The Largest Infrastructure Bet in Tech History
Amazon ($200B), Alphabet ($175-185B), Meta ($115-135B), Microsoft ($120B+), Oracle ($50B). Add the numbers and the conclusion is staggering: the five largest cloud and AI companies will pour nearly $700 billion into capital expenditure in 2026 alone. That figure is six times what hyperscalers spent in 2022 and represents the largest single-year infrastructure investment surge in the history of technology.
Moody’s Ratings has taken notice, warning that investors are growing uneasy about the risk of overbuild and weak returns. Goldman Sachs projects total hyperscaler capex from 2025-2027 will reach $1.15 trillion — more than double the $477 billion spent from 2022 through 2024. The question every technology leader and investor must confront: is this a justified supercycle, or an infrastructure bubble inflating in real time?
The Spending Breakdown
Amazon Web Services leads with an estimated $200 billion, the bulk directed at expanding its global data center footprint and securing GPU capacity. Alphabet follows at $175-185 billion, aggressively scaling custom TPU infrastructure. Microsoft tracks toward $120 billion or more, fueled by Azure AI demand and its OpenAI partnership. Meta has guided $115-135 billion for its Llama ecosystem and AI-powered advertising.
Oracle is the most dramatic transformation story. The company plans to spend roughly $50 billion on AI infrastructure, raised through debt and equity — starkly contrasting with the sub-$2 billion annual cost of its legacy database business. Oracle has secured major contracts, including a reported $30 billion annual deal with OpenAI.
Approximately 75% of aggregate capex will fund AI-related infrastructure, representing roughly $450 billion in AI-specific spending. To finance this, Big Tech has issued over $100 billion in bonds so far in 2026. Moody’s also flagged $662 billion in off-balance-sheet data center lease commitments among the top five hyperscalers — hidden financial exposure that does not appear as current liabilities.
Advertisement
The Bull Case: Supply Cannot Meet Demand
Hyperscalers are not spending blindly. Every major cloud provider reports supply-constrained markets, not demand-constrained ones. Microsoft disclosed an $80 billion backlog of Azure orders that cannot be fulfilled due to power constraints alone.
Goldman Sachs estimates AI-related investments will contribute nearly 40% of total US real GDP growth throughout 2026. The firm calculates an $8 trillion present-discounted value for capital revenue unlocked by AI productivity gains. In this view, $700 billion is not extravagance — it is the minimum ante to capture a generational economic shift.
There is also a competitive argument. Every hyperscaler perceives underinvestment as an existential threat. If one scales AI capacity and competitors do not, enterprise customers migrate. Unlike the dotcom era — where startups burned cash on speculative models — today’s spending comes from profitable incumbents with strong balance sheets. The infrastructure being built is largely redeployable even if specific AI workloads shift.
The Bear Case: Returns Are Not Materializing
AI-related services generated only about $25 billion in revenue in 2025 — roughly 10% of infrastructure spending that year. Only about 25% of enterprise AI initiatives have delivered their expected ROI, and fewer than 20% have scaled across entire organizations.
Moody’s warns that massive capex is eroding historically strong free cash flow and driving higher borrowing, flagging a potential “reassessment of creditworthiness” if profit growth fails to materialize. Goldman Sachs CEO David Solomon acknowledged he expects “a lot of capital that was deployed that doesn’t deliver returns.”
Power constraints add another layer of risk. Lack of readily available electricity will constrain AI capacity through at least 2027, meaning some infrastructure being built today may sit partially idle waiting for grid connections. Data center power demand is forecast to increase 50% by 2027, from 55 GW to 84 GW.
History offers cautionary parallels. The telecom industry invested over $500 billion in fiber optic networks in the late 1990s. When the bubble burst, most capacity sat dark for years. The infrastructure eventually proved valuable — but the companies that built it often did not survive to benefit.
2026 as the Proof Point
Goldman Sachs frames 2026 as the critical Phase 3 transition year — where AI-enabled revenue models must prove their worth. The first two phases (chip buildout, then infrastructure deployment) were capital-intensive by design. Phase 3 demands that enterprises convert AI capabilities into measurable productivity gains and new revenue.
The most likely outcome sits between extremes. Demand for AI compute is real and growing, but not every dollar will earn a return. Some data centers will be overbuilt in regions where power never arrives on schedule. Some GPU clusters will depreciate before workloads justify their cost. The winners will be hyperscalers with the most efficient capital allocation — not necessarily the ones who spend the most.
Frequently Asked Questions
Why are hyperscalers spending $700 billion on AI infrastructure in 2026?
All major cloud providers report that demand for AI compute significantly exceeds supply. Microsoft alone has an $80 billion backlog of unfilled Azure orders due to power constraints. Hyperscalers view underinvestment as an existential competitive risk — if one scales capacity and competitors do not, enterprise customers migrate. Goldman Sachs projects AI investments will drive 40% of US GDP growth in 2026, framing the spending as capturing a generational economic shift.
What is the overbuild risk that Moody’s warns about?
Moody’s warns that the $700 billion capex surge is eroding historically strong free cash flow and driving unprecedented borrowing. AI-related services generated only $25 billion in revenue in 2025 — roughly 10% of infrastructure spending. The agency also flagged $662 billion in off-balance-sheet data center lease commitments. If enterprise AI revenue fails to grow at the pace capex implies, hyperscalers face margin compression, stranded assets, and potential credit downgrades.
How does 2026’s AI spending compare to the dotcom bubble?
The scale is larger, but the structure differs fundamentally. Dotcom-era spending came from cash-burning startups with unproven models. Today’s $700 billion comes from profitable incumbents with strong existing revenue. The infrastructure is largely redeployable. However, the telecom fiber overbuild of the late 1990s ($500B+ invested, most sat dark for years) offers a closer cautionary parallel — the infrastructure proved valuable eventually, but many companies that built it did not survive.
Sources & Further Reading
- US Hyperscaler Capex to Top $700B in 2026 — Moody’s via Data Center Dynamics
- Why AI Companies May Invest More Than $500 Billion in 2026 — Goldman Sachs
- Moody’s Flags $662B Off-Balance-Sheet Risk in Data Center Buildout — Fortune
- AI Capex 2026: The $690B Infrastructure Sprint — Futurum Group
- Tech AI Spending Approaches $700B, Cash Taking Big Hit — CNBC
- Oracle Eyes $50B for AI Infrastructure in 2026 — Data Center Knowledge
- Don’t Fear the AI Bubble, It’s About to Unlock $8 Trillion — Fortune / Goldman Sachs
