A huge bubble in China is about to burst

China Unfolded
14 followers
May 21, 2025
A huge bubble in China is about to burst
In mid-May, during the 90-day tariff suspension window,
a silent war for AI computing dominance suddenly exploded.
Server prices went wild, swinging violently, rising 15% to 20% in a short span.
With the suspension of tariffs, we planned to resume sales at the original price,
a chip supplier in southern China revealed to Huxu.
Meanwhile, new variables hit the supply side.
Huxu exclusively learned that NVIDIA's Hooper series of high-end chips
and the even more advanced Blackwell series have quietly entered the domestic market.
The Hooper chips started surfacing around September 2024, and Blackwell only recently.
A relevant executive from Huarui Zhisuan remarked,
Different suppliers have different channels for obtaining supply.
But the shadowy web of supply chains behind this remains untouchable.
Since October 17, 2023, Washington has tightened its grip,
phasing out NVIDIA chips like A100, A800, H800, H100, and H200 from being sold to China.
Most recently, the last H20 chip of the Hooper line was added to the restricted export list.
The Hooper series flagship, the H200, is a souped-up version of the H100,
30% more efficient and costing just over 200,000 yuan more.
But the Blackwell series, especially the B200, takes it to a whole new level,
priced over 3 million yuan, extremely restricted and traded through backchannels.
Both series are critical for training large AI models,
with the B200 becoming nearly impossible to get, a ghost card in the market.
Rewind to April 2024, when a photo of NVIDIA's Jensen Huang with OpenAI's
Sam Altman and Greg Brockman went viral on Twitter.
It marked the first H200 delivery, hand-delivered by Huang himself.
OpenAI, I was the very first customer.
Five months later, H200s are showing up in China.
Domestic suppliers now claim they can deliver up to 100 servers per week,
with the H100 discontinued.
Demand is surging toward the H200.
According to one supplier, fewer than 10 companies have access to these chips,
and the supply-demand gap is widening fast.
The H200 is the most scarce item in the market now,
a seasoned computing power veteran with 18 years of experience told Huxu.
A cloud factory has been desperately hunting them down recently.
His company has long provided services to tech giants like Baidu,
Alibaba, Tencent, and ByteDance.
In this high-stakes arms race, deals happen in the dark.
A top domestic supplier told Huxu that contracts no longer mention specific hardware,
only abstract computing units called P.
Tiger Sniff Note, P is the calculation unit for computing power.
Server transactions are converted into computing power agreements,
and the cards themselves are kept out of sight.
Deeper down the supply chain, the underground network comes into view.
Some media revealed that Chinese dealers use special procurement,
layered reselling, and packaging tricks to smuggle servers in via the backdoor.
Huxu further learned that others go a step further,
embedding modules into products via third-party companies
to acquire chips through alternative loopholes.
Beneath these turbulent currents, China's computing power industry is morphing.
Where did this AI arms race really begin?
At the end of 2023, the U.S. ban on NVIDIA chips hit like a meteor.
It shattered the calm, and ignited a hidden scramble for core computing firepower.
The first few months were chaotic.
The market became a gold rush.
Driven by massive profit margins, savvy risk-takers dove in.
Back then, everyone was a supplier.
Overseas students, solo traders, anyone with inside info,
said one anonymous industry insider.
Their deals were crude and basic,
still underground but nowhere near the layered subcontracting chain we see now.
These early pioneers exploited information asymmetry,
and a patchwork of grey channels to flood the market with NVIDIA's high-end graphics cards.
Prices skyrocketed overnight.
Some media revealed that individual resellers brazenly priced NVIDIA A100 graphics cards
at 128,000 yuan, far above the official MSRP of around $10,000.
In an even bolder move, someone flaunted an H100 chip on social media,
boasting that a single unit sold for a staggering 250,000 yuan.
These acts weren't just transactions, they were a public spectacle of power and access.
Amid this underground surge, large computing power providers
began establishing their own shadowy supply chains.
This fueled a frenzy across the nation.
From 2022 to 2024, a gold rush of intelligent computing centers sprang up.
In 2024 alone, over 458 projects were launched, reflecting the fever pitch of the boom.
But the bubble expanded too fast.
By late 2024, the tide began to turn.
With domestic AI models like DeepSeek proving their worth through cost efficiency,
many players who had merely hoarded GPUs or lacked deep technical foundations
found their stories collapsing.
The intelligent computing dream started to fray at the edges.
Data from Q1 2025 shows the aftermath.
165 intelligent computing centers were in progress across China,
but only 16 were actually operating or in trial use, less than 10%.
The rest remained stuck in planning or early construction, hinting at a looming overbuild.
And the problem isn't just domestic.
Over the past six months, global tech titans like Meta and Microsoft
have hit the brakes on several major data center projects.
Behind the AI hype lies an unsettling truth.
Idle machines and wasted investment.
The current activation rate of intelligent computing centers
is under 50%, an insider told Hooksyou.
Domestic chips can't handle pre-training, and some centers are running on outdated servers.
This cards-in-hand-but-nowhere-to-play crisis stems from what experts call a structural mismatch,
not a glut of computing power, but a lack of usable high-end capacity.
Much of the infrastructure remains locked, unusable due to tech shortfalls,
underdeveloped ecosystems, or weak operational capabilities.
Still, amid the wreckage, the tech giants aren't retreating.
They're doubling down.
Reports suggest ByteDance is gearing up to pour over 12.3 billion
into AI infrastructure in 2025, with 40 billion yuan earmarked for domestic AI chips
and 50 billion for NVIDIA's.
ByteDance later told Hooksyou the report was inaccurate,
but the scale of ambition speaks for itself.
Alibaba is going all-in on AI.
On February 24, CEO Wu Yongming made waves by announcing that Alibaba would pour
into building AI infrastructure over the next three years,
a jaw-dropping number that surpasses its total investment in the past decade.
But as capital floods in, the supply side is showing clear signs of strain.
The market can't keep up with demand from the tech giants.
Even after contracts are signed, many companies still fail to deliver,
a salesperson from an intelligent computing supplier told Hooksyou.
Ironically, this massive new wave of AI investment clashes sharply
with the deflating intelligent computing bubble.
On one side, A-share-led suppliers are freezing or scaling back major computing projects.
On the other, tech giants are doubling down, racing to build the next-generation AI backbone.
The reason behind this divergence is crystal clear.
The intelligent computing cooldown coincided with DeepSeek's breakout moment.
Since then, no one dares talk about a
100-model war anymore.
DeepSeek exposed the illusion of endless training demand.
Now, only the real players—big firms and serious AI model startups—remain at the table.
Feng Bo, managing partner at Changlai Capital, echoed this view to Hooksyou.
When training demand dries up, only those with true capability and scale,
like Alibaba and ByteDance, are still buying cards.
The rest, lacking core competencies, vanish from the scene,
leaving behind piles of unused computing power. That's the real bubble.
To avoid tariffs imposed by President Trump, Apple has been pushing iPhone production to India.
This strategic move helps reduce the impact of the ongoing US-China trade war.
At the same time, other tech giants like Amazon, Foxconn,
are also rethinking their reliance on China.
Rising geopolitical tensions have made China a risky place for global supply chains.
These companies are now actively seeking alternative manufacturing hubs.
India and Vietnam are quickly becoming their new production strongholds.
Apple has already started assembling 20% of its iPhones in India,
worth $22 billion, a massive leap from just 1% two decades ago.
This shift reflects Apple's growing determination to reduce dependence on Chinese factories.
Amazon, too, has been diversifying its suppliers across Southeast Asia.
They no longer see China as the irreplaceable
world's factory. The search for stable and cost-effective alternatives is now urgent
and irreversible. High labor costs in China are one major reason for this exodus.
Average factory wages have jumped from $150 per month in 2005 to over $1,200 in 2025.
In comparison, labor costs in India and Vietnam remain far lower. Apple's accountants see better
margins outside of China. The dream of cheap, large-scale manufacturing in China is fading.
Apple and Amazon are making tough decisions to protect their bottom lines.
But labor costs are not the only issue pushing companies out.
Younger Chinese generations are turning away from factory work, preferring jobs in tech and services.
This cultural shift has resulted in a worsening labor shortage, especially in manufacturing hubs.
As factories struggle to fill positions, wages climb even higher.
For Apple and Amazon, the long-term risks in China are growing unbearable.
Exit becomes not just a choice, but a necessity.
The US-China trade war has further accelerated the departure. With tariffs as high as 245%
on Chinese goods, companies are seeing profits evaporate. In retaliation,
China imposed its own restrictions, tightening the noose.
Amazon and Apple are particularly exposed, given the scale of their Chinese operations.
Global supply chains have come under pressure, and moving production is the fastest way to adapt.
China's manufacturing dominance is clearly under threat.
COVID-19 was the tipping point. When China locked down entire provinces,
companies like Apple and Amazon were caught in crisis.
Their dependence on a single country backfired, delayed shipments,
broken supply chains, and inventory shortages became the norm.
This shock forced companies to rethink their global strategies.
Now, resilience and diversification are key, not just cost efficiency.
Apple and Amazon's retreat marks the end of an era.
China's once unshakable role as a manufacturing giant is collapsing.
Southeast Asia and India are rising as the new production centers.
The ripple effects will reshape the global economy.
Other companies are already following their lead.
The great supply chain exodus from China is just beginning, and it's picking up speed.
A factory can ignite the prosperity of a city, or it can silence it overnight.
Foxconn's parent company, Hanhai, recently quietly launched a major move,
planning to set up its first wholly-owned factory in northern India in Uttar Pradesh,
UP. It is rumored that the scale may even exceed the existing Bangalore facility,
which is known as the second-largest factory in the world.
At the same time, Foxconn factories in China are laying off employees, relocating, and going silent.
Netizens shouted,
Don't let Foxconn run away!
But the truth is, it has left a long time ago, and we're only seeing the results now.
Foxconn, one of the largest electronic component manufacturers in the world,
is shifting a significant portion of its production lines out of China
to mitigate risks from escalating trade tensions and political issues.
The company has been ramping up investments in countries like India and Vietnam,
where labor costs are lower,
and the manufacturing environment is less impacted by the stringent policies in China.
This shift is causing Foxconn's factories in China to experience a devastating situation,
with many production lines being scaled down or shut down, leaving thousands of workers unemployed.
In China, Foxconn's factories, once central to the production of major brands like Apple and Sony,
are now struggling with a severe lack of orders and a drastic decline in production.
Some factories have been forced to lay off workers or reduce operations,
leading to a dire economic situation in these industrial zones.
This not only affects Foxconn, but also impacts its subcontractors,
who are unable to maintain contracts and production.
If Foxconn's move is just a business choice,
then what's happening behind it signals the structural rupture of the entire made-in-China.
Could this be the beginning of a long-term recession in the Chinese economy?
Meanwhile, in countries like India and Vietnam,
Foxconn is making substantial investments to restructure its supply chain.
New factories producing iPhones in Devanahalli, India, and other facilities in Vietnam
are being built with local government support.
This clearly signals Foxconn's exit from China, where geopolitical risks
and an increasingly difficult manufacturing environment are making it harder to compete.
The factories in China appear to no longer be competitive in the face of the growing
global trend of production relocation.
The Chinese government has tried a wide range of tactics to prevent Foxconn
from relocating its production lines to countries like Vietnam and India.
Initially, they attempted to entice the company with financial incentives,
including tax breaks and land subsidies, to keep its factories within China.
They also offered Foxconn favorable terms for land use and expanded factory space,
hoping to maintain their stronghold as the primary manufacturing hub for global tech companies.
However, these efforts were ultimately unsuccessful,
as Foxconn recognized the long-term risks of remaining in an increasingly unstable environment.
As Foxconn continued to press forward with its plans to diversify its production base,
the Chinese authorities resorted to more aggressive measures.
They implemented bureaucratic obstacles, delayed permits,
and even launched heavy inspections at Foxconn's existing facilities,
creating an environment of constant uncertainty and tension.
Despite these efforts to tighten control, Foxconn's leadership was undeterred.
The company made it clear that the pressures of operating in China,
coupled with rising labor costs and political volatility, had forced their hand.
In the end, Foxconn's decision to move its operations to other countries
was not swayed by the Chinese government's tactics,
signaling a clear shift in the global manufacturing landscape.
