The AI revolution’s first year: has anything changed?

Once every decade or so the computing world finds itself on the cusp of a revolution.

Whether it’s the commercialisation of the internet in the 1990s or the birth of mobile and cloud computing in the first decade of this century, a collective belief takes hold in the transformative potential of a new technological paradigm.

For many in the tech industry, 2023 will go down as the year that generative AI changed everything. The ability of computers to automatically generate text or images with the apparent facility of a human first came to widespread attention with the launch of OpenAI’s ChatGPT late in 2022.

Not since the debut of the iPhone has a single product fuelled such powerful hopes for a new era of technology. By the end of this year, it had led to a race in the tech industry to bring generative AI from the research lab into everyday use, embedding it in many of the most widely used digital products and services.

For Microsoft chief executive Satya Nadella, encountering ChatGPT was like the time, 30 years ago, when he first saw a web browser. Describing that moment to the FT earlier this year, he said: “It just clicks that, wow, it’s a different day.”

Google chief executive Sundar Pichai was also caught by the wave of optimism, declaring that artificial intelligence, of which generative AI is merely the latest instalment, will be more important than fire or electricity.

Erik Brynjolfsson, a professor at Stanford University who has studied the adoption of other historically important new technologies, predicts generative AI’s impact on working practices could spur a productivity boom across the global economy.

But even as the buzz around the technology has intensified, serious doubts about its practical usefulness have surfaced.

The large language models that form the foundation of generative AI deal in probabilities, not the hard logic of traditional computing systems. They are capable at times of breathtaking artistry, whether writing computer code or poetry. But they also have an alarming tendency to return inaccurate information and “hallucinate” by generating plausible-sounding responses that have little relation to reality.

Even some companies racing to deploy the technology admit that these inherent shortcomings will limit its usefulness, even as the tech industry hurries to find ways to mitigate the problems. “It will be useful, but not as radically game-changing as many people hope,” says Peter Schwartz, head of strategy at Salesforce, one of the software companies seeking to embed generative AI into many of their products.

Whether generative AI turns out to be as revolutionary as the boosters claim, or merely a useful addition to the IT arsenal with limited applications, should start to become clearer in 2024. The technology has been a catalyst for a powerful tech stock rally, helping to turn a small group of leading tech companies into this year’s undisputed stars on Wall Street. Without strong momentum behind generative AI’s adoption, that could be shortlived.

After a year in which it was obligatory for every tech company to come up with a generative AI strategy, the time is fast approaching when Wall Street will start to demand real revenue and profits from the technology.

“2024 is going to be the year where we see who is just playing the AI card, as opposed to having a real business model,” says Jim Tierney, a growth stock investor at AllianceBernstein.

Boom or boomlet?

While the hopes swirling around AI have been an important catalyst in the change of mood on Wall Street, they were not the most direct cause of the 2023 tech stock rebound.

Falling interest rates and the durability of Big Tech’s profits, along with upward earnings revisions, were the main factors carrying stocks higher, says Tierney. As hopes for a soft landing in the US economy have heightened and growth investments have come back into favour, tech has carried the overall market higher.

For the tech companies themselves, the turnaround has been dramatic. After shedding 40 per cent of their combined value, or $3.7tn, in 2022, the five biggest US tech companies — Apple, Microsoft, Alphabet, Amazon and Meta — have gained back $3.9tn this year.

AI fervour supported this rebound, while also having a more direct impact on one company at the heart of the AI boom. The stock market value of Nvidia, which has a clear lead in making the chips needed to train the latest AI models, jumped by $800bn, making it the biggest percentage gainer among large tech companies in a banner year.

Nvidia’s sales reflect a common feature of new tech waves. Investment floods into the infrastructure needed to support AI-powered applications, even before it is clear what the most important applications will be or whether they will generate sufficient profits to justify the investment.

The tech world is grasping to understand the scale of the boom that is unfolding. Lisa Su, chief executive of Nvidia’s rival chipmaker AMD, predicted in December that sales of AI chips for data centres would soar to $400bn in 2027 — a significant increase from the $150bn prediction she made less than four months ago — and a figure equal to the entire global semiconductor market in 2019.

Such estimates are the kind of stabs in the dark that people usually make when a new technology is taking off, according to Tierney. “The euphoria over the potential [of AI] is about as high as it’s going to be right now,” he adds.

In 2023, private investment into new AI companies also picked up sharply, though the volume has not matched previous tech booms. According to PitchBook data, $27bn flowed into private AI companies like OpenAI, which are building the large language models that underpin generative AI.

Bill Janeway, a veteran tech investor and former head of Warburg Pincus, calls the wave of generative AI investment a “boomlet”, and “not that big a deal in terms of financing”.

If anything, the lack of a bigger AI investment bubble may actually hold back the development of the technology, says Janeway, limiting the amount of “trial and error” that takes place when capital is thrown about more freely.

A slow start

After a year in which many companies laid the foundations for wider use of AI, investors are now starting to look ahead to 2024, and to come to terms with what is likely to be slow adoption of the technology, at least in the short term.

Cautious forecasts from some of the tech companies expected to lead the new generative AI wave have tempered expectations. Software company Adobe, whose shares had jumped nearly 90 per cent since the start of the year, was the latest to disappoint, with a revenue forecast for 2024 that fell short of Wall Street’s hopes.

Microsoft, which has moved faster than most to embed AI into its software, has also worked hard to damp expectations, saying that a pick-up in sales from the new “Copilot” features in its software is not likely until the second half of the year.  

One reason for caution is the lingering issue of large language models producing inaccurate results, undermining their value in many business settings. “It’s a big enough problem and enough people are working on it in parallel” to suggest that solutions will eventually be found, says Schwartz at Salesforce.

He and others point to two methods in particular that offer hope: ensuring that work involving generative AI always has a “human in the loop” to catch mistakes, and linking the language models to factual databases so that they can return verifiably accurate answers when needed.

Another issue likely to slow the adoption of generative AI in 2024 is a lack of preparedness on the part of many potential customers. Experts including Janeway say that much of the value companies stand to get from generative AI will come from training the models on their own in-house data.

But according to Julie Sweet, chief executive of Accenture, many companies hoping to use AI lack the technical knowhow. “Most companies do not have mature data capabilities, and if you can’t use your data, you can’t use AI,” she said in an interview with the FT this month.

Cost is also likely to slow uptake of the technology. In one of the strongest displays of confidence in the value of generative AI, Microsoft slapped a price tag of $30 a month on the use of the technology in its Office suite of productivity software, a move that nearly doubles the cost of the software for some customers.

The software company claims that the pricing reflects the huge increases in worker productivity that will come from the software. But analysts warn that the high price will cause customers to limit the technology to small groups of workers, at least until they have clear evidence of its value.

Issues like these mean that, at least in the short term, the lift to tech companies’ revenue and profits from new AI products and services could be muted.

Spending on generative AI next year will amount to little more than $20bn, or 0.5 per cent of total global IT spending, says John-David Lovelock, chief forecaster at IT research firm Gartner. By comparison, IT buyers will spend five times as much on security, he adds.

Sweet at Accenture told her company’s investors this month that tech spending among customers is picking up, though “it’s not increasing as fast as it was increasing a couple of years ago” and much of the spending on AI reflects customers “reprioritising” their existing spending.

Seeking demand

Yet signs that the early take-up of generative AI may be slow have done nothing to damp the hype in the tech industry. Many claim it will find its way into mainstream use more quickly than other important new technologies. According to Sweet, for example, most companies are likely to adopt generative AI faster than they did cloud computing.

Compared with other technologies in their early days, like the internet, AI also benefits from existing computing and communications systems, says Brynjolfsson of Stanford. “Things will happen somewhat faster this time around because the infrastructure is in place,” he says.

Macro forecasts tell a positive story. Analysts at Goldman Sachs estimate that after a slow start, with investment in AI hovering below half a per cent of GDP, spending will jump in the latter part of this decade to reach more than 2.5 per cent of GDP by 2032.

Exactly how and when this will translate into demand for real tech products and services, or which tech companies will see the biggest benefits, is another matter. 

For consumers, there is no “highly monetised killer app” yet that could turn generative AI into a big moneymaker, says Oren Etzioni, former head of the Allen Institute for AI, a research and engineering non-profit organisation. That makes it similar to the early internet, when there was a preponderance of free services and little in the way of online advertising — though Etzioni predicted that ways to make money would quickly take off, just as they did on the internet. 

Many investors, meanwhile, predict that the bigger opportunity will come from business use of the technology. “It looks like AI will get monetised in the enterprise,” says Kevin Walkush, a portfolio manager at Jensen Investment Management. That would make it an extension of the cloud computing wave that has seen many companies invest hugely in shifting their IT to the cloud.

The tech companies themselves could be the first to record significant gains from using the technology, reducing their hiring as they adopt generative AI in their own businesses. IBM chief executive Arvind Krishna said earlier this year that his company would “pause” its recruitment of back-office staff in anticipation that many of their jobs would be replaced by AI.

Both Amazon and Microsoft have designs on using generative AI to make their workforces more productive, adds Tierney, potentially bringing an end to the years of rapid headcount growth that limited an expansion in profit margins. “Doing more with less is the new mantra [for tech companies]. From an investor perspective, that’s pretty darned welcome,” he says.

Some early research into the use of generative AI suggests that business gains could come quickly. A study co-authored by Brynjolfsson at Stanford this year found that the average productivity of call centre workers who used the technology rose by 14 per cent within a matter of weeks. For workers with the lowest skill levels, the gains averaged 35 per cent.

The promise of productivity improvements like this has led many CEOs to press their organisations to study how best to adopt the technology, leading to a close focus on how specific tasks can be adapted to make use of the new AI, Brynjolfsson adds.

Many workers may soon get their first taste of a technology that can help them write reports, analyse corporate data and summarise meetings.

One of the most striking findings from Microsoft’s early trials with generative AI has been how “viral” the technology is among workers, according to Jared Spataro, a vice-president at the software company. People take to the technology quickly when they see others around them using it, and once they have tried it out in their own jobs they don’t want to give it up, he says.

The high cost of the technology and uncertainty about how best to integrate it into everyday business means that the generative AI revolution will not come quickly. But 2024 could be the year when a new way of working starts to take shape.