Amazon is borrowing money in Canadian dollars because the AI arms race costs more than even Amazon's cash flow can handle.
The Summary
- Amazon just raised C$14 billion ($10 billion) in Canadian dollar bonds, the largest such offering on record, and followed it with $17.5 billion in bank loans.
- The back-to-back debt raises signal that US hyperscalers are looking beyond their home market to fund AI infrastructure, a sign that domestic capital markets alone can't keep pace with spending velocity.
- Debt is climbing across the industry as companies burn cash to stay competitive in AI, raising questions about how long this leverage cycle can sustain itself.
The Signal
Amazon just pulled off a double capital raise that tells you everything about the current state of AI spending. First, a record C$14 billion bond sale in Canadian dollars. Then, immediately after, $17.5 billion in bank loans. That's $27.5 billion in fresh debt in a matter of days. For context, Amazon generated $143 billion in operating cash flow last year. They don't need to borrow. They're choosing to.
The Canadian bond sale is the tell. When one of the world's most cash-rich companies goes to Canada for $10 billion, it's not about diversification or currency hedging. US hyperscalers are looking beyond their home market because they need capital faster than any single market can provide it. The AI buildout is happening at a pace that exceeds even the deepest pools of available credit.
"US hyperscalers are looking beyond their home market to fund artificial intelligence spending."
This is what the agent economy looks like at the infrastructure layer. Not elegant. Not efficient. Just relentless capital deployment to build the compute layer that everything else will run on. Amazon Web Services isn't just competing with Microsoft and Google for cloud share anymore. It's competing for the right to host the next generation of autonomous agents that will handle customer service, logistics routing, inventory management, and a thousand other tasks currently done by people or clunky software.
The debt strategy is rational but risky. Borrow cheap, build fast, lock in the infrastructure advantage before the window closes. But debt is climbing across every hyperscaler. The playbook assumes that AI workloads will generate returns that justify the leverage. If they don't, or if they take longer to materialize than the market expects, someone's going to get caught holding very expensive data centers running very expensive chips that aren't generating enough revenue to service the debt.
Key risks in this leverage cycle:
- AI demand growth could slow before infrastructure costs are recouped
- Interest rate changes could make servicing $27B+ in new debt more expensive
- Competitors are making identical bets, which means overcapacity is a real possibility
The Implication
Watch the debt-to-infrastructure ratio across Amazon, Microsoft, and Google over the next two quarters. If they keep raising at this pace without corresponding revenue announcements from AI services, it means they're still in land-grab mode and no one has proven the business model yet. That's not necessarily bad, but it means we're still early, and the winners aren't decided.
For anyone building on AWS, this is good news. Amazon is betting the farm on having the best, fastest, most available compute infrastructure. If you're building agents, you're going to have options.