TMTPOST -- Amazon.com Inc. earlier this week raised 15billion in its first U.S. dollar bond offering in three years, joining a wave of Big Tech's tapping debt markets to fund artificial intelligence (AI) infrastructure.

AI Generated Image

The new Amazon bond sale was to attracted about 80 billion in demand at its peak, Bloomberg reported on Monday. The deal will reportedly finance acquisitions, capital expenditures and share buybacks as the e-commerce and cloud computing leader races to expand its data center capacity.

The bond sale, which exceeded initial estimates of $12 billion, comes amid an aggressive AI buildout across the tech sector. According to Bloomberg, the issuance spree from technology companies has helped push global debt issuance to a record of more than $6 trillion this year.

Amazon sold notes in six parts, according to Bloomberg. Pricing on the longest portion, a 40-year bond, tightened to 0.85 percentage point above Treasuries from 1.15 percentage point initially. Orders were cut in half as borrowing costs fell during the sale process. Goldman Sachs Group Inc., JPMorgan Chase and Morgan Stanley managed the transaction.

JPMorgan Chase expects the wave of spending to finance AI investments to drive issuance in the U.S. high-grade market to a record $1.81 trillion next year. JPMorgan noted this is an "opportune" time for Amazon, which has primarily relied on its own cash flow, to incorporate debt into its capital structure for funding flexibility.

The proceeds will support business investments, fund future capital expenditures and repay upcoming debt maturities, Amazon said. The company last tapped the U.S. investment-grade market in November 2022 when it raised $8.25 billion.

Amazon's bond sale follows similar moves by major technology firms seeking to finance AI infrastructure. Google parent Alphabet Inc. earlier this month sold $25 billion of debt in the U.S. and Europe. Meta Platforms Inc. issued30 billion of corporate bonds last month, the biggest offering of the year, while Oracle Corp. raised $18 billion through investment-grade notes in September.

Amazon is the world's largest seller of rented computing, critical to powering artificial intelligence systems. The company has been investing heavily in data centers and chips to build and run AI models capable of generating text or images and automating processes.

Amazon in late October said it made capital expenditures (Capex) of nearly $90 billion in the first nine months of the year and expected to spend $125 billion for the full year. Earlier this month, the company's cloud unit signed a $38 billion deal to supply OpenAI access to hundreds of thousands of Nvidia Corp. graphics processing units (GPUs) as part of a seven-year computing agreement.

On October 29, Amazon said one of its largest AI data center investments, Project Rainier, was fully operational. The $11 billion facility is intended to train and run models from Anthropic, in which Amazon has invested $8 billion. Amazon said the startup will use 1 million of its custom Trainium2 chips by the end of 2025.

Amazon's Capex are expected to top $147 billion next year—roughly three times the level seen in 2023—according to analyst estimates compiled by Bloomberg. The company's data center power capacity has doubled since 2022, and CEO  Andy Jassy expects it to double again by 2027.

Amazon reported on October 30 that sales for the third quarter rose 13% year-over-year (YoY) to $180 billion, while net profit soared 39% YoY to $180 billion. Revenue from Amazon Web Services (AZW) climbed 20% to $33 billion for the September quarter—the fastest growth rate since 2022--exceeding analyst estimates.Operating income at AWS rose 9% YoY to $11.4 billion, accounting for about two-thirds of Amazon's total operating profit.

Amazon expected fourth-quarter sales between $206 billion and $213 billion, and operating profit between $21billion and $26 billion as it enters the busy holiday shopping season.

Jassy told analysts that customers were buying cloud-computing and AI services as fast as Amazon could make them available given capacity constraints. "Today overall in the industry, maybe the bottleneck is power," Jassy said. "I think at some point it might move to chips."