AI infrastructure investment by 4 US hyperscalers is expected to expand to approximately ¥113 trillion annually in 2026. Attention is also focusing on how the funds are being raised and the risks to the market.
The 4 companies involved in AI infrastructure investment, Alphabet (Google), Amazon, Meta and Microsoft, are expected to revise their 2026 capex plans, bringing their combined spending to $720 billion–$745 billion. This is on a scale approaching Japan's general account budget for a single year, while their order backlogs are also enormous. The combined order backlog at Azure, Google Cloud and AWS alone totals $1.688 trillion, underscoring the scale of overall demand.
A variety of methods are being used to raise the necessary funds. Meta is increasing capex while reducing cash and maintaining its dividend, and its corporate bond balance is rising sharply. Investment in the Hyperion data center is being made through an investment fund. Although Meta's stake is limited to 20%, the agreement includes a provision requiring Meta to pay the difference if the value falls below a specified level. Moody's has warned of the associated debt risks.
There are also support agreements involving companies such as NVIDIA, in addition to the 4 major companies. The circular investment structure surrounding OpenAI is making it difficult to assess returns on investment. NVIDIA is supporting emerging cloud companies through partnerships with CoreWeave, Sharon AI and Firmus Technologies, among others, while some have pointed out that CoreWeave's debt is close to junk-bond status. Japanese companies will need to monitor developments both in Japan and overseas closely.
At the same time, the effects of the funding cycle are emerging as a risk factor. As the funds supporting AI investment circulate within a closed environment, the correlation between actual demand and investment could become unclear. AWS's review of GPU reservation pricing could also affect investment trends, and the outlook remains uncertain.