What happened to the artificial-intelligence investment boom?
Many economists believe that generative artificial intelligence (AI) is about to transform the global economy. A paper published last year by Ege Erdil and Tamay Besiroglu of Epoch, a research firm, argues that 「explosive growth」, with gdp zooming upwards, is 「plausible with ai capable of broadly substituting for human labour」. Erik Brynjolfsson of Stanford University has said that he expects ai 「to power a productivity boom in the coming years」.
許多經濟學家認為，生成式人工智慧（AI）即將改變全球經濟。 Epoch 研究公司的Ege Erdil和Tamay Besiroglu去年發表的一篇論文認為，隨著GDP飆升，「AI能夠廣泛替代人力勞動」的「爆炸性增長」是「合理的」。史丹佛大學的Erik Brynjolfsson表示，他預計AI「將在未來幾年推動生產力繁榮」。
For such an economic transformation to take place, companies need to spend big on new software, communications, factories and equipment, enabling AI to slot into their production processes. An investment boom was necessary to allow previous technological breakthroughs, such as the tractor or the personal computer, to spread across the economy. From 1992 to 1999 American nonresidential investment jumped by 3% of gdp, for instance, driven in large part by extra spending on computer technologies. Yet so far there is little sign of an ai splurge. Across the world, capital expenditure by businesses (or 「capex」) is remarkably weak.
要發生這樣的經濟轉型，公司需要大舉投資新軟體、通訊、工廠和裝置，使AI能夠融入他們的生產過程。以前的技術突破，如拖拉機或個人電腦，要在經濟中廣泛傳播，就必須經歷投資熱潮。例如，從1992年到1999年，美國非住宅投資躍升了GDP的3%，主要是由於額外的電腦技術支出。然而，目前還沒有看到AI狂潮的跡象。全球企業的資本支出（或「capex」）非常疲軟。
After sluggish growth in the years before the covid-19 pandemic, capex increased as lockdowns lifted (see chart). In early 2022 it was rising at an annualised rate of about 8% a year. A mood of techno-optimism had gripped some businesses, while others sought to firm up supply chains. Capex then slowed later the same year, owing to the effects of geopolitical uncertainty and higher interest rates. On the eve of the release of OpenAI’s GPT-4 in March 2023, global capex spending was growing at an annualised rate of about 3%.
在新冠疫情之前的幾年裏，capex增長緩慢，隨著封鎖措施的解除，capex有所增加（見圖表）。在2022年初，它的年增長率約為8%。一些企業抱有技術樂觀主義，而其他企業則尋求鞏固供應鏈。同年晚些時候，由於 地緣政治 不確定性和更高利率的影響，capex增長放緩。在OpenAI於2023年3月釋出GPT-4的前夕，全球capex支出的年化增長率約為3%。
Today some companies are once again ramping up capex, to seize what they see as the enormous opportunity in ai. This year forecasters reckon that Microsoft’s spending (including on research and development) will probably rise by close to 20%. Nvidia’s is set to soar by upwards of 30%. 「AI will be our biggest investment area in 2024, both in engineering and compute resources,」 reported Mark Zuckerberg, Meta’s boss, at the end of last year.
如今，一些公司再次加大capex投入，以抓住他們所看到的AI巨大機遇。今年，預計微軟的支出（包括研發）可能會增長近20%。Nvidia的支出預計將增加超過30%。去年年底，Meta的老板馬克·祖克柏報道說：「AI將是我們2024年最大的投資領域，包括工程和計算資源。」
Elsewhere, though, plans are more modest. Exclude firms driving the AI revolution, such as Microsoft and Nvidia, and those in the S&P 500 are planning to lift capex by only around 2.5% in 2024—ie, by an amount in line with inflation. Across the economy as a whole, the situation is even bleaker. An American capex 「tracker」 produced by Goldman Sachs, a bank, offers a picture of businesses’ outlays, as well as hinting at future intentions. It is currently falling by 4%, year on year.
然而，在其他地方，計劃則更為謹慎。排除推動AI革命的公司，如微軟和Nvidia，標普500指數中的公司計劃在2024年僅將capex提高大約2.5%——即與通脹水平相當。在整個經濟中，情況甚至更加黯淡。高盛銀行制作的一個美國capex「追蹤器」提供了企業支出的情況，並暗示了未來的意圖。目前它正在以4%的年率下降。
Surely, with all the excitement about generative AI’s potential, spending on information technologies is at least soaring? Not quite. In the third quarter of 2023 American firms’ investment in 「information-processing equipment and software」 fell by 0.4% year on year.
毫無疑問，隨著人們對 生成式AI 潛力的興奮，資訊科技的投資至少在飆升嗎？並非如此。在2023年第三季度，美國公司在「資訊處理裝置和軟體」上的投資同比下降了0.4%。
Similar trends are observable at a global level. According to national-accounts data for the oecd club of mostly rich countries, which go up to the third quarter of 2023, investment spending—including by governments—is growing more slowly than in the pre-pandemic years. A high-frequency measure of global capex from JPMorgan Chase, another bank, points to minimal growth. With weak capex, it is no surprise that there is little sign of productivity improvements, according to a real-time measure derived from surveys of purchasing managers (see chart).
在全球範圍內也可以觀察到類似的趨勢。根據經合組織（主要是富裕國家的俱樂部）的國民帳戶數據，截至2023年第三季度，包括政府在內的投資支出增長速度比疫情前的年份慢。摩根大通另一家銀行提供的全球capex高頻測量指標顯示最小增長。在capex疲弱的情況下，據采購經理調查衍生的即時測量顯示生產力改進的跡象很少（見圖表）。
An official survey in Japan does point to sharply higher capex growth in the future, after years of sluggishness. Yet this probably reflects factors specific to that country, such as reforms to corporate governance. And in most places outside America the situation is rather less encouraging. A worsening outlook for the economy in Europe does not help. Investment intentions of services companies in the European Union are less than half as ambitious as they were in early 2022. British businesses plan to raise capex by a mere 3% over the next year, compared with 10% when asked in early 2022.
日本的官方調查確實指出未來capex增長將顯著提高，結束多年的低迷。然而，這可能反映了該國特有的因素，如對公司治理的改革。在美國以外的大多數地方，情況則不那麽樂觀。歐洲經濟前景的惡化並沒有幫助。歐盟服務公司的投資意圖不到2022年初的一半。與2022年初詢問時相比，英國企業計劃在未來一年內將capex增加僅3%。
These trends suggest one of two things. The first is that generative AI is a busted flush. Big tech firms love the technology, but are going to struggle to find customers for the products and services that they have spent tens of billions of dollars creating. It would not be the first time in recent history that technologists have overestimated demand for new innovations. Think of cryptocurrencies and the metaverse.
這些趨勢表明兩種可能性之一。第一種可能性是生成式AI是一個失敗的投資。大型科技公司熱衷於這項技術，但可能會發現很難找到客戶來購買他們花費數百億美元建立的產品和服務。這並不是近代科技人員第一次高估了對新創新的需求。想想加密貨幣和元宇宙。
The second interpretation is less gloomy, and more likely. The adoption of new general-purpose technologies tends to take time. Return to the example of the personal computer. Although Microsoft released a groundbreaking operating system in 1995, American firms only ramped up spending on software in the late 1990s. Analysis by Goldman Sachs suggests that while only 5% of chief executives expect AI to have a 「significant impact」 on their business within one to two years, 65% think it will have an impact in the next three to five. AI is still likely to change the economy, but with a whimper not a bang.
第二種解釋不那麽悲觀，也更有可能。新的通用技術的采用往往需要時間。回到個人電腦的例子。盡管微軟在1995年釋出了突破性的作業系統，美國公司直到1990年代末才加大對軟體的投資。高盛的分析表明，雖然只有5%的執行長期望AI在一到兩年內對其業務產生「重大影響」，但有65%的人認為它將在接下來的三到五年內產生影響。AI仍然有可能改變經濟，但這將是一個悄然而非轟動的變化。
