimi, whenever a phenomenal AI product appears, many people will marvel at the "iPhone moment" that AI is about to usher in.

After a year and a half of catalysis, AI, dubbed the "Fourth Industrial Revolution," how much impact does it really have on ordinary people?

On April 2, a team led by Goldman Sachs economist Jan Hatzius released a report titled "One Year into AI Transformation: Progress Smooth but Impact Still Years Away." The report stated that the actual application of AI is still limited, with less than 5% of enterprises using generative AI in production. The current level of application is not sufficient to significantly boost productivity.

Goldman Sachs pointed out that the low application rate of AI limits its impact on the labor market. Preliminary data shows that the development of AI has slightly increased labor market demand, with negligible effects on the unemployment rate, thus exerting a mild positive push on the labor market.

For early AI users, Goldman Sachs stated that data shows AI can significantly improve productivity. Both academic research and business cases demonstrate that labor productivity on average increases by 25% (with a median of 16%) after adopting AI.

The Massive Impact of AI Will Take Time to Materialize

Goldman Sachs noted that over the past year, they have argued through a series of articles that generative AI can enhance labor productivity and drive global growth. However, the emergence of this growth largely depends on the widespread application of AI technology and AI-related investments, both of which may take several years to fully realize:

We have previously pointed out that the AI wave will be a significant driver of global productivity, with global productivity expected to increase by over 1.5% annually over the next 10 years, driving $7 trillion in economic growth.

However, achieving a significant boost to macroeconomic growth largely depends on the extensive application of technology and the scale of investments related to AI. As we have previously mentioned, both of these may take several years to fully materialize.

Goldman Sachs also highlighted that the financial markets view AI as a milestone technology, initially driving investments and subsequently leading to productivity improvements:

Looking at the upward trend in the past 6 months, the gains in the US stock market have mainly focused on AI hardware suppliers (such as semiconductor companies) and software suppliers (such as cloud service providers), benefiting from the early growth in AI-related investments, while potential beneficiaries of AI productivity have only seen more moderate gains.

As emphasized by our US investment portfolio strategy team, these gains (especially for hardware companies) mainly reflect improvements in fundamentals driven by the demand for GPUs and data centers, rather than valuation increases driven by investors' improved outlook on company prospects Goldman Sachs believes that technology companies are among the first to benefit from AI applications. However, based on the mention of AI in various companies' financial conference calls, an increasing number of industries have incorporated AI into their future strategies, including IT and communications, industrial, consumer, financial, and medical fields. This indicates that the potential efficiency improvements brought by AI applications are receiving widespread attention:

While current AI applications are mainly concentrated in the technology industry, the growing interest from other industries and the rapid increase in AI-related investments suggest that the widespread application of AI and the resulting productivity improvements may gradually be realized in the coming years.

By 2025, new investments in AI hardware could reach $250 billion, equivalent to 9% of U.S. corporate investments or 1% of GDP. This is consistent with previous forecasts. Within the AI hardware sector, the main focus is on semiconductors (with Nvidia accounting for over 75%) and the increasing demand for cloud services, servers, and network equipment.

Goldman Sachs believes that the stock market is providing an optimistic signal, indicating that the AI investment cycle may have already begun. However, in the official national accounts data (the main basis for GDP accounting), there has not been a significant increase in AI-related investments. This suggests that factors other than AI, including the demand for non-AI technologies and cyclical factors, may currently be playing a more important role in driving overall capital expenditure.

AI is not yet widely adopted

According to data released by the well-known web market analysis platform similarweb, in early 2023, ChatGPT's global traffic exceeded 1 billion visits. By February of this year, ChatGPT's global traffic reached 1.6 billion visits. Goldman Sachs believes that this data indicates that over a quarter of Americans use artificial intelligence tools informally at least once a week.

Goldman Sachs points out that despite the surge in AI application traffic, the formal adoption of AI by enterprises remains low, except for a few specific high-tech industries. According to the recent "Business Trends and Outlook Survey" AI supplement report released by the Census Bureau, less than 5% of companies formally use AI-generated technology to launch products and provide services, while this percentage reaches 10-15% in information, professional services, and financial companies:

There is a greater disparity in AI usage rates in more specific sub-industries. In the technology industry, over 20% of companies use AI-generated tools in production, while in other digital fields such as film and sound production, the current and expected AI usage rates are higher.

Goldman Sachs points out that in the next six months, most industries are expected to accelerate the adoption of AI, with many companies investing in AI applications to be used in the coming years:

An IT spending survey shows that although only 12% of CIOs currently plan to allocate more than 5% of their IT budget to generative AI, this proportion is expected to exceed half in the next three years, and the share of AI in IT budgets will also double.

Significant Efficiency Improvement for Early AI Users

Goldman Sachs points out that although the adoption of AI is still in its early stages and there is still a high degree of uncertainty about the ultimate impact of widespread AI use, evidence from early users suggests that AI may bring significant efficiency improvements:

We evaluate the impact of AI on productivity from academic research and company reports, and several points should be noted.

Firstly, selection bias should be noted. Early adopters are usually the group that can benefit the most from new technologies, so their effects may be higher than those of general users. Currently, most cases are concentrated in professions with high task repeatability suitable for AI automation, which may lead to overly optimistic estimates.

Secondly, publication bias exists, as the academic and corporate sectors may be more inclined to publish and report positive effects while ignoring insignificant or negative results, which may also lead to overestimation. In addition, there is a possibility of underestimation, as current cases only consider the automation of certain tasks, and if AI applications expand to more tasks, the potential for improvement may be even greater.

Goldman Sachs estimates that AI can increase labor productivity by 9-56%, with a median of 16% and an average of 25% according to academic research; the median expected productivity improvement in corporate reports is 25%, with an average of 26%. Some studies have found that AI not only increases output but also improves work quality.

At the same time, Goldman Sachs states that the productivity improvement effect of AI on new employees is more significant than on experienced staff, as AI may be more helpful in accelerating learning and improving the efficiency of less experienced employees, while its assistance to highly skilled employees is relatively limited