We’re at the point in the AI cycle where the narrative is running ahead of the math, and everybody is quietly hoping a miracle buyer shows up to keep the GPU supercycle running.
There’s really only one buyer left on Earth with pockets deep enough to support another year of Nvidia’s current momentum:
China.
And here’s the uncomfortable part nobody wants to say out loud:
Even if China buys $20–30B worth of GPUs, it only buys Nvidia a quarter or two.
If China doesn’t buy at all, the air starts leaking immediately.
Let’s break down why.
1. U.S. hyperscalers are already hitting capacity and power limits
AWS, Google, Meta, Microsoft… they’ve all signaled the same things:
- Capex is massive, unsustainable
- Efficiency gains reduce GPU-per-model
- In-house accelerators are replacing third-party GPUs
- Power and data center constraints are hitting the wall
- ROI scrutiny is rising
They aren’t going to double GPU purchases forever.
The slope is flattening, even if the headlines pretend otherwise.
2. The ONLY untapped mega-buyer left is China
And that window is closing fast.
Why?
- Huawei Ascend is ramping
- RISC-V investments accelerating
- Domestic AI stacks maturing
- Partnership with Google emerging
- Export restrictions killed trust
- Beijing wants self-reliance, not deeper dependence
China will buy something, maybe $5B, maybe $20B … but no scenario exists where China commits the $100B+ needed to keep the U.S. AI bubble inflated.
China buying chips is a Band-Aid, not a lifeline.
The bubble stays inflated only if:
- U.S. hyperscalers keep expanding
- Cloud companies keep buying aggressively
- Startups keep raising billions
- Model sizes keep increasing
- Inference costs stay high
- AI adoption accelerates
- Margins stay sky-high
- Competitors (TPU, Trainium, MTIA, Huawei) don’t steal share
But right now?
NONE of that is guaranteed.
Google has TPUs.
Amazon has Trainium.
Meta has MTIA.
Microsoft has Maia.
China has Ascend + RISC-V.
OpenAI is working on efficiency over scale.
There’s no new explosive market left to support infinite GPU demand.
And it’s not strategically rational for China to be Nvidia’s savior.
3. The math is brutal
Nvidia’s data center revenue run-rate:
~$120B per year
Global AI capex:
~$400B per year
A Chinese purchase of $20–30B sounds massive, but in reality:
- It buys Nvidia one or two quarters
- It does not sustain the supercycle
- It does not replace U.S. hyperscaler slowdown
- It does not reverse China’s domestic chip momentum
It’s a temporary sugar high … not a cure.
4. The bubble doesn’t need a crash — it just needs demand to grow slower than expected
Nvidia’s valuation is priced for:
- explosive YOY growth
- endless model scaling
- unlimited GPU appetite
- zero meaningful competition
But here’s the crack forming:
Every major buyer is shifting from “more GPUs” to “more efficiency.”
That alone bends the curve.
If China doesn’t show up with a reindeer-sized credit card,
the curve bends a LOT faster.
5. You can already hear the pressure escaping
If “Santa China” doesn’t come sliding down Nvidia’s chimney with a giant sack of H200 orders, then Jensen better start warming up the panic flute, because that’s the ONE storyline Wall Street is still pretending will magically fix everything.
And the funny part?
There is no Santa.
China’s not coming.
They’re not even warming up the sleigh.
They’re building their own workshop in Shenzhen and humming RISC-V Christmas carols.
That’s where we’re heading.
6. Final thought
If Santa doesn’t show up in China buying tens of billions in GPUs,
the rally’s already out of cookies.
Because that’s EXACTLY how Wall Street is thinking right now:
- China orders = Christmas miracle
- No China orders = Grinch stole the rally
The problem?
China isn’t shopping.
China is manufacturing its own Santa.