Morgan Stanley: AI Capex Could Reach $10-15 Trillion Over the Next Decade
nashnova research
Morgan Stanley CEO Ted Pick projects $10-15 trillion in AI capital spending over the next decade, but PIMCO warns the financing load will eventually become excessive — Wall Street is splitting over whether this buildout is worth the bill.
How big is $10-15 trillion, really?
Morgan Stanley CEO Ted Pick estimates top hyperscalers could deploy $800 billion to $1 trillion per year on data centers and compute infrastructure over the next 7-12 years.
This means → a handful of tech giants would spend roughly the GDP of a mid-sized country on AI infrastructure every single year.
Pick put it in baseball terms: "We're probably in the second inning of a nine-inning game." In plain terms = the buildout is nowhere near its peak — most of the money has yet to be spent.
Does the math hold up?
Pick argues that $10-15 trillion, while enormous, looks reasonable against the S&P 500's total market cap of roughly $70 trillion.
Some $2-3 trillion has already been committed or deployed — this is not all projection.
He estimates AI-driven productivity gains could add 50-70 basis points to U.S. real GDP growth annually. This means → Morgan Stanley's case is that the spending pays for itself through economy-wide efficiency gains.
Why is PIMCO pushing back?
PIMCO — one of the world's largest bond fund managers — CEO Emmanuel Roman warns AI-related financing could swell to roughly $10 trillion over five years and eventually become "too large."
His math: AI capex financing could generate about $2 trillion in annual funding needs, against a U.S. bond market of roughly $12 trillion. In plain terms = one industry's borrowing would consume a sixth of the entire U.S. bond market.
Roman said bluntly that "sooner or later the tide goes out" — and rising U.S. rates, Middle East tensions, or a eurozone crisis could be the trigger.
State Street's view — bubble or real demand?
State Street CEO Ron O'Hanley says it is "hard" to call AI a bubble right now, because the data centers being built have real orders and real corporate customers behind them.
But he flagged the real risk: if AI becomes efficient enough to need far less compute, the capacity being built today could turn into overcapacity.
This reflects a deeper split on Wall Street — not over whether AI works, but over whether the current pace of building will actually be absorbed by real-world usage.
Is $10-15 trillion in AI spending over a decade justified?
BULL
Returns are quantifiable
Morgan Stanley estimates AI adds 50-70 bps to U.S. GDP annually — the spending has a payback case.
Demand is real
State Street notes data centers have actual orders — this is not the dot-com era.
Scale is proportionate
$10-15 trillion sits against $70 trillion in S&P 500 market cap — large but not absurd.
BEAR
Bond market gets crowded
PIMCO calculates AI financing would take up one-sixth of the U.S. bond market.
Overcapacity is the real risk
If AI efficiency leaps forward, today's compute buildout may never be fully utilized.
External shocks lurk
Rising rates, geopolitics, or a eurozone crisis could pull the rug at any time.
In plain terms = the two sides aren't arguing over whether AI works — they're arguing over whether the current spending pace is running too far ahead of actual demand. The answer hinges on how fast AI efficiency improves and how much the bond market can absorb, and right now, nobody knows for sure.
市场有风险，内容仅供研究参考，不构成投资建议。