Amazon Raises $25 Billion in Bonds to Fund AI Infrastructure, Joining Big Tech Debt Boom

Amazon returned to the bond market on July 7, 2026 with its largest single debt offering of the year: an eight-part benchmark transaction targeting at least $25 billion. The books opened with $62 billion in peak demand before banks trimmed spreads and the final order book settled at approximately $41 billion — roughly 1.6 times oversubscribed. The deal is the clearest data point yet on what “AI infrastructure investment” means in practice for the world’s largest cloud provider: not equity, not operating cash flow, but debt, in quantities and at a pace that would have seemed extraordinary five years ago.

The Numbers Behind the Build

Amazon has projected its 2026 capital expenditures at approximately $200 billion, up from $131 billion in 2025 — a 53% increase in a single year. The overwhelming majority of that increase goes to AI infrastructure: data centers, custom AI chips, networking, and the real estate, power, and cooling systems that the physical plants require.

To fund that build without fully depleting operating cash flow, Amazon has turned aggressively to the bond market. The July 7 offering is not a standalone event; it sits within a year of heavy debt issuance. Amazon had already raised roughly $54 billion in U.S. and European bonds earlier in 2026, plus an additional $10 billion in Canada in June. Adding the $25 billion July offering, Amazon’s year-to-date bond issuance exceeds $89 billion.

That scale puts AI infrastructure spending in the same category as the long-term industrial investments that historically characterized oil companies, railroads, and utilities — sectors where the capital requirements are so large, and the payback periods so long, that equity alone cannot fund the build. The bond market, with its appetite for investment-grade paper from a company of Amazon’s credit quality, becomes the only realistic mechanism.

AWS as the Engine

The spending rationale is straightforward: Amazon Web Services (AWS) is growing its AI-related revenue faster than the conventional cloud business that powered Amazon’s first decade of dominance. Enterprises are not just moving workloads to the cloud — they are building and deploying AI applications that consume vastly more compute per dollar of revenue than traditional databases or web hosting.

AWS’s custom AI chip program — the Trainium line for training and Inferentia line for inference — has become a strategic priority. Amazon designs these chips in-house at Annapurna Labs and manufactures them at TSMC, targeting customers who want lower per-token costs than Nvidia GPU-based inference provides. But running those chips at scale requires the same power, cooling, and networking infrastructure as any other data center build.

The $200 billion capex target also reflects AWS’s positioning in sovereign AI: government and regulated-industry customers who need data to remain within specific geographic boundaries are building private AI environments on AWS infrastructure, a workload category that requires dedicated rather than shared capacity.

The Broader Big Tech Debt Wave

Amazon is not alone. The bond market in 2026 has become the de facto mechanism by which Big Tech finances the AI arms race.

Alphabet (Google’s parent) completed a $10 billion bond offering in May, its largest in several years, directed toward Google’s AI data center expansion. Google reported a 37% surge in electricity consumption in 2026, driven almost entirely by AI compute — a figure that illustrates why “AI infrastructure” requires power plants and transmission lines, not just racks and chips.

Meta has similarly tapped the debt markets, and Nvidia — despite having the highest profit margins in the semiconductor industry — issued bonds to accelerate its own AI infrastructure investments, including supporting the buildout of Nvidia-branded AI supercomputer clusters for hyperscalers.

The pattern across all four companies reveals something important: even companies with exceptional free cash flow find that the pace and scale of AI infrastructure investment outstrips what operating cash alone can fund without constraining other priorities like R&D, acquisitions, and shareholder returns.

Who’s Buying the Bonds

The $62 billion in peak demand for Amazon’s $25 billion offering tells a story about where institutional capital is flowing. Investment-grade corporate bonds from hyperscalers are effectively low-risk proxies for AI infrastructure investment — buyers get the credit quality and liquidity of Amazon paper while indirectly financing the data centers, chips, and power systems that AI depends on.

Pension funds, insurance companies, and sovereign wealth funds that cannot directly invest in NVIDIA stock or AWS equity — either due to mandate restrictions or concentration limits — can access AI infrastructure exposure through the bond market. The heavy oversubscription of Amazon’s offering, and those of its peers, suggests substantial latent demand for exactly this kind of instrument.

The Risk Calculus

Not everyone is comfortable with the debt levels accumulating across the sector. Amazon’s total long-term debt was approximately $130 billion entering 2026; after this year’s issuances, it will approach $220 billion. For a company with AWS’s cash generation, that is manageable — AWS operating income in 2025 was approximately $40 billion — but it represents a significant bet that AI infrastructure investment will generate returns commensurate with the capital committed.

The bull case is intuitive: AI workloads will continue growing, AWS will capture a meaningful share of that growth, and the data centers being built today will generate operating leverage for a decade or more. The bear case is more unsettling: if AI adoption plateaus faster than current projections, or if a significant fraction of AI workloads consolidate on a competitor’s infrastructure, the fixed costs of the capital build become a liability rather than an asset.

Bond covenants generally do not restrict how Amazon deploys capital — the company retains full operational flexibility. But the sheer scale of the commitment creates a strategic path dependency: having borrowed $89 billion in a single year to fund AI infrastructure, Amazon is deeply committed to a world where that infrastructure pays for itself.

What $200 Billion in Capex Actually Builds

To make the numbers tangible: at approximately $7.8 million per Nvidia Vera Rubin NVL72 rack, Amazon’s incremental $69 billion capex increase in 2026 could theoretically purchase roughly 8,800 of those racks. In reality, the spending is spread across land acquisition, power infrastructure, building construction, networking, cooling, and multiple chip generations — but the arithmetic illustrates why data center spending has become the defining economic event in the semiconductor and power sectors.

Amazon’s bond offering closed without incident on July 7, with spreads tightened from initial guidance — a signal that the market remains deeply confident in AWS’s ability to monetize the infrastructure being built. For now, the AI infrastructure debt boom continues, with Amazon’s $25 billion offering unlikely to be its last word from the bond market in 2026.