Big Tech는 보증을 통해 장부에서 3000억 달러의 AI 노출을 유지합니다.

An FT investigation finds Meta, Nvidia, Broadcom and peers issued up to $300bn of guarantees that keep AI data center and chip debt off balance sheets.

Some of the biggest names in technology have issued up to $300 billion of guarantees tied to AI data centers and chips over the past year, keeping most of that debt off their own balance sheets, according to a Financial Times investigation published on Saturday.

The report describes a financing pattern that has become a defining feature of the AI infrastructure boom: instead of borrowing the enormous sums needed for data centers and accelerated computing hardware directly, technology companies increasingly backstop debt raised by separate financing vehicles. The parent company promises to make lenders whole — often by guaranteeing a minimum future value for the chips or facilities involved — while the borrowings themselves sit on the vehicles' books, not the tech giant's.

How the Guarantee Structure Works

The mechanics are straightforward even if the scale is not. A data center project or an AI cloud startup raises debt through a special-purpose vehicle. The vehicle borrows against the residual value of the assets — GPUs, servers, buildings. A big technology company then issues a guarantee that those assets will retain a certain value, or that the debt will be repaid.

The effect, as the FT reports, is that lenders behave in part as though they are lending to the guarantor — a Meta, an Nvidia, a Broadcom — rather than to a young AI startup or a single-purpose data-center company. That lowers borrowing costs dramatically and unlocks capital at a pace the AI sector could not otherwise access.

It also means the exposure barely registers on the guarantor's balance sheet. The companies report the guarantee commitments in the footnotes of their filings, while the associated debt is consolidated nowhere.

Alphabet's Fast-Moving Guarantee Footprint

The FT's reporting includes a striking data point about Alphabet. The company's guarantees tied to data-center leases reportedly jumped from $16.9 billion to $43.8 billion in about six months, with less than 2% of the associated obligations booked on its balance sheet, according to the investigation.

Meta's massive Hyperion data-center project in Louisiana, a development valued at roughly $50 billion, has also been financed through a structure that keeps the debt off Meta's books, per the FT's reporting on the deal.

Taken together, the FT calculates that guarantees issued by Meta, Nvidia, Broadcom and other technology companies over the past year total as much as $300 billion — a figure that sits within a much larger universe of AI infrastructure commitments that industry-wide estimates now put near $3 trillion.

Who Actually Owns the Risk?

The arrangements are not hidden from the people whose job it is to assess credit risk. According to the FT, credit-rating agencies explicitly include the guarantees in their modeling of the companies involved. The disclosure exists; it is simply far less visible than balance-sheet debt, and it concentrates correlated risk in a single scenario: a sharp fall in the value of used AI chips and data-center capacity.

That scenario is not hypothetical. The entire structure depends on GPUs holding their value through successive hardware generations. If chip resale values collapse — because demand disappoints, or newer hardware obsoletes existing fleets — the guarantees would convert from footnote entries into real cash obligations, potentially all at once.

The reporting has already stirred debate among analysts and investors. A Hacker News discussion of the FT article drew hundreds of comments, with readers arguing over whether the structure is prudent capital recycling or a systemic risk that echoes earlier eras of off-balance-sheet financing. Defenders note that guarantees are a standard tool of project finance and that rating agencies see the full picture. Critics counter that the AI buildout is now so large and so interconnected that a downturn in chip values would propagate quickly through vehicles that were designed, in part, to keep that risk at arm's length.

A Pattern That Keeps Growing

FT 조사는 보증 지원 AI 금융 뉴스가 꾸준히 나오는 가운데 시작되었습니다. AI Buzz Wire가 이전에 다룬 것처럼 이전 버전의 약속 규모가 최대 2,500억 달러로 조정된 후 Nvidia 자체는 OpenAI 계약과 관련된 데이터 센터 보증을 1,200억 달러 미만으로 축소한 것으로 알려졌습니다. 칩 공급 약정, 데이터 센터 합작 투자, 네오클라우드 용량 계약 등 각각의 새로운 대규모 거래는 투자자가 관찰하는 전통적인 부채 총액 외에 또 다른 조건부 노출 계층을 추가합니다.

현재 자본 시장은 여전히 지지적입니다. 보증 구조는 금융 비용을 낮추고 건설이 기록적인 속도로 계속 진행되도록 합니다. FT가 제기하는 질문은 AI 수요가 흔들리면 해당 수학에 어떤 일이 발생하느냐는 것입니다. 최대 3000억 달러의 보증에 따르면 답은 더 이상 각주에만 국한되지 않습니다.