Daniel and Nova look at how the current AI buildout is actually
financed, going beyond headline capex numbers from Alphabet,
Amazon, Meta and Microsoft to trace where the money really comes
from: company cash flow, bond markets, and private credit and
special purpose vehicles that do not appear on any balance sheet.
They use IMF and OECD financing data, a detailed look at Meta's
Louisiana data center deal with Blue Owl, and the circular
revenue flows between Nvidia, the hyperscalers, OpenAI and
Anthropic to show how much of the reported growth rests on
companies investing in and buying from each other.
The second half turns to what this means for practitioners and
investors. They break down why API token prices and subscription
prices diverge so sharply at OpenAI and Anthropic, why agentic
tools like Claude Code, GitHub Copilot and OpenClaw forced a wave
of pricing changes in June, and how KPMG data shows most
companies have no real visibility into their AI spending. They
also cover the rise of cheaper Chinese open-weight models such as
DeepSeek, Kimi, Qwen and GLM, the risk of export control driven
model outages, and how automatic index inclusion means anyone
with an MSCI World or Nasdaq 100 savings plan is already exposed
to AI bets like SpaceX and a possible Anthropic IPO.
Key topics:
- How the roughly 650 billion dollars in 2026 hyperscaler data
center spending from Alphabet, Amazon, Meta and Microsoft is
actually funded, based on IMF and OECD financing breakdowns
- The rise of private credit and off balance sheet special
purpose vehicles, illustrated by Meta's Louisiana data center,
financed mostly by Blue Owl
- The circular deal structure between Nvidia, the hyperscalers,
OpenAI and Anthropic, and what share of hyperscaler order
backlogs comes from those two companies
- Why Amazon's reported profit relied heavily on unrealized gains
from its Anthropic stake while free cash flow collapsed
- The gap between profitable API token pricing and loss making
subscription plans at OpenAI and Anthropic, and the June pricing
changes at Anthropic, GitHub Copilot and Copilot Cowork
- Real world cost blowouts from agentic AI usage at Uber and
Meta, and survey data showing most companies lack cost visibility
into their AI spending
- The growing competitiveness of Chinese open-weight models like
DeepSeek, Kimi, Qwen and GLM, their pricing advantage, and the
state subsidies and cheap power behind them
- The risk of export control related access outages for cloud
hosted models versus locally run open weight models
- How automatic inclusion in indexes like MSCI World, FTSE
All-World and the Nasdaq 100 exposes ordinary retirement savers
to AI bets such as SpaceX and a possible Anthropic IPO
Key takeaway: AI technology works and is likely here to stay, but
that does not guarantee the current wave of investors gets paid
back. Daniel argues IT leaders should budget for token costs
several times higher than today's market share pricing, treat
local compute as insurance against price hikes and access cutoffs
rather than a pure cost saving, and recognize that most retail
investors are already exposed to this bet through ordinary index
fund savings plans, whether they know it or not.
Kommentare (0)
Melde dich an, um einen Kommentar zu schreiben.