研究員稱，歷史性 AI 建設融資在美國引發系統性風險

霍華德・施奈德報導By Howard Schneider

華盛頓，9月24日（路透）——一項新研究指出，人工智慧（AI）建設所需占美國產出的比重，將超過電力、鐵路、州際公路或網際網路的普及建置；而其金融結構也日益複雜，可能構成系統性風險。WASHINGTON, Sept 24 (Reuters) - The artificial intelligence buildout is on track to require a larger share of US output than the rollout of electricity, railroads, interstate highways or the internet, with an increasingly complicated financial structure that poses potentially systemic risks, according to a new study.

哥倫比亞商學院金融與不動產教授 Stijn Van Nieuwerburgh 在一篇為本週布魯金斯研究院研討會準備的論文中寫道，原本由亞馬遜（Amazon.com）、Meta Platforms 以及 Alphabet 旗下 Google 等公司以累積現金支付的支出，如今已演變成一波擴張，將在2032年前每年消耗約國內生產毛額（GDP）的3.6%，金額超過10兆美元，並且正在採用愈來愈複雜的融資安排。What had been paid for out of the cash stockpiled by companies like Amazon.com, Meta Platforms and Alphabet's Google has morphed into an expansion that will consume around 3.6% of gross domestic product annually through 2032, or more than $10 trillion, and is using ever more intricate financing arrangements, Stijn Van Nieuwerburgh, a finance and real estate professor at Columbia Business School, wrote in a paper prepared for a Brookings Institution conference this week.

這項估算高於19世紀末鐵路建設所吸收的每年GDP 2.2%；鐵路是成本次高的通用技術建置。這也略高於美國自1950年代開始興建州際公路系統、或1990年代中期開始的電信擴張各自每年約略超過1%的水準。That estimate is higher than the 2.2% of annual GDP absorbed by railroads in the late 1800s, which was the next most costly rollout of a general technology, and a bit more than 1% annually each for construction of the US interstate highway system beginning in the 1950s or the telecommunications expansion that started in the mid-1990s.

Van Nieuwerburgh 寫道，正如鐵路與電信擴張曾導致顯著泡沫與泡沫破裂，AI 建設的規模、仍未經考驗的營收來源，以及圍繞 AI 浮現的複雜融資結構，都意味著它可能已醞釀下挫風險。Just as the rail and telecoms expansions led to notable bubbles and busts, Van Nieuwerburgh wrote that the extent of the buildout, the still-untested revenue streams, and the intricate financing structure emerging around AI mean it could be primed for a fall.

他在向記者簡報時談到 AI 公司、大型科技超大規模雲端服務業者、銀行、私募信貸機構、不動產公司，以及參與建設的其他眾多業者之間正在形成的安排時說：「這實在複雜得嚇人。」他保守估計，未來七年將新增183吉瓦的資料中心容量，相較之下，目前已建置容量約為57吉瓦。"This is freaking complicated," he said in a briefing with reporters of the arrangements emerging between AI firms, major tech hyperscalers, banks, private credit lenders, real estate firms, and a host of other players involved in building what he conservatively estimated at 183 gigawatts worth of new data-center capacity over the next seven years, compared with about 57 gigawatts currently installed.

資料中心建設及 AI 相關風險已成為美國政治與經濟辯論的核心議題。一些地方愈來愈不願接納這類設施，並擔憂地方資源承受壓力；聯準會官員也在評估這波建設熱潮是否正在推升通膨。一些 AI 業界主管則表示，放慢發展步調可能較為安全。Data center construction and the risks around AI have become a central issue in US political and economic debates, with some localities increasingly reluctant to host the facilities and worried about strains on local resources, and Federal Reserve officials considering whether the construction boom is adding to inflation. Some AI executives have suggested a slower pace of development might be safer.

「顯著下行風險」'MEANINGFUL DOWNSIDE RISK'

Van Nieuwerburgh 在這篇將於週五發表的論文中指出，目前正在進行的投資，已經超出主要業者能以自身現金流支應的範圍。轉向外部融資提高了槓桿、將風險重新分散至整體經濟，並使這項事業仰賴尚未獲得證明的營收來源。The investment underway already has outstripped what the major players can fund from their own cash flows. The shift to outside financing has increased leverage, redistributed risks across the economy, and made the venture dependent on revenue streams that have yet to be proven, Van Nieuwerburgh noted in the paper, which will be presented on Friday.

他在與記者談話時說：「所有這些特殊目的機構（SPV）的不透明性，多少令人想起次級房貸危機中發生的事。」當時複雜的房屋貸款融資安排出現問題，惡化程度震撼全球金融體系，並引發美國2007至2009年的經濟衰退。"This opacity of all these special purpose vehicles is somewhat reminiscent of what happened in the subprime mortgage crisis," he said in the conversation with journalists, when complex home mortgage financing arrangements went bad at rates that rocked global financial systems and triggered the 2007-2009 recession in the US.

他寫道：「這些發展並不意味著金融困境迫在眉睫。AI 應用強勁成長、高使用率，以及模型能力持續改善，都可能支撐預期中的基礎建設，並產生穩定現金流。」「但不確定的需求、快速的技術變化、執行瓶頸與高槓桿相結合，使得一旦預期遭到修正，便會產生顯著下行風險。」"These developments do not imply that financial distress is imminent. Strong growth in AI applications, high utilization, and continued improvements in model capability could support the projected infrastructure and generate stable cash flows," he wrote. "But the combination of uncertain demand, rapid technological change, execution bottlenecks, and high leverage creates meaningful downside risk if expectations arerevised."

他舉例寫道，AI 產業到2032年必須達到約3.7兆美元的年營收，才能實現預期投資報酬；而「以目前估計 OpenAI 與 Anthropic 年營收合計約1,000億美元來看，營收必須以每年約80%的速度成長」。As an example, he wrote that the AI industry will need to be earning about $3.7 trillion in annual revenue by 2032 to achieve the expected return on the investment, and "given current estimates of annual combined revenues of OpenAI and Anthropic of around $100 billion, revenues would need to grow at roughly 80% per year."

（霍華德・施奈德報導；保羅・西芒編輯）(Reporting by Howard Schneider; Editing by Paul Simao)