Is all this AI Investment Headed for a Ho-Hum Ending?

Big Tech AI investment has been nothing short of colossal.1 Few companies can throw this kind of money at a still largely unprofitable venture, and while traditional AI investments were considerable, the capital expenditures on data centers and chips to train and deploy generative AI (GenAI) models are approaching historic market gambles.

As opposed to traditional AI, GenAI requires uncommon levels of computational power. While a traditional, predictive AI tool can analyze a painting and determine who the artist is, GenAI can produce a new painting in the same style as the artist. GenAI mimics human intelligence and creativity and as such large language GenAI models like ChatGPT are driving unprecedented interest.

The PricewaterhouseCoopers 27th Annual Global CEO Survey paints a bullish picture, with most US CEOs indicating that 2024 will be the year they finally reap investment returns on GenAI.2 The optimism is notable, but when Sequoia, the influential Silicon Valley venture capital firm, estimated the industry spent $50 billion in 2023 on Nvidia chips to train AI, yet only registered $3 billion in revenue, it’s easy to understand why murmurs of an AI bubble are brewing.3

CEOs are looking to grow their AI investments at least through 2025. AI efficiency gains are top of mind, as productivity can certainly be bolstered with smartly deployed GenAI tools.

At the same time, GenAI technology is pricey, so while efficiency gains are welcome, translating GenAI to additional dollars and cents is easier said than done.

In their most recent earnings reports, Microsoft and Google reported an increase in cloud services revenue, a significant portion of which comes from powering other companies' AI tools and infrastructure.4 This is positive news, but sustaining these revenue gains depends on companies continuing to spend billions of dollars to run these AI systems.

Roughly one-third of companies pay for at least one AI tool.5 While this is up from a year ago, the figure suggests a massive gulf still exists between employees who are playing with AI, to actually relying on it, and most importantly, paying for it. OpenAI, for example, is valued at nearly $90 billion. Appfigures, an analytics firm, noted that the firm’s most recent demo of its voice-powered features resulted in a 22% one-day subscription spike.6 What’s clear is OpenAI can generate attention, but with a $90 billion valuation and an estimated $2 billion in annual revenue, how many users are simply tinkering with something novel as opposed to widely adopting it?7

One prominent voice hitting pause on the AI investment euphoria is author and economist Daron Acemoglu. In a recent interview with Goldman Sachs, Acemoglu estimated that AI would ultimately grow US GDP by a paltry 0.9% over ten years and affect fewer than 5% of all human tasks.8 On the other hand, Goldman Sachs internal analysts point to a 6% GDP growth and the automation of a much more robust 25% of human tasks over the same period.

Estimates aside, Goldman analysts point to some interesting AI takeaways that are hard to ignore:

• The Internet’s early days featured solutions that were cheaper on Day 1 than the way individuals were doing business at the time.
• Smartphones, albeit more expensive than non-smart phones, replaced pricier products such as car GPS systems.

At the moment, AI can point to efficiency gains, but cannot solve the types of tasks that would be game-changing, like those of the Internet or a smartphone. Moreover, Nvidia absolutely dominates the AI hardware supply, so without additional market competition, running AIs will likely not get any cheaper in the short term.

Clouds are fluffy, light, and unintrusive. But when someone’s data is backed up to the “cloud,” it's neither a light nor fluffy destination. Data centers are massive facilities replete with a multitude of servers running 24/7. Pre-AI, data centers had been expanding at a steady clip. Bitcoin mining played a significant role, but GenAI is presenting an entirely new challenge.

A ChatGPT query requires roughly ten times as much electricity to process than a typical search engine search. Traditional data centers feature 5 to 10 kilowatts per rack of average density, while an AI center requires up to 60 kilowatts. Goldman Sachs estimates US utilities will require approximately $50 billion in new generation capacity investments to support just the data centers alone.9

While Big Tech AI investment capital is spurring data center growth, the US electric utility industry is highly regulated - code for slow and bureaucratic. Last year, the total capacity of power projects waiting to connect to the grid expanded by roughly 30%. Wait times vary from 40 to 70 months, and when power crunches occur, the utilities and the state will be forced to choose who receives power and who doesn’t.10

In Dublin, for example, a state-owned power operator paused new data center grid expansion until 2028, while Amsterdam instituted new rules that would levy fines on centers that don’t turn off idle servers to conserve additional energy.11

For the time, Big Tech AI investment is clearly benefiting Nvidia. Designing, training, and deploying capable AIs rely on graphic processing units (GPUs), and Nvidia controls approximately 92% of the market. Amazon, Microsoft, Google, and Oracle make up over 40% of the company’s revenue, but ironically enough, all are aiming to compete with Nvidia chips.12

While bubble murmurs could continue to grow, the GenAI arms race will undoubtedly continue. McKinsey Global Institute analyzed 63 use cases where productivity increases around customer support, the creation of creative collateral for marketing, advertising, and sales divisions, and drafting software code have the potential to augment the value of productivity by 15% to 40%.13

This is clearly where the bet lies, and the same report posits the biggest impact could be felt in the banking, high tech, and life sciences industries. For the skeptics, it’s important to remember that Uber eventually made money. The ramp-up, however, took 15 years. That’s a time horizon reserved nearly solely for the tech giants.

In many ways, today's AI boom mirrors the dot-com bubble of the late 1990s, a period marked by both incredible innovation and significant pitfalls. Just as the internet revolution reshaped entire industries and economies, paving the way for a digital transformation that affected everything from retail to communication, AI is poised to similarly transform every sector it touches—be it healthcare, finance, transportation, or entertainment. The potential for AI to enhance efficiency, drive productivity, and unlock new capabilities is immense. However, amid the excitement and rapid growth, it’s crucial to recognize that not every AI venture will succeed. The landscape is filled with startups and projects that, while promising, may not have the sustainable business models or technological foundations to endure in the long run.

During the dot-com era, skilled hedge fund managers played a pivotal role. They meticulously identified the winners—those companies with the right mix of innovation, market demand, and operational excellence—while expertly shorting the numerous companies that were destined to fail. This same level of expertise is invaluable in navigating the current AI landscape, where discerning wise investments from fleeting trends is not just beneficial but crucial. As we continue to witness the rapid proliferation of AI technologies and applications, investors need to be astute in recognizing which companies are genuinely pioneering advances versus those that are simply riding the wave of hype.

As we move forward, the role of hedge funds in identifying and investing in sustainable AI innovations cannot be overstated. These financial institutions bring a wealth of experience and analytical prowess, allowing them to evaluate the potential impact of various AI projects critically. By investing wisely in the AI market and leveraging seasoned professionals, investors can seize opportunities while effectively mitigating risks. The journey ahead will undoubtedly be filled with both challenges and rewards, and making informed decisions will be key to capitalizing on the transformative power of AI.

Sources:

1. De Vynck, Gerrit and Nix, Naomi. April 24, 2024. “Big Tech keeps spending billions on AI. There’s no end in sight.” The Washington Post.
2. PricewaterhouseCoopers. January 15, 2024. “PwC’s 27th Annual Global CEO Survey.”
3. Jin, Berber. March 30, 2024. “A Peter Thiel-Backed AI Startup, Cognition Labs, Seeks $2 Billion Valuation.” Wall Street Journal.
4. Mims, Christopher. May 31, 2024. “The AI Revolution Is Already Losing Steam.” Wall Street Journal.
5. Macomber, Ian. May 2, 2024. “AI spending grew 293% last year. Here’s how companies are using AI to stay ahead.” Ramp.
6. Appfigures. May 17, 2024. “GPT-4o Leads ChatGPT to Biggest Spike in App Revenue Ever!”
7. Mims, Christopher. May 31, 2024. “The AI Revolution Is Already Losing Steam.” Wall Street Journal.
8. Goldman Sachs – Global Macro Research. June 25, 2024. “Gen AI: Too Much Spend, Too Little Benefit?
9. Goldman Sachs. May 14, 2024. “AI is poised to drive 160% increase in data center power demand.”
10. Goldman Sachs – Global Macro Research. June 25, 2024. “Gen AI: Too Much Spend, Too Little Benefit?
11. Ibid.
12. Leswing, Kif. June 2, 2024. “Nvidia dominates the AI chip market, but there’s more competition than ever.” CNBC.
13. Chui, Michael and Yee, Lareina. July 7, 2023. “AI could increase corporate profits by $4.4 trillion a year, according to new research.” McKinsey Global Institute.