高盛将2027年人工智能项目融资预测上调80%以上,并警告称融资成本将不断上升

Listen to this article

Goldman Sachs raises 2027 AI project financing forecast by over 80% to $545 billion and warns of rising financing costs

cls.cn· Nov 18 18:30

①Goldman Sachs has raised its 2027 global AI project financing forecast from $300 billion to $545 billion; ②Chip financing is expected to reach $320 billion, nearly seven times this year’s figure to date; ③Goldman Sachs advises investors to remain patient with data center credit and to exercise prudent selection, assessing each project on its specific merits.

Cailian Press, November 18 (Editor: Li Ying) — Goldman Sachs recently raised its forecast for AI project financing next year by more than 80%, with chip financing accounting for the largest share. However, some chip financings rely on vendor guarantees, and NVIDIA stated during its earnings call that it will rigorously assess future guarantees. The market is watching to see whether Goldman Sachs’s forecast can materialize.

According to media reports on Wednesday local time, Goldman Sachs’ credit strategy team released a report on Tuesday, raising its 2027 global AI project financing forecast from $300 billion to $545 billion. Of this amount, $225 billion is allocated to data center construction, and $320 billion to computing power financing for GPUs and other chips.

Note: Global AI-related project financing scale; dark color represents data center construction financing, light color represents chip/computing power financing; the figure shows Goldman Sachs’ broader global statistics.

According to Goldman Sachs, AI project financing has totaled about $150 billion so far this year, with chip financing accounting for only $48 billion. By this calculation, next year’s computing power financing will be nearly seven times this year’s level to date, and data center construction financing will roughly double.

According to reports, major cloud computing companies have issued $295 billion in bonds globally this year, surpassing Goldman Sachs’ previous full-year forecast two months ahead of schedule. Investors already hold a large volume of these bonds, and continued purchases now demand higher yields, meaning these companies’ borrowing costs are significantly higher than those of other creditworthy enterprises.

Goldman Sachs explained that highly rated companies face constrained issuance capacity in the corporate bond market, and the market has become more selective toward lower-rated borrowers. Consequently, an increasing share of AI capital expenditures will be financed through higher-margin structured financing, such as borrowing by specially established project entities repaid using project contract revenues.

There are hidden concerns about financing.

A common practice in chip financing is for a separate project entity to borrow funds to purchase chips and then lease them out, with the lender recovering principal and interest through rental payments.

Chips depreciate rapidly, and lenders are reluctant to accept them as collateral, so transactions involving weaker-quality clients often require chipmakers to provide guarantees. Goldman Sachs notes that such guarantees typically span several years, far exceeding the period over which manufacturers recognize related revenue; currently, this does not show up as leverage on the balance sheet, but rating agencies have already flagged the accumulation of related risks across multiple deals.

According to reports, after NVIDIA founder Jensen Huang stepped down as CEO on the 12th of this month, interim CEO Colette Kress stated during Tuesday’s earnings call that the company will rigorously evaluate future investments, commitments, and guarantees, maintaining discipline in capital spending while continuing to provide financing assistance or guarantees to certain highly creditworthy customers. Kress disclosed that NVIDIA’s current total guarantee amount is approximately $118 billion.

In August this year, NVIDIA teamed up with several institutions, including Goldman Sachs, to launch a $500 billion chip financing platform that provides guarantees of up to 25% for each transaction, but no transactions have been completed in the past three months.

The lenders want NVIDIA to increase the collateral ratio and cite chipmaker Broadcom as an example: in May this year, Broadcom guaranteed $30 billion of a $35 billion chip financing for AI company Anthropic.

Media reports indicate that Goldman Sachs has also acknowledged that tighter deal conditions have slowed the pace of AI project financing issuance this quarter, with pending deals piling up.

Financing for data center construction is also facing rising costs: as borrowing rates increase, tenants must pay higher rents; if tenants refuse to accept this, projects can only reduce debt and increase equity investment.

Goldman Sachs advises investors to remain patient with data center credit, choose carefully, assess on a project-by-project basis, and expects that the spreads on such bonds are more likely to widen than to narrow significantly.