Even for the rollercoaster world of AI, last week was particularly volatile. Investors scrambled to keep up with developments that threatened the dominance of the largest western chipmakers — and in the process, exposed a circular, opaque AI economy that rests heavily on the fate of a single company.
🔍 THE BOTTOM LINE
Nvidia has become what analysts are now calling the “central bank of AI” — funding its own customers, underwriting the data centres that buy its chips, and propping up a market structure that nobody fully understands. When China’s CXMT surged 466% on its Shanghai debut and Beijing claimed a lithography breakthrough, the resulting sell-off revealed how fragile that structure is. The AI economy is not a free market. It is a closed loop, and the loop is tightening.
A Week That Shook the Chip World
The turbulence began on Monday, July 28, when Chinese memory chipmaker CXMT floated on the Shanghai stock market. Its shares soared 466%, reaching a valuation of 3.3 trillion yuan (£365 billion). The same day, reports emerged that China had developed its own deep-ultraviolet lithography tools — the precision lasers essential to the chip supply chain over which the Dutch company ASML had held a monopoly.
The reaction was swift. AI-linked shares dropped globally. South Korea’s KOSPI fell 11.5% on Tuesday and a further 6% on Wednesday, dragged down by semiconductor giants SK Hynix and Samsung Electronics. The US Nasdaq briefly entered correction territory. Nvidia lost more than 5% by Thursday, at which point it had been overtaken by Apple as the world’s largest listed company.
Then came the rebound. Strong financial results from Amazon and Microsoft calmed traders’ nerves. The KOSPI jumped nearly 20% on Friday — though the week’s slump still meant it recorded its worst month since October 2008.
What CXMT Actually Is — and Isn’t
What is CXMT? ChangXin Memory Technologies is China’s largest domestic maker of DRAM memory chips — the components that store data in phones, computers, and AI systems. Its 466% debut was the largest IPO in Chinese mainland history.
Here is the crucial distinction: CXMT makes dynamic random-access memory (DRAM) chips, which store data. It does not make graphics processing units (GPUs) — the “brains” of an AI system that Nvidia dominates. CXMT and Nvidia produce complementary, not competing, products.
That means CXMT is arguably more of a boon to the global AI economy than a threat. There is a global shortage of DRAM chips, which is why phones and computers are expected to get far more expensive. As Forrester analyst Alvin Nguyen put it: “SK Hynix, Micron, others, they can’t produce enough memory chips to begin with … the demand keeps growing even higher.”
The more serious concern is the lithography news. If China can manufacture the machines that etch the thinnest lines in the world into silicon wafers, Beijing could — in theory — produce GPUs that rival Nvidia’s. But as Nguyen cautioned: “Fabs take years to develop.” Chris Beauchamp, chief market analyst at IG, agreed: “These Chinese chip companies appear poised to do to the big chipmakers what they have done to steel, automobiles and a host of other industries, namely undercut them and outcompete them on price.”
The Nvidia Problem: Circular and Opaque
The sell-off was an overreaction to the China news. But it was also a reasonable response to an AI economy that is increasingly difficult to understand — and increasingly dependent on one company.
The most revealing detail of the week was the Wall Street Journal report that Nvidia is considering a $250 billion backstop for OpenAI’s massive data centre project. This comes roughly half a year after a $100 billion deal between the two companies fell apart.
The anxiety is that Nvidia has become the “central bank of AI” — holding up vast parts of the economy and the global stock market in ways that most people, investors included, do not understand. The chipmaker is simultaneously the supplier, the customer (through its own AI products), and now the lender. Morningstar noted that the OpenAI backstop was a significant factor in Nvidia’s decline over the week.
This is not how markets are supposed to work. When the supplier of a critical input is also the financier of the buyer, the price signal that usually governs supply and demand is replaced by a relationship. The market stops being a market and becomes a vertically integrated closed loop — one where the only company making a profit on AI is the one at the centre.
Why This Matters Beyond Silicon Valley
The implications extend well beyond chip stocks. Nvidia is the only company making a profit on AI. The entire AI buildout — the data centres, the energy contracts, the model development, the enterprise software — depends on Nvidia chips. When Nvidia’s shares wobble, the knock-on effects touch KiwiSaver funds, pension systems, and retail investors worldwide.
The comparison to 2008 is imperfect but instructive. The global financial crisis exposed how much of the economy depended on instruments and relationships that regulators and investors did not fully understand. The AI economy in 2026 has a similar opacity problem. Nvidia’s $250 billion guarantee for OpenAI is not a regulated bank loan. It is a private commercial arrangement between two private companies, one of which is about to go public at a valuation exceeding $1 trillion.
Nguyen’s assessment is blunt: “Nvidia knows the gravy train’s going to run out. Everybody’s waiting for them to fall apart. I don’t know that they will because what they do still has value … at some point in the future, they’ll no longer be one of the most valuable companies in the world. Maybe they’ll be worth only $2tn. It’s still pretty good.”
NZ Angle
New Zealand investors are exposed to this opacity through KiwiSaver default funds and global index trackers, which hold Nvidia, Apple, Microsoft, and Amazon as their largest positions. The volatility of the past week — the KOSPI’s 18% swing, the Nasdaq’s flirt with correction territory — flowed through to NZX-listed passive funds. The deeper question for NZ is whether the AI buildout that Nvidia underwrites will eventually deliver the productivity gains that justify the spending, or whether the closed-loop structure means the returns stay inside the loop.
❓ FAQ
Is CXMT a threat to Nvidia? Not directly. CXMT makes DRAM memory chips, not GPUs. They are complementary products. The real threat to Nvidia would be China developing its own GPU manufacturing capability — which the lithography breakthrough makes possible, but is still years away.
Why is Nvidia lending money to OpenAI? Nvidia is reportedly considering a $250 billion backstop for OpenAI’s data centre project. The motivation is likely strategic: keeping OpenAI as a customer for Nvidia’s chips ensures continued demand. But it creates a circular relationship where the supplier is also the financier.
Should investors be worried about the AI economy’s structure? The concern is not that AI is overhyped — Amazon and Microsoft’s earnings suggest the demand is real. The concern is that the financial structure around AI is opaque, circular, and concentrated in one company. That concentration creates systemic risk, similar to the role AIG played in 2008.
What does the China lithography breakthrough mean? If confirmed, it means China can manufacture the precision tools needed to make advanced chips, breaking ASML’s monopoly. This is a symbolic victory for now — matching western reliability will take years. But it signals that US export controls have accelerated, not slowed, China’s push for self-sufficiency.
🔍 THE BOTTOM LINE
The AI economy is not a market. It is a relationship — one where Nvidia sells the chips, funds the buyers, and is the only company making a profit. Last week’s turmoil did not break that structure. It revealed it. The question for investors, regulators, and anyone whose pension is tied to tech stocks is whether the structure can hold, or whether the “central bank of AI” is about to discover what central banks discover when the loans go bad.
