Citi: Bond Market Drives the Narrative — AI Infra Financing Under Pressure but Demand Remains Strong
nashnova research
Citi software co-head Tyler Radke says the 10-year Treasury yield is up ~30% from its March low, squeezing valuations of leveraged tech names — yet underlying AI infrastructure demand sits at record highs, making this a financing problem, not a demand problem.
Why is the bond market calling the shots for tech?
The 10-year U.S. Treasury yield has risen ~30% from its March low, the single biggest driver of recent market volatility.
This means → sub-investment-grade tech firms and neocloud providers — smaller platforms that rent out GPU compute — are hit hardest. Valuation recovery is hostage to rate direction.
Radke put it bluntly: "It's hard to fight this, especially ahead of the Fed meeting and the midterms."
What is the real bottleneck for AI infra — funding or demand?
Radke's verdict: it is fundamentally a financing issue, not a demand issue. Nvidia H100 GPU lease spot prices sit at record highs; pre-IPO AI labs show strong business activity.
"These companies are locking in contracts at the highest prices they've ever seen. Margins are higher." In plain terms = downstream clients are scrambling to sign, at prices above previous levels — real demand has not weakened.
He expects the pressure to ease over time: higher financing costs will ultimately be passed through into client contracts, shifting the burden downstream.
Infrastructure bottlenecks — who actually benefits?
Power supply limits, grid capacity, and regulatory hurdles increase the scarcity value of resources already held by existing neocloud providers.
This means → if building new data centers gets harder, platforms that are already built and can deliver compute immediately become more valuable — tighter supply favors incumbents.
This reflects a counterintuitive dynamic: near-term construction barriers act as a natural moat for players who have already secured their position.
Where does Citi have the highest conviction?
Radke's highest-conviction bet is on pure-play software companies, not capital-intensive hardware or neocloud names.
Microsoft and Palantir carry "the best margin profiles in the world" and remain free-cash-flow positive despite heavy capex.
In plain terms = software stocks remove the need to judge GPU depreciation or financing trajectories — earnings visibility is far higher than in hardware, making this Citi's top-conviction investment theme.
What is the key variable that validates this call?
One core variable: whether financing costs can pass through smoothly into contract pricing.
If pass-through works and demand holds, Radke's "financing problem, not demand problem" thesis stands.
If clients refuse price increases and contract signings slow, financing pressure will feed back into demand — and the nature of the problem changes.
市场有风险，内容仅供研究参考，不构成投资建议。