Goldman Sachs Deep Report: Generative AI - Hype or Real Transformation?

When it comes to the sector with the most rapid growth this year, AI must rank first.

Driven by the wave of AI technology investment, the market value of the "Big Seven" - Apple, Microsoft, Alphabet-C, Amazon, Meta, NVIDIA, and Tesla - has skyrocketed by 60% to a staggering $11 trillion.

And this "rocket-like" surge has also propelled the Nasdaq to a 34% increase this year, leading most of the gains in the S&P 500 index.

AI concepts have soared, and doubts have begun to arise. The loudest question is whether the hype around AI is excessive. Does generative AI truly have disruptive potential? Is it currently worth the enthusiasm of investors?

In Goldman Sachs' recent "Top of Mind" report, Goldman Sachs strategists Allison Nathan and Jenny Grimberg conducted in-depth conversations with several AI industry professionals, including Sarah Guo, founder of AI venture capital firm Conviction, Gary Marcus, professor at New York University and CEO/founder of startup Robust.AI, Kash Rangan and Eric Sheridan, software and internet analysts at Goldman Sachs, in an attempt to answer these questions.

In addition, they also explored the most attractive investment opportunities in the AI field at present, as well as the risks that investors should pay the most attention to.

The Revolutionary Changes of AI

The fundamental difference between generative AI and traditional AI lies in the former's ability to create content by understanding natural language, while the latter relies on programming languages. According to Kash Rangan, software analyst at Goldman Sachs, this is the key transformative feature of generative AI technology.

First, it can generate new content in the form of text, images, videos, audio, and code, while traditional AI systems are trained to predict human behavior and business outcomes.

Second, it allows humans to communicate and interact with computers in their own natural language, which has never been possible before; traditionally, computers use programming languages as prompts.

Guo further explains that in the era of Software 1.0, humans needed to write code to perform specific tasks, and in the era of Software 2.0, neural networks were trained by "painstakingly" collecting data. Now, humans have entered the era of Software 3.0:

Basic models can be used through open source or APIs, with natural language capabilities, reasoning abilities, and knowledge about the world.

In this mode, companies do not need to collect nearly as much training data, making the technology more useful, more accessible, and cheaper.

Since ChatGPT burst into popularity last year, many people have already experienced the power of generative AI technology. Analysts believe that generative AI could reshape the way society operates and add a new growth engine to the global economy.

Guo stated that the transformative potential of generative AI is already turning into reality. Any AI investment company can now invest in these models to enhance their business or undergo transformation.

Rangan estimates that in certain cases, developers have increased productivity by 15-20% by using generative AI tools.

With the widespread adoption of AI, Guo predicts that more fields, especially traditional service industries such as law, data analysis, image, speech, and video generation, will increasingly be served by AI.

Peter Callahan, a TMT industry analyst at Goldman Sachs, points out that retail investors believe generative AI technology possesses all the elements for platform transformation and has the potential to completely change the experience for businesses and consumers.

Furthermore, Joseph Briggs, a senior global economist at Goldman Sachs, suggests that this transformative potential could have profound effects on the macroeconomy.

He estimates that after the widespread adoption of generative AI technology in the United States and other developed economies, labor productivity growth could increase by approximately 1.5 percentage points per year over the next 10 years, ultimately leading to a 7% increase in global GDP.

Ryan Hammond and David Kostin, equity strategists at Goldman Sachs, believe that the U.S. stock market will also benefit from this trend, with a broader rebound expected in the medium to long term. They anticipate that the fair value of the S&P 500 index will be approximately 9% higher than it is now.

Artificial Intelligence is still far from intelligent, beware of excessive hype

In the long run, the transformative nature of AI technology is beyond doubt, but given the current stage of its development, is the market hype surrounding it excessive?

Marcus's answer is "yes" because "current artificial intelligence is still far from intelligent."

He points out that the functionality of current AI neural networks is completely different from that of the human brain.

Although AI can perform "reflexive" statistical analysis, it lacks mature reasoning capabilities. These machines can learn, but to a large extent, they revolve around statistical analysis of words and correct responses to prompts, rather than abstract concepts. Moreover, they do not possess an "internal model" like humans do to understand the world around them. **

Marcus issued a warning to investors:

Be cautious about AI, it may not be as magical as many people imagine.

I wouldn't say it's too early to invest in AI now; some companies with smart founding teams and a good understanding of product market fit may succeed, but there will also be many losers.

Marcus said that Artificial General Intelligence (AGI) may eventually be achieved, but humans are still far from this goal, and no investment can change that.

In addition, investors can learn from history.

Goldman Sachs market strategists Dominic Wilson and Vickie Chang mentioned that during past periods of innovation-driven productivity booms, such as electricity (1919-1929) and the proliferation of personal computers and the internet (1996-2005), stock prices and valuations soared and eventually led to a burst bubble.

Guo believes that even today, there are pricing errors in certain areas of the private market. Although investors have a deeper understanding of these areas, they still generally adopt the same investment approach.

She warned that misjudging the timing of change is a common trap in investing. As an early-stage investor, she is less concerned about valuations and instead chooses markets, products, and companies that she believes are meaningful.

Eric Sherida, an internet analyst at Goldman Sachs, has a slightly different view.

He believes that the trading prices of most outstanding AI concept stocks are still reasonable relative to Generally Accepted Accounting Principles (GAAP) earnings per share multiples.

Rangan also believes that AI may not be in a speculative cycle because this wave is led by tech giants rather than startups:

This technology cycle is not dominated by newcomers (AI), so it is unlikely to end with a whimper or take a long time to get started.

The transition from mainframes to distributed systems in the early 1990s and from distributed systems to cloud computing in the early 21st century took longer than many people expected because large established companies were the key voices of opposition.

As Rangan said, OpenAI, the company behind ChatGPT, is supported by Microsoft, Alphabet-C has launched Bard, and investments have been made in AI startups like Anthropic. Meta has launched LLaMA, and domestic giants like Baidu and Alibaba have also released their own models. The global AI competition is in full swing.

"Picks and Shovels"

Amidst the skepticism, what are the most promising investment opportunities in AI?

According to Rangan and Sheridan, the opportunities lie not only in large tech companies developing foundational AI models, but also in "picks and shovels" enterprises.

"Pickaxes and Shovels" is one of the investment strategies favored by the legendary figure in the investment industry, Peter Lynch, which involves investing in companies that indirectly benefit from a particular trend.

Rangan and Sheridan believe that in the current AI boom, companies serving the field, such as semiconductor companies, cloud computing giants, and infrastructure companies, can gain advantageous positions in the current "building" phase.

Guo shares a similar view, but also sees opportunities throughout the entire stack, and is particularly excited about the application layer.

Many investors are uncertain about this layer and believe that all the value lies in model training itself. However, a great deal of creativity and work is required to make non-deterministic models work in production use cases. Currently, both startups and existing app companies will leverage these capabilities in many areas... and we are very excited.