Jul. 7 at 4:22 PM
SemiAnalysis ests cumulative AI capital spending (across IT & datacenters) will hit roughly
$11.1T from 2024 to 2029
As this arms race shifts from cash reserves to debt-financed expansion, outstanding AI-related debt is projected to reach about
$7.1T by 2029
Goldman Sachs ests about
$7.6T of global AI infrastructure investment from 2026 to 2031 across compute, data centers & power
McKinsey ests that data centers will require
$6.7T worldwide by 2030, w/
$5.2T tied to AI workloads alone
Morgan Stanley ests a cumulative
$4.8T–
$5.2T from 2024 to 2029, tracking just the top 5 hyperscalers (capturing ~45% of SemiAnalysis’s total global footprint)
Bank of America ests a cumulative
$5.9T–
$6.4T from 2024 to 2029, but looks strictly at traditional on-balance-sheet corporate Cloud + AI CapEx-
--- Note: Each f/cast counts the AI buildout differently, so the totals are not directly interchangeable
$NVDA acts as a pseudo-"central bank" by offering a 6-year minimum GPU rental rev guarantee program. This structure allows "neocloud" operators to secure massive credit without prime corporate credit ratings
--- This structure serves as Nvidia's structural hedge against legacy hyperscalers who are aggressively designing internal custom ASIC chips. By underwriting these neocloud balance sheets,Nvidia secures downstream platform economics. Nvidia is building a "neocloud proxy army" to protect itself
It wouldn't be a surprise if Nvidia eventually (10+ years) consolidates these dependent networks to step forward as a dominant first-party hyperscaler itself
Note: When Nvidia previously tried to build DGX Cloud as a direct cloud service, it caused severe friction w/ AWS & Azure. This friction forced Nvidia to fold DGX Cloud into internal R&D operations. Hyperscalers acct for 45%-50% of Nvidia’s total corporate revs currently
$DRAM $AMZN $GOOGL $ARTY