Big Tech companies, including Google, Amazon, Microsoft, and Meta, are facing scrutiny over their substantial AI expenditures, which are projected to exceed $700 billion this year. Investors are keen to see tangible results and returns on these investments, rather than just continued spending. The challenge is amplified by increasing costs for AI infrastructure, with memory chip prices surging and expenses rising for power equipment, construction materials, skilled labor, and electricity connections. For example, Morgan Stanley estimates the cost of building one gigawatt of AI capacity has increased by approximately 20%, with one Nvidia-based setup climbing from $29 billion to $35 billion, and a newer version from $41 billion to $49 billion.
Adding to the financial pressure, the five largest builders of AI data centers—Alphabet, Amazon, Meta Platforms, Microsoft, and Oracle—have collectively doubled their debt load by some $350 billion over the past five years to finance this spending spree. KB Securities projects that the capital expenditures (CAPEX) for these five hyperscalers will reach $725.1 billion in 2026, a 4.9-fold increase from $148.1 billion in 2023. This expansion is driven by growing demand for agentic AI and bottlenecks in memory, power infrastructure, liquid cooling systems, and networking equipment, which are pushing up per-unit costs. For Microsoft, GPU expenses account for 60-70% of CAPEX, and memory makes up 13%.
While analysts from KB Securities and Samsung Securities believe that the AI investment expansion will continue through 2027, with CAPEX-to-revenue ratios peaking around 38% in 2027 before easing, concerns are rising about the shift from equity-like funding to debt-based financing, such as corporate bond issuance. Hyperscalers' corporate bond issuance has rapidly increased, reaching $194.1 billion by July this year, up from $20.1 billion in 2024 and $108.4 billion in 2025. This growing debt raises questions about interest burdens and potential credit risk, though most analysts suggest that credit concerns for companies like Microsoft, Alphabet, Meta, and Amazon are premature, given their stable debt-to-EBITDA ratios (e.g., Microsoft and Alphabet at 0.6 times).
Oracle stands out with a higher total debt-to-EBITDA ratio of 5.6 times as of March 2026, and its corporate bond issuance of $25 billion by July this year has caused its credit default swap premium to jump, signaling market concern. Despite these issues, financial analysts estimate that most hyperscalers have at least two to four years of funding capacity through bond issuance before credit risk materializes. The expectation is that AI-related revenues, currently growing at around 30%, will eventually help normalize CAPEX pressures as earlier investments begin to generate revenue, allowing for a gradual easing of the investment cycle post-2027.
