AI capex surge ($49B H1, $100-120B FY); Japan $10B commit, expansions, ESG/talent risks
Key Questions
What is the scale of AI-related capital expenditures (capex) by hyperscalers in recent periods?
Hyperscalers reported Q2 capex of $37.5B, up 66%, and H1 total of $49B. Projections show $726B for 2025 (+57%) and over $1T in 2026, with Microsoft at +76% growth and AWS holding 32% share targeting $200B.
How much is Microsoft investing in Japan for AI infrastructure?
Microsoft is committing $10B (1.6 trillion yen) from 2026-2029 for Japan's AI future, partnering with SoftBank on Sakura GPUs, cybersecurity ties, and aiming for 1M AI talent by 2030.
What expansions are hyperscalers pursuing in Asia-Pacific?
Microsoft is investing $5.5B in Singapore, including free Copilot for 200K students, with additional expansions in Thailand and Australia. These support regional AI infrastructure growth.
What are the main ESG risks for hyperscalers' AI data centers?
Investors are pressing Amazon, Microsoft, and Google on high water and power usage in US data centers, alongside memory costs and carbon deals. Shareholder scrutiny focuses on sustainability amid AI scale-up.
How has Azure performed in terms of growth and traffic?
Azure grew +39% with 58% traffic increase, emphasizing resilience patterns for AI scale via features like those discussed by Mark Russinovich. This contrasts with outage risks from talent shortages.
What talent challenges are affecting hyperscalers like Microsoft and Amazon?
US H-1B visa crisis has led to 9K layoffs, hiring freezes, and slowdowns, with juniors overloaded and ex-engineers claiming Azure relies on manual fixes post-talent exodus. This raises stability risks versus AWS and GCP.
How is the AI capex surge impacting cloud margins and stock performance?
Capex surge of +66% is compressing margins to 68%, sparking debates on overhang versus cloud growth. Mag7 rotation pressures have hit stocks, with Evercore noting over 40% capex intensity.
What do backlogs indicate about AI infrastructure demand?
Hyperscaler backlogs show strong demand for AI infrastructure, with data center capex hitting record highs and growing 57% last year per Dell'Oro Group.
Hyperscalers $726B 2025 (+57%), >$1T 2026 (MSFT +76%, AWS 32% share/$200B); Q2 $37.5B +66%/H1 $49B; Japan $10B 2026-29 (SoftBank/Sakura GPUs, cyber ties, 1M talent by 2030); Singapore/Thailand/AU; Azure +39%/58% traffic; water/power scrutiny/shareholder push/memory costs/carbon deals; resilience patterns for AI scale; H1B crisis/layoffs/slowdown/juniors overload/outages risk stability vs AWS/GCP; 9k cuts/freezes; Evercore >40%; capex overhang vs cloud growth debates (+66% surge compresses margins to 68%); Mag7 rotation pressures.