Nvidia is raising at least $20bn in its first corporate bond sale since June 2021, when it raised $5bn. The offering spans seven tranches with maturities ranging from two to 30 years, with the longest notes maturing in 2056. Price talk on the 30-year tranche is around 0.9 percentage points above Treasuries.
Goldman Sachs, JPMorgan and Morgan Stanley are running the deal. Proceeds will be used for general corporate purposes, including the repayment and refinancing of existing notes.
Nvidia had $13.24bn in cash and equivalents at the end of April. It is the most profitable semiconductor company in history, with data centre revenue hitting $62.3bn in its most recent quarter alone. It does not need the money.
It is borrowing it anyway.
Why borrow when you are sitting on cash
A company borrows at investment-grade rates when it believes the return on deploying that capital exceeds the cost of the debt. Nvidia's cost of borrowing, even on 30-year paper, is modest relative to the revenue growth its AI chips are generating.
The bond sale gives Nvidia financial flexibility without diluting shareholders. It preserves the cash pile for opportunistic uses, acquisitions, equity stakes in partners, or the kind of strategic investments the company has been making across the AI ecosystem, while locking in long-term capital at rates that may not be available if market conditions change.
Nvidia took a $2bn stake in Marvell three months ago. It has provided financial backing for OpenAI's proposed 10-gigawatt data centre campus with SoftBank. It is pitching Vera CPUs to Chinese customers. Each of these moves requires capital or the credible promise of it.
The tech debt wave
Nvidia is not alone. Meta filed for up to $30bn in bonds in October. Alphabet sold yen-denominated bonds for the first time last month and raised $85bn in equity. Amazon and Google have collectively borrowed hundreds of billions to fund AI infrastructure expansion.
Combined hyperscaler capex is expected to exceed $700bn this year, up from $400bn in 2025. The bond market is funding the physical infrastructure that the AI revolution requires, from chips to data centres to power generation.
What it signals
When Nvidia borrows $20bn across maturities stretching 30 years into the future, it is making a statement about the duration of the AI buildout. This is not a company funding a two-year product cycle. It is financing an infrastructure commitment that it expects to last decades.
The chip cycle is famously volatile. Nvidia is borrowing as though it is not. Either the company sees demand visibility that justifies 30-year debt, or it is locking in cheap capital while the market still believes the cycle is structural rather than cyclical.
Both readings lead to the same trade: buy the bonds while the market is willing to lend at 90 basis points over Treasuries, and deploy the capital before the window closes.
Nvidia does not need the money. It needs the optionality. Twenty billion dollars of it, stretching to 2056.
