When OpenAI launched the generative AI chatbot ChatGPT for public use on Nov. 30, the S&P 500 was worth $5 trillion less than now, tech spending was deep in a post-pandemic hangover, and the economy appeared headed for recession or persistent high inflation.
That single day provided just an inkling of generative AI's potential for transformative impact. The S&P 500 shot up more than 3% as tech stocks with artificial intelligence products rumbled. OpenAI investor Microsoft leapt 6%, and AI chipmaker Nvidia climbed 8%. Google parent Alphabet also jumped 6% that day, and Meta Platforms ran up nearly 8%.
Now the tech hangover is giving way to a new "gold rush," Wedbush Securities analyst Dan Ives argues. Ives thinks ChatGPT opened the door to another $1 trillion in artificial intelligence-related spending over the coming decade that wasn't on Wall Street's radar.
And the economic impact could prove even more far-reaching. Some analysts are even talking about a new Roaring '20s fueled by AI. Experts say generative AI could launch a surge in productivity after a 17-year slump. A productivity pickup couldn't come at a better time, as a worker shortage, an aging population and deglobalization fan inflationary pressures.
"We are in desperate need of a new source of growth," Deutsche Bank economists wrote in a June 14 report. "Despite near-term pessimism, we remain enthusiastic about AI's potential to change the nature of our economies," they wrote, calling it "an immense source of optimism" as the decade progresses.
The consensus is that generative AI will change the world. But productivity growth has sputtered through recent waves of new technology, including the Apple iPhone, cloud computing and robotics. Why should this time be different? And what might stand in the way of a productivity boom and S&P 500 bull run?
Generative AI Productivity Boom
Generative AI could boost U.S. labor productivity growth by 1.5 percentage points per year for a decade, Goldman Sachs economists predict. That would double the lackluster 1.4% rate since 2005. Goldman says S&P 500 profit margins could widen by four percentage points, which would be a huge increase from the 12% average over the past few years. Global GDP could grow by an extra 7 percentage point, or $7 trillion, over the same period.
Solita Marcelli, chief investment officer for UBS Global Wealth Management in the Americas, says increased investment spurred by generative AI strengthens the case for a secular bull market that's "comparable to the Roaring 1920s."
That could mean "sustained growth above 2%, falling inflation, strong employment gains, rapidly rising productivity and revolutionary technological change."
The lightning-fast adoption of ChatGPT, which gained its first 100 million users in a record-shattering two months, partly explains the conviction that the U.S. is on the cusp of a productivity boom.
Goldman economists figure that technological advances begin to yield broad productivity gains in the 10 years after half of businesses adopt a technology.
If earnings calls are an indication, that threshold could arrive quickly. Just 53 S&P 500 firms mentioned artificial intelligence in Q3 calls, before ChatGPT's release. That jumped to 110 by Q1, according to FactSet.
That huge interest and rapid takeup reflect not just generative AI's potential but the relatively low barriers to adoption. Unlike the branch of artificial intelligence known as machine learning, generative AI responds to natural language commands.
Artificial Intelligence News And AI Stocks To Watch
Generative AI Vs. Machine Learning
For routine, repetitive tasks, such as some manufacturing work, computers can be programmed with explicit instructions. Advances in machine learning in the last decade have enabled automation of complex tasks. That means programming a computer to recognize patterns in specialized data sets in order to make accurate predictions. Among the early successes were computers reading CT scans and producing diagnoses. Self-driving cars, which use data from sensors and cameras to predict the next movement of other vehicles and pedestrians, are an example of machine learning AI with a more mixed record of success so far.
Generative AI is seen as potentially world-changing because it aims to automate something thought to be beyond reach: creative content.
Instead of training with a focus on specialized data, GPT-3.5, the brains behind the ChatGPT chatbot, soaked up a big chunk of the books, articles and websites on the internet to create what is known as a large language model. On top of that, it is trained to try and fool humans from thinking its output is computer-generated.
Some people have described ChatGPT as an advanced autocomplete. Instead of just suggesting the next word, it can offer up a cover letter, essay, story, marketing pitch, travel itinerary or legal brief.
Goldman Sachs economists estimate that 44% of legal work could be automated via generative AI.
But generative AI isn't limited to responding in natural language. It can also generate images, computer code and complex designs.
How Will AI Change Knowledge Work?
Microsoft's GitHub Copilot leverages OpenAI's latest GPT-4 large language model to turn software developers' prompts into snippets of code, speeding up their work by 55%, according to an internal study published last year.
A 2022 paper from the NASA Goddard Space Flight Center hailed AI-enabled generative design as "a paradigm shift in the design process." Highlighting OpenAI's DALL-E foundation model that turns natural language prompts into images, the paper explained how engineers working with Autodesk Fusion 360 software designed optical instruments in a tenth of the typical time, while improving the instruments' performance.
Engineering, architecture and even drug discovery are all fields where generative AI is already making an imprint.
There's a common thread with most of these new applications, according to a McKinsey & Co. analysis.
"Generative AI's impact is likely to most transform the work of higher-wage knowledge workers because of advances in the technical automation potential of their activities, which were previously considered to be relatively immune from automation," McKinsey found.
Goldman Sachs economists Joseph Briggs and Devesh Kodnani conclude that "generative AI could substitute up to one-fourth of current work."
A study by OpenAI arrives at a similar place, finding that 49% of today's jobs could have half or more of their tasks accomplished with generative AI. The authors note, however, that complementary software, such as Casetext for lawyers, and human overseers are needed to catch factual mistakes and avert other issues.
What Jobs Could AI Replace?
| Industry | Share of jobs |
|---|---|
| Office and administrative support | 46% |
| Legal | 44 |
| Architecture and engineering | 37 |
| Life, physical and social science | 36 |
| Business and financial operations | 35 |
| Community and social service | 33 |
| Management | 32 |
| Sales | 31 |
| Computer and mathematical | 29 |
| Farming, fishing and forestry | 28 |
| Protective service | 28 |
| Health care practitioners and technical | 28 |
| Educational | 27 |
| Health care support | 26 |
| Arts, design, entertainment, sports, media | 26 |
| All industries | 25 |
| Personal care and service | 19 |
| Food preparation and serving | 12 |
| Transportation and material moving | 11 |
| Production | 9 |
| Construction and extraction | 6 |
| Installation, maintenance and repair | 4 |
| Building and grounds cleaning/maintenance | 1 |
Source: Goldman Sachs
Still, such estimates may understate the coming impact on work. That's because they omit ongoing efforts to automate hands-on work in industries such as manufacturing, logistics and transportation via nongenerative AI, or machine learning, and robotics.
Long before ChatGPT made its stunning entrance, McKinsey researchers in 2019 had envisioned that 50% of today's work activities could be automated by 2055. Now that generative AI has opened a door to automating knowledge work, McKinsey has moved up that timeline by a decade to 2045.
Such predictions come with a wide margin of error, partly because earlier productivity booms were slow to materialize. The advent of personal computers in the late 1970s, followed by spreadsheets in the early '80s, seemed to be a productivity miracle. But statistics didn't begin to back that up until the mid-1990s.
"It was not until the world's computers were integrated into what eventually became the Internet that the true potential of computers was realized," wrote Peter Berezin, chief global strategist at BCA Research.
The release of the Netscape Navigator browser in 1994 unleashed the potential of the internet and opened the door to new business models, Marcelli of UBS wrote.
She sees ChatGPT playing a similar role, "bringing AI to the masses in an accessible and intuitive way."
Berezin sees reason to think predictions about the impact of generative AI may actually prove too timid.
Join IBD experts as they analyze actionable stocks on IBD Live
Generative AI Reflects 'Exponential' Progression
"They extrapolate linearly from what AI can do today to what it can do tomorrow," Berezin said. "But AI's progression is following an exponential curve, not a linear one."
OpenAI says that its latest GPT-4 model, released in March on a subscription basis, scored in the top 10% of test takers in a simulated bar exam. The GPT-3.5 model that's behind ChatGPT scored in the bottom 10%.
GPT-4's training covered 1 petabyte of material, or more than 20 times the material mastered by GPT-3.5.
Token limits are another measure of generative AI's usefulness. ChatGPT has a limit of 4,096 tokens. That means it can process queries and generate responses with only that many tokens. A typical token may be a letter, word, number, punctuation mark or other symbol. Once a conversation reaches that threshold, ChatGPT resets, which is why it's said to have no short-term memory.
OpenAI's GPT-4 model was introduced with more than 8,000 tokens, allowing deeper analysis. In May, AI startup Anthropic introduced the latest version of its Claude generative AI model with a 100,000-token limit. That equates to roughly 75,000 words, enabling the analysis of SEC 10-K filings, for example.
Massive Demand For AI Chips
The generative AI model behind ChatGPT, already left in the dust by newer models, required 10,000 Nvidia graphics chips to train ahead of its deployment, according to UBS analyst Timothy Arcuri.
Yet the far-bigger expense, known as inference costs, comes from tapping into that artificial brain power with user queries. "The costs to inference ChatGPT exceed the training costs on a weekly basis," according to the SemiAnalysis research firm.
Alphabet chairman John Hennessy has said that queries using Google's Bard AI chatbot cost the company 10 times as much as regular Google search. SemiAnalysis' work suggests that's an understatement.
The cost of using generative AI means that companies must expect a payoff, such as more revenue, labor and other expense savings, or both.
Wall Street is betting that those payoffs are coming. Morgan Stanley analysts see the market for AI semiconductors tripling from $43 billion to $125 billion over the next three years. The broader AI technology market should grow from $90 billion today to $275 billion by 2027.
In late May, Nvidia forecast $11 billion in revenue for the July-ended second quarter, crushing estimates for $7.2 billion, largely on AI chip demand.
Microsoft's AI Ramp
On July 18, Microsoft announced it will charge $30 per month for Microsoft 365 users to upgrade to Microsoft 365 Copilot. The generative AI version of the business software suite allows users to order up article outlines or drafts in Word, emails and invites in Outlook, meeting summaries in Teams and more. If just 20% of customers sign up, Microsoft could realize as $9 billion in annual revenue.
Still, Microsoft stock slipped after its fiscal Q4 earnings on July 25 amid guidance for higher capital spending led by AI costs but only "gradual" AI revenue growth.
Negative gross margins, with spending exceeding revenue, is to be expected at the front end of a large CapEx cycle, Goldman Sachs analyst Kash Rangan said in a July 5 research report.
"It took a 10-year investment cycle for gross margins in Microsoft's cloud business to go from negative to where they are today, which is well north of 50% to 60%. That's probably the baseline for AI."
Tech Manufacturing Boom Underway
Tapping into the huge potential of AI or getting left behind may be what finally pushes companies to update legacy systems, wrote Ed Yardeni, chief investment strategist at Yardeni Research. He cited a Rackspace Technology survey indicating that 83% of retail-sector IT executives said that benefiting from AI will require modernizing their technology. That backs Yardeni's thesis that "a new technology capital spending cycle — and the Roaring 2020s — are underway!"
The AI investment boom is taking off at a time when businesses are investing heavily to build semiconductor and EV production capacity in the U.S. Spending on manufacturing construction hit an annualized $194 billion in May, up 76% from a year earlier, Commerce Department data shows. That's getting an assist from the Inflation Reduction Act and the Chips Act, both approved in 2022.
Artificial Intelligence Drives Bull Market
While consumer spending has cooled this year amid more Fed rate hikes, helping to moderate inflation, the AI buzz and strength in business investment have unexpectedly given life to a new S&P 500 bull market.
In June, Goldman Sachs boosted its year-end S&P 500 target to 4500 from 4000. Earlier this month, Credit Suisse lifted its S&P 500 year-end target to 4700 from 4050, citing an improved inflation outlook and stronger tech earnings.
Through the July 31 close of 4589, the S&P 500 has rallied 31.4% from its Oct. 13 bear-market low and now stands just 4.8% off its all-time high set on Jan. 4, 2022. The Nasdaq has been even hotter, surging 37% year-to-date — nearly doubling the S&P 500's 19.5% gain.
So far in 2023 through July 31, Nvidia stock has skyrocketed 220%. Microsoft stock has surged 40%.
The question now is whether the market is "getting too hot," Yardeni wrote. He thinks this new bull market has legs and sees potential for the S&P 500 to rise as high as 5400 by the end of 2024. But there's still scope for a near-term correction, he says. (Be sure to read IBD's The Big Picture column every day to stay in sync with the stock market direction and what it means for your trading decisions.)
Short-Term Inflation Pressure?
Over time, the productivity gains from generative AI should help hold down inflation. But, for now, strengthening capital investment and the rebound in stock market wealth could work the other way. This is among near-term and longer-term factors that complicate the bullish AI-driven market outlook.
As with the late 1990s dot-com boom, "The AI craze could temporarily lift bond yields by pushing out the onset of the next recession," BCA's Berezin wrote.
Back then, rising equity wealth and strong capital spending produced a very tight labor market. The Fed ultimately raised its key rate to 6.5% by May 2000, helping to burst the bubble.
This time there's no bubble to prick. The AI boom is in its first year, led by the highly profitable "magnificent seven" stocks, including Microsoft, Nvidia, Google and Meta Platforms. Still, it's unclear whether stock market gains and stronger business investment will help the Fed achieve a soft landing by offsetting consumer weakness or whether they'll give consumption a second wind and compel policymakers to hike rates more than expected.
So far the tech spending rebound appears narrowly focused on AI. Taiwan Semiconductor, which makes most of the world's advanced chips for companies like Nvidia, AMD, Broadcom and Apple, said July 20 that sales will fall 10% this year. While its AI business is booming, that makes up only 6% of sales.
Generative AI Impact On Job Market
With so much potential for AI to displace workers, another key unknown is how the job market will adapt. Historically, providing workers with technology to work faster and better has increased living standards as employees reaped a share of higher profits.
Productivity gains could enable an era of faster, noninflationary growth. That could help provide jobs for displaced workers, including new types of work.
Some 60% of today's jobs are in occupations that didn't exist in 1940, MIT economist David Autor has found. That has minimized the fallout from technological progress.
"In past technological transformations, workers who lost their jobs could transition to new jobs, and on average pay increased," Stanford professor Erik Brynjolfsson and co-authors wrote in a recent paper for Brookings Institution.
"However, given the scale of the impending disruption and the labor-saving nature of it, it remains to be seen whether this will be the case in the age of generative AI."
The government response to worker displacement will have big market implications. Tax increases could erode higher profit margins enabled by AI. Spending to buttress income-support programs or provide a universal basic income could raise rates and undo AI's disinflationary effects, Goldman says. "Higher interest rates also could counteract much of the potential increase in fair value for the S&P 500."
AI Chip War Poses Big Risk
But geopolitics is by far the biggest risk. With so much riding on the AI investment boom, any imminent threats to the supply of crucial chips could trigger serious bouts of volatility for the S&P 500 and Nasdaq. Unfortunately, supply chain threats are a fact of life because AI chips are at the heart of the U.S.-China chip war.
The U.S. has formed a united front with key allies to keep advanced semiconductors made by Nvidia and others out of China's hands. The same goes for equipment that can produce the chips. That has only deepened a rift between China and the U.S. over the status of Taiwan. That's where TSM produces the chips that power the AI revolution.
Despite U.S. diplomacy efforts, the conflict continues to ratchet up. In recent weeks, the Biden administration signaled it will widen chip export curbs. Beijing struck back with its own restrictions on exports of gallium and germanium, two chip inputs whose production China dominates.
The hard-to-gauge risk of a war over Taiwan would have incalculable economic consequences. While the risk hasn't diluted the fervor for AI stocks, markets may reassess after Taiwan elections slated for January. A shift in leadership could move the island closer to China. A status quo election could narrow nonmilitary options for Beijing to assert its sovereignty.
