One-Line Summary

AI infrastructure spending is being financed through multi-currency corporate bonds while unpredictable usage-based bills push companies to favor smaller, cheaper AI models.

Industry & Biz

Banks push multi-currency bonds as AI debt surges

Big tech companies are borrowing more money to build data centers and buy chips, and banks are finding new ways to sell that debt. Reuters reports AI-related borrowing is close to 15% of investment‑grade bond issuance in 2026, as Amazon and Alphabet sold about $60 billion of bonds over the past 12 months, often in non‑U.S. currencies to reach more investors. ¹

Amazon raised €14.5 billion ($16.56 billion) in March in the largest ever euro corporate bond sale, while Alphabet set borrowing records in yen, Canadian dollars, Swiss francs and sterling, and issued the first 100‑year tech bond since 1997, according to LSEG data cited by Reuters. Morgan Stanley’s Teddy Hodgson said Alphabet and Amazon diversified issuance into Europe, Canada and Asia; demand remains strong in dollars but may face limits in other currencies, and AI exposure is still a small slice of broad credit indexes. ¹

Why it matters: the AI buildout is being underwritten by global fixed‑income investors, keeping capital flowing to chips, cloud and data centers even as U.S. markets absorb heavy supply. For operators and finance teams, multi‑currency issuance signals competition for funding and possible differences in borrowing costs by currency bloc, factors that can influence vendor pricing and contract terms. ¹

Rising AI bills nudge buyers toward smaller, cheaper models

Companies that embraced AI tools are rethinking model choices as bills jump under usage‑based pricing. Despite falling token prices, Reuters reports that shifting from flat subscriptions to metered use makes spending unpredictable and often higher; executives at Microsoft, Palo Alto Networks and Coinbase say smaller, cheaper models can cover a big share of needs, and reports say Uber burned through its entire 2026 AI budget in four months before capping usage. ²

Analysts quantify the spread: Gartner estimates AI coding costs could exceed the average developer salary by 2028, and Citi says some Chinese models charge about 18 cents per million tokens versus roughly $4 for top models. For buyers, the practical metric is cost per successful task, not price per token — a shift that elevates procurement discipline and finance oversight. ²

What This Means for You

Budgets for AI are moving from “how many seats” to “how many calls.” With vendors leaning into usage‑based pricing, set hard caps, alerts, and default models by task to avoid runaway spend as intensity rises. This is the cloud‑metering playbook applied to AI assistants and APIs. ²

Model selection becomes a sourcing decision, not hype‑driven. Run a short bake‑off on a top workflow to compare accuracy and total cost to completion between a smaller, cheaper model and a premium option — especially where speed and context needs are modest. Executives and analyst estimates cited by Reuters support the value of this comparison. ²

If your product or workloads depend on cloud capacity, the bond market signals that funding for chips and data centers is flowing from investors beyond the U.S. dollar. That broad base can support capacity buildouts — and may shape pricing and contract terms differently by region and currency. ¹

Action Items

1. Run a one-hour model bake-off: Take 20 real prompts from your top workflow and compare a smaller, cheaper model vs. your current default for accuracy, latency, and total cost per completed task.
2. Set spending guardrails: Turn on usage alerts and monthly caps in your AI tools or admin consoles; assign default models by task to prevent overuse of premium options.
3. Ask vendors for predictable pricing: Request per-seat or committed-use discounts and a clear cost-per-task estimate for your key use cases before expanding rollout.
4. Build a basic cost dashboard: Export last 30 days of AI usage logs into a spreadsheet, group by use case, and calculate cost per task to spot quick savings.