Goldman Sachs expects nearly $1 trillion in new artificial intelligence-related debt in 2027 as spending on chips, computing infrastructure and data centres increases. The estimate combines projected AI project financing of $545 billion with approximately $440 billion in borrowing by the five largest hyperscalers, according to the research figures cited in the supplied update.
The bank raised its estimate for AI project financing in 2027 to $545 billion from an earlier projection of $300 billion. Around $320 billion of the projected amount is expected to go towards chips. The increase highlights the growing financing requirements associated with the expansion of artificial intelligence infrastructure.
Goldman Sachs also estimated AI project financing at approximately $150 billion in 2026. The 2027 projection is therefore more than three times the 2026 estimate. Chip financing in 2027 is expected to be almost seven times the 2026 level, according to the supplied figures.
AI project financing: 2026 vs 2027
Goldman Sachs: AI Project Financing
Caption: Goldman Sachs expects a sharp increase in financing for AI projects, with chips accounting for a large share.
Chips and hyperscaler borrowing drive financing needs
The five largest hyperscalers are expected to borrow around $440 billion in 2027. Hyperscalers are large technology companies that operate extensive cloud and computing infrastructure. Their borrowing is being considered alongside the financing required for dedicated AI projects.
The supplied update also cited approximately $295 billion in hyperscaler bonds already issued. The distinction between existing bond issuance and projected future borrowing is important: the $295 billion figure describes reported issuance, while the $440 billion figure is an estimate for 2027 borrowing by the five largest hyperscalers.
Goldman Sachs’ report, titled “The Credit Line: From Shell to Compute: An Update on AI Project Finance”, examined the financing structures supporting AI infrastructure. The report discussed arrangements involving guarantees and other forms of credit support.
Nvidia’s guarantee commitments were estimated at approximately $118 billion, subject to a 25% cap, according to the supplied material. Broadcom was reported to have guaranteed most of the $35 billion financing associated with Anthropic.
Projected AI-related financing in 2027
| Category | Estimated amount |
|---|---|
| AI project financing | $545 billion |
| Of which: chip financing | $320 billion |
| Borrowing by five largest hyperscalers | $440 billion |
| Combined AI-related debt estimate | Nearly $1 trillion |
Caption: The combined estimate adds AI project financing and borrowing by the five largest hyperscalers. Chip financing is included within project financing and is not an additional amount.
Financing costs and credit risks increase
The expansion of AI infrastructure borrowing is taking place alongside higher financing costs. High-yield bonds issued by data-centre companies were reported to offer yields of approximately 8% to 10%, compared with 6% to 8% in the earlier comparison supplied.
Comparison compiled from figures stated in this article.
Higher borrowing costs can increase the expense of financing data centres, chips and other computing infrastructure. The use of guarantees and financing structures can help support funding arrangements, but it also creates obligations that investors need to understand.
MarketWatch cited the Goldman Sachs analysis in the supplied update. The figures are projections and reported financing arrangements, not confirmation that every forecast borrowing transaction will take place.
The expected increase in debt raises questions about how AI infrastructure investment will be financed and how the associated costs will be distributed among technology companies, project developers and lenders. Investors will be watching whether projected financing needs translate into actual borrowing and whether higher yields affect project economics.
What It Means
AI investment is increasingly connected to debt markets rather than being funded only through operating cash flows or equity. A larger financing requirement could create opportunities for lenders, bond investors and infrastructure developers, but it also raises the cost of building computing capacity. Data-centre bond yields of 8% to 10% show that some financing is already expensive. Investors will need to distinguish between announced investment plans, projected borrowing and debt that has actually been issued. The scale of the expected financing also makes credit guarantees and other funding arrangements important parts of the AI investment cycle.
What to Watch
- Whether Goldman Sachs’ revised $545 billion project-financing estimate changes further.
- Actual borrowing by the five largest hyperscalers.
- Financing allocated to chips and data-centre infrastructure.
- Yields on high-yield data-centre bonds.
- Further disclosures on Nvidia and Broadcom guarantee commitments.
Market Impact
- 2027 AI project financing: $545 billion — Higher projected funding needs for AI infrastructure.
- Chip financing: $320 billion — A major component of expected AI project funding.
- Hyperscaler borrowing: $440 billion — Adds to expected technology-sector debt issuance.
- Combined AI-related debt: Nearly $1 trillion — Shows the potential scale of financing activity.
- High-yield data-centre bond yields: 8%–10% — Indicates elevated financing costs.
- Nvidia guarantee commitments: Approximately $118 billion — Highlights the role of credit support in AI financing.
How much AI-related debt could be raised in 2027?
Goldman Sachs expects nearly $1 trillion in new AI-related debt in 2027. Its estimate combines $545 billion in AI project financing with approximately $440 billion in borrowing by the five largest hyperscalers. These are projections, not confirmed future borrowing transactions.
How much will AI chip financing reach in 2027?
Approximately $320 billion of the projected $545 billion in AI project financing is expected to go towards chips. Goldman Sachs also estimated that chip financing in 2027 could be almost seven times the 2026 level, according to the supplied figures.
Why are hyperscalers borrowing more?
The supplied analysis links borrowing needs to the expansion of artificial intelligence infrastructure, including chips, data centres and computing capacity. The five largest hyperscalers are expected to borrow around $440 billion in 2027, although the figure remains a projection.
What are the financing costs for data-centre bonds?
The supplied update cited yields of approximately 8% to 10% for high-yield data-centre bonds, compared with 6% to 8% in the earlier comparison. These yields indicate the financing costs associated with the reported bond category.
What role do Nvidia and Broadcom play in AI financing?
The supplied report cited approximately $118 billion in Nvidia guarantee commitments, subject to a 25% cap. Broadcom was reported to have guaranteed most of the $35 billion financing associated with Anthropic. These arrangements are examples of credit support within AI financing structures.







