한국을 비롯한 글로벌 증시가 급락한 배경에는 그동안 기술주 랠리를 이끌어 왔던 인공지능(AI) 테마에 대한 회의론이 자리잡고 있다는 분석이 나온다. 주가가 과열됐다는 투자자들의 ‘깨달음’이 연쇄 투매를 불렀다는 것이다. 하지만 무너져 내린 것은 AI 기업의 가격과 기대감일 뿐, AI 산업의 수요와 수익모델 등 ‘펀더멘털(내재가치)’은 견고하다는 반론이 나온다.
5일(현지시간) 미 증시에서 애플 주가는 4.82% 떨어졌다. 엔비디아도 6.36%, 구글 모회사 알파벳은 4.61% 하락하면서 몇년간의 증시 상승세를 이끌어왔던 기술 랠리가 마침표를 찍는 모습이다.
그 배경으로는 엔화 강세에 따른 유동성 쇼크, 미국 고용지표 악화 등이 꼽힌다. 세계 경기 전망에 대한 공포가 매도 행렬로 이어졌다는 분석이다.
이런 매크로(거시경제) 분석에 더해, AI 테마에 대한 우려도 주가 하락에 적잖은 영향을 끼쳤다. 이른바 ‘AI 거품론’이다. 지난 2022년 챗GPT 등장 이후 AI 산업이 급격한 관심을 받았다가, ‘닷컴 버블’처럼 차갑게 식고 있다는 논리다. AI 투자를 이끄는 빅테크 4개사(구글·메타·아마존·마이크로소프트)를 비롯한 미국 대형 기술기업 7개사(매그니피센트7)의 시가총액이 스탠더드앤푸어스(S&P)500 지수에서 차지하는 비중은 30%가 넘는다. 이들의 가치 하락은 전체적인 시장 붕괴로 이어질 수 있다.
그 발단은 기대에 못 미치는 실적이었다. 지난주 열린 2분기 실적 발표회에서 구글은 분기당 120억달러(약 16조원)에 달하는 AI 투자를 단행했으면서도 수익 실현 시점을 두고는 불투명한 답변을 내놓았으며, 마이크로소프트 역시 AI 클라우드 매출이 전망치를 밑돌았다. 그러면서도 이들 기업들은 엔비디아 그래픽처리장치(GPU) 구입을 비롯한 AI 설비투자 금액은 더 늘릴 것이라고 밝혔다.
애덤 사르한 ‘50파크 인베스트먼트’ 최고경영자(CEO)는 “투자자들은 ‘나에게 보여줘(show me)’ 단계에 접어들고 있으며, AI가 수익과 생산성에 미치는 영향에 대한 구체적인 증거를 찾고 있다”고 말했다.
‘거품이 아니다’라는 반론도 나온다. 캐피탈이코노믹스의 수석 경제학자 존 히긴스는 이날 “닷컴 버블이 터진 2000년보다는 주가가 일시적으로 폭락한 1998년과 더 비슷해 보인다”고 밝혔다. 1998년에도 미국의 실업률이 약간 상승하고 엔화가 급등해 증시가 조정을 겪은 바 있는데, 지금은 그때와 달리 시스템적인 불안 요소는 없다는 주장이다. 히긴스는 “경제는 우려했던 것보다 더 잘 견뎌낼 것이며 투자자들도 AI에 대한 열정을 재발견함에 따라 주식시장이 회복될 것”이라고 덧붙였다.
시장이 AI 기업 실적에 과민반응을 보이고 있다는 시각도 있다. 구글 등 AI 투자를 주도하는 빅테크 기업들의 2분기 실적이 전망치에는 다소 못 미쳤으나, 그렇다고 수익성이 악화하고 있다는 근거도 희박하기 때문이다. 이들 기업이 ‘밑빠진 독에 물 붓듯’ 투자하는 상황도 아니다. 기존 캐시카우(수익사업)에서 번 여유자금을 투자하고 있기 때문이다. 글로벌 투자은행 UBS는 “(기술 기업들의)펀더멘털은 여전히 견고하며 이들은 2분기에 전년대비 24%의 수익 성장을 보고할 것”이라며 “AI 수익화가 회복되고 있다는 증거가 더 많이 있으며 내년 AI 칩에 대한 수요가 강할 것으로 예상된다”고 밝혔다.
4개 빅테크 업체 모두 마진이 상당폭 확대되고 현금흐름이 증가하는 가운데 비용을 굉장히 신중하게 배치하고 있다. 빅테크의 AI 데이터센터 투자 기조는 변한 것이 없다. AI 구현에 핵심 역할을 맡고 있는 AI 가속기와 고대역폭메모리(HBM) 공급 과잉, 수요 둔화를 우려하기엔 아직 이른 시점이다.
Are We in a Dot-Com Style Artificial Intelligence Bubble?
Although artificial intelligence has been a familiar concept to most of us for at least a decade or more, there hasn’t been a day since November 2022 that I haven’t seen a news article or social media post about it.
Of course, November 2022 was when ChatGPT was released, and it promised to change everything. We are approaching 18 months since that date, and it’s safe to say that the world hasn’t changed much, apart from AI-mania still being at an all-time high.
Meet Your Salesforce Safety Net 1
In the days and weeks following ChatGPT’s release, there was an instant response by internet side hustlers selling ChatGPT courses and changing their LinkedIn status to “AI Expert” or “Prompt Engineer”. On the surface there isn’t much wrong with this, but the marketing behind these courses is usually some form of overzealous promise on how this is going to benefit your career or ill-mannered scaremongering of how you will lose your job if you don’t buy it.
But come on, let’s be serious. Talking to ChatGPT or any other AI Chatbot isn’t exactly rocket science. For those who have grown up with the internet, it’s almost like being good at using Google.
Although I am bought into AI-mania as much as the next technology professional, the hype and language used by some of these side hustlers has been an early warning signal for me that we are in a bubble, not too dissimilar to a canary in a coal mine.
Are We in an AI Bubble?
It’s inherently human to get excited by the next big thing and throw all your energy into it. We’ve had it with crypto, NFTs, the metaverse, and now artificial intelligence. Whilst a few of us have been burned on buying digital art, a virtual racehorse, or some land in the metaverse (my friends and I have been burned on all three), artificial intelligence felt very different. This is mainly due to the fact there was an instant practical application of ChatGPT that went way beyond what we were used to with search engines.
But fast forward 18 months and we’re being told by many that it’s not AI that will be taking our jobs, but someone who is using AI. So where exactly are we?
One of the biggest differences from 18 months ago is tech stock prices. Nvidia is leading the charge at over 400% increase since Nov 2022, AMD is at 112%, Microsoft and Google are both sitting pretty at around 60% increases, and Salesforce has grown a whopping 84%.
Now, when I question whether we are in an AI bubble, I am talking not only about the financial markets, but also the hype that is being pushed to us daily by those who have something to sell.
Recently, there have been a number of articles discussing the AI bubble theory. Whilst ChatGPT has a huge amount of practical application, are the current valuations and hype justified?
In an interesting article by John Naughton, From boom to burst, the AI bubble is only heading in one direction, ChatGPT is asked about the stages of the bubble: displacement, boom, euphoria, profit-taking, and panic. It states that we are currently in the euphoria stage, moving into the profit-taking phase – but there is little profit to be had unless you are selling hardware like Nvidia.
On the other hand, Quartz argues against the theory that the AI boom is anything like the dot-com era: “today, AI is capable of generating substantially greater revenues than the internet was in the 1990s and early 2000s”.
Whilst I largely agree with this statement, AI doesn’t come without faults – hallucinations being the big one. The ChatGPT style language has also led to a new name being coined by Alex Hern at the Guardian: “AI-ese”. Whilst text generated by ChatGPT is grammatically correct and easy to read, there is a lot of waffle, and it uses certain words so often that it’s become very easy to spot when text has been AI-generated: words such as “’explore’, ‘tapestry’, ‘testament’ and ‘leverage’ all appear far more frequently in the system’s output than they do in the internet at large”.
Moreover, has the hype cycle put pressure on companies to dive into GenAI now, even if they aren’t ready? In a recent study conducted by Slack, they found that “nearly all executives feel pressure to integrate AI tools into their organization, with half of all executives saying they feel a high degree of urgency to incorporate AI tools”.
So it comes as no surprise Gartner predicts that by 2025, 90% of enterprise deployments of GenAI will slow, as costs exceed value, and 30% of those projects will be abandoned after proof of concept (POC) due to poor data quality, inadequate risk controls, escalating costs, or unclear business value. Accenture is one company reaping the rewards of this hype, with over $1 billion in AI bookings in six months.
How Does the Bubble Impact Salesforce?
Our beloved Salesforce is a fantastic case study for many of the points I’ve outlined here. They have caught on early to the fact that data, and huge amounts of it, is the only way to power the next generation of artificial intelligence applications. It’s not enough to use the most generic LLM possible (GPT3/4); you need to use specific LLMs for specific purposes and use information that is specific to your business.
This is why Salesforce has rebranded its platform “Einstein 1”, combining artificial intelligence in the form of Einstein, as well as their Data Cloud. Salesforce were looking to further bolster their data product armory with the acquisition of Informatica, but talks have since fallen through.
Salesforce are also a great case study for another point – their GPT products have only been generally available for a couple of months now. It took Salesforce exactly a year from announcing their GPT products at TrailblazerDX 2023 to then release them at TrailblazerDX 2024. Whilst this is an impressive feat, this shows that doing things properly takes time.
As Salesforce have only just released their GPT products to the world, it’s going to take time for customers to evaluate, implement, and integrate these products into their existing business processes – especially since AI is still in its infancy, and has very obvious issues when it comes to hallucinations and trust. It’s also going to take time to translate to dramatic revenue growth for Salesforce, as well as productivity boosts for its customers.
There are a few companies that are profiting in real terms, such as Nvidia, ARM, Amazon, Microsoft and Palantir. But these companies are the exception, not the rule. With so much VC money flowing into startups, Futurism quotes that the real losers will be those “who are raising money on the promise of selling their services for $20/user/month”.
Why Does a Bubble Matter?
Bubbles are defined as an economic cycle, characterized by the rapid escalation and subsequent decline of asset values such as the stock market. So unless you are investing in financial markets, why does it matter?
Well, bubbles also create hype, and hype can massively impact the decisions you are making for your business or your career.
It took 48 years for electricity to reach 100% of American households, and it took half the time for the internet to go from 10% to 88% adoption in the states. ChatGPT racked up a record 100 million users in only a couple of months, becoming the fastest-used technology in history.
Whilst it’s only logical that AI is going to be adopted faster than the internet due to the fact that technologies and ideas can now spread faster than ever, strong foundations still have to be laid out. If AI is adopted without a proper understanding of use cases, and LLMs that aren’t fit for purpose are being used, at best, you could be throwing money down the drain. And at worst, you could put company data or processes at risk.
Even Salesforce, who have been some of the biggest ‘hypemen’ around for AI, have started to cool their jets. At the Amsterdam World Tour on April 18, Ed Thompson, an Ex-Gartner analyst who now works for Salesforce, suggested that we could be heading past the hype.
In the classic Gartner Hype Cycle that observes how new exciting technologies enter the market, the stereotype is for the hype to get out of control, before falling back down to earth, slowly reaching a level of maturity. Thompson suggested with the use of some headlines such as Amazon’s AI chatbot leaking data, and chatbots being vulnerable to indirect prompt injection attacks, that we are entering the trough of disillusionment.
Summary
As much as some individuals and companies would like to tell you that AI is growing at a high speed (unless you pay X money or buy Y product), the truth is that whilst the implementation of AI is moving faster than it took people to adopt the internet, or for cars to become ubiquitous, it’s only natural that the implementation of groundbreaking new technologies will take time.
Just as infrastructure and security protocols had to be built to accommodate the internet and cars, the foundations now have to be laid for AI to be useful for companies going forward. We need systems integrated, data organized, and AI use cases to be fleshed out and experimented with. The whole world is only just getting started on this journey.
Should you ignore AI? Of course not! As many have said, this is potentially bigger than the internet itself, and a new revolution is underway. But is buying a single course on prompt engineering for hundreds of dollars going to help you? I doubt it.
The enthusiasm for artificial intelligence (AI) investment is rapidly cooling down. This is due to the spread of pessimism that the third Ice Age (AI Winter) is coming instead of the rosy prospect that it will revolutionize human life. The AI industry has experienced two recessions in the 1970s and late 1980s as it faced technical limitations in the past.
Wall Street in New York has recently been pouring out AI bubble theory. The perception that the astronomical funds poured out by big tech companies for two to three years are not leading to actual sales has spread, and the prospect of not increasing productivity as expected by AI for quite some time is gaining momentum.
The Economist analyzed that investors have increased the market value of the five major big techs by more than $2 trillion over the past year, but AI-related sales will be around tens of billions of dollars this year. At this scale of investment, annual sales should be between $300 billion and $400 billion, but there is virtually no profit model for the entire industry.
Jim Covelo, head of Goldman Sachs' global stock research, is a representative "AI skeptic." He believes AI will not bring about revolutionary changes as it did when smartphones and the Internet appeared in the past. His comments, an analyst who has dealt with the tech sector for more than 30 years, sparked the "AI overinvestment theory."
"The innovative technological transition in history has been to replace very expensive solutions with inexpensive solutions," Covelo said in a recent Bloomberg interview. "Replacing jobs with technology that costs a lot (like AI) is the opposite."
Contrary to expectations that AI will replace people's jobs and increase productivity, actual AI is expensive and does not have much productivity improvement effect.
Appearing on Goldman Sachs' own podcast, he disparaged, "Years have passed, but so far, there is no use of AI for cost-effective purposes."
Investment bank Barclays also said in a report, "Only ChatGPT and GitHub Copilot have achieved great success in the consumer and business sectors 20 months after ChatGPT came out," adding, "Wall Street's skepticism is growing."
With the recent return of the earnings release season of big tech companies, Wall Street investors have focused their attention on how to monetize funds that have been excessively invested in AI infrastructure.
At Microsoft's earnings conference call on the 30th, Morgan Stanley analyst Keith Weiss asked CEO Satya Nadella about the capital investment plan, saying, "The industry is currently debating whether capital and monetization related to Generative AI will meet expectations." CEO Nadella tried to calm investor anxiety by saying that capital investments are made in line with customer demand and the growth of cloud AI.
When Alphabet (Google) announced its earnings on the 23rd, investors asked about AI investment, but Alphabet's stock price plunged when the answer fell short of expectations.
MIT economics professor Darren Aussimoglu, famous for "Why the State Fails," also joined the AI bubble theory. "The large-scale language model (LLM) is more impressive than many people expected," he said in an interview, but pointed out, "It takes a big leap of faith to be able to get something as smart as the AI of the movie '2001 Odyssey' just by predicting the next word." He also argued in a paper published in May that "the productivity improvement caused by Generative AI is only about 5% over the next 10 years."
Sequoia Capital, one of Silicon Valley's top venture capital firms, has also warned of an AI bubble. David Khan, partner of the company, emphasized in a post titled "AI's $600 billion problem" that the gap between infrastructure investment costs and expected sales of big tech companies is gradually widening.