
17 November 2026
Over the past day or two, developments surrounding NVIDIA have multiplied, and it has become increasingly difficult to consider them separately. Since Jensen Huang left the CEO position, some Silicon Valley technology leaders, NVIDIA employees and investors have called for his return. Trump has also expressed support, or at least indicated that he would not oppose it. At the same time, the United States has tightened its oversight of advanced AI chips and overseas computing services, transactions involving Southeast Asian partners have come under investigation, and NVIDIA and other AI-related shares have faced pressure. Consider these developments alongside the enormous investment in data centres and the debate prompted by recent major IT outages, and the controversy now extends across corporate governance, national security, technical reliability and capital markets.
The most pressing need is to resist rushing towards a single conclusion about the whole affair. These issues have converged in time, but that does not mean they share a common cause, nor does it establish that NVIDIA’s senior executives orchestrated an unlawful scheme. If risks in different areas begin to reinforce one another, however, their combined impact could be far greater than that of any individual event. That is precisely why the recent developments deserve continued attention.
Huang’s importance to NVIDIA extends beyond his long tenure as CEO. He has become almost synonymous with the company’s technological direction, commercial capabilities and public identity. Its rapid growth has been closely connected with his judgement, organisational approach and customer relationships. Many people want him back because they trust him personally, but practical interests are also involved. Employees are concerned about the company’s direction; customers want certainty about product supply and construction plans; investors fear that a leadership change could undermine expectations; and other technology companies do not want prolonged instability at one of the AI industry’s most important suppliers.
The board faces a different set of considerations. If investigations relating to national security and export controls remain unresolved, immediately restoring Huang’s previous authority could make it difficult to explain to shareholders and regulators how an independent investigation would remain free from interference. Political support can influence public opinion, and petitions from industry leaders may affect market sentiment, but neither can substitute for legal and compliance procedures. Even if Huang is ultimately found to have broken no law, the board would still have governance grounds for maintaining some distance at this stage.
The more serious question concerns the extent to which the chip supply and computing services network NVIDIA has developed through Southeast Asia has approached or crossed the boundaries of US export controls. Discussions of chip exports once tended to focus on a shipment of GPUs moving from one country to another. The situation is now much more complicated. Chinese customers may rent computing capacity from data centres in places such as Singapore and Malaysia without bringing restricted chips into China. From the customer’s perspective, they still obtain advanced AI computing power. From a regulatory perspective, however, exports of physical equipment, effective control of a data centre and the remote provision of computing services may fall under different legal rules.
Two situations need to be distinguished. If a company uses false end-user information to divert controlled chips through a third country to an unauthorised destination, that may constitute a clear export-control violation. A different situation arises when equipment remains in an approved overseas data centre and an overseas operator supplies services to customers. Whether that arrangement is unlawful depends on the rules in force at the time, the licence conditions, the customer’s identity and the services actually provided. An arrangement may raise policy concerns, but stricter rules introduced later do not establish that every earlier transaction was illegal.
The recent tightening of US oversight of transactions involving Singapore, Malaysia, other Southeast Asian countries and Brazil deserves particular attention. It points towards a policy that increasingly considers who ultimately uses the chips and whose activities their computing power serves, as well as where the chips are sold. Tougher licensing requirements differ substantially from an outright export ban, but even a higher licensing threshold can affect data centre construction, financing arrangements and customer expectations. For projects that have already committed substantial funds, the ability to obtain licences in future can matter more than whether equipment can be delivered today.
How far the US investigations proceed will depend on the evidence. If particular distributors or operators concealed where equipment was going, the main issues would concern those transactions and the people involved. If someone inside NVIDIA knew in advance about unlawful diversion and nevertheless assisted it, the company’s compliance system and the legal responsibilities of those involved would need reassessment. Evidence that senior executives deliberately organised, tolerated or concealed such conduct would make the matter considerably more serious and could even raise questions of criminal liability. These are hypotheses of different severity. The most serious possibility must not be treated as an established fact.
For the same reason, whether Huang can return does not depend simply on how many people support him. If the investigations find no personal wrongdoing, calls for his return may become stronger. If they reveal problems involving the company’s longstanding internal decisions, the board would find it difficult simply to restore the previous management arrangements, however strong his commercial reputation. A deeper question is whether NVIDIA has concentrated too much of its strategy, customer relationships and critical judgement in one individual, to the point that his departure itself becomes a risk for the entire industry.
The stock market has begun responding to these uncertainties. NVIDIA’s approaching earnings announcement is particularly relevant. Investors previously focused mainly on chip sales, revenue growth and gross margins. Those measures still matter, but they may no longer be sufficient to steady market confidence. Earnings describe the quarter just completed, while investors are worried about export licences, orders, financing and governance over the quarters ahead. Even strong profits may leave the market believing that future growth carries much greater risk. Equally, if both results and guidance substantially exceed expectations, the share price could rebound. It would be wrong to assume that it must fall regardless of the results.
What concerns me more is whether doubts about NVIDIA will gradually spread to the AI industry’s current investment model. In recent years, enormous sums have been committed around the world to building data centres, buying high-end chips, and expanding electricity and network infrastructure. Every part of this system is preparing for the next stage of AI demand. Yet who will ultimately pay for that investment, and how long it will take to recover the cost, remain insufficiently tested.
A capital cycle within this process warrants caution. Chip companies sell GPUs; cloud providers and data centres buy them; financial institutions provide financing; and some upstream companies may also invest in their downstream customers. Those customers then use the funds to expand their purchases of computing capacity. The resulting revenue supports growth expectations and valuations for the suppliers. Higher valuations encourage further investment, giving the whole chain renewed momentum to expand.
There is nothing necessarily wrong with such a cycle. Many new industries need capital investment before end-user demand and profitability develop. Investment and procurement relationships between companies do not, by themselves, establish that demand is artificial. But if downstream businesses remain unable to generate enough revenue from the AI services they actually provide, and must continually raise new funds to buy more equipment, while upstream growth becomes increasingly dependent on those purchases, the cycle may begin to resemble a game of passing the parcel. The crucial question is whether enough real customers will ultimately pay for computing power and AI applications on a continuing basis.
Assessing whether that risk is emerging requires more than looking at how far share prices have fallen or how much revenue NVIDIA earns. It requires examining actual data centre utilisation, computing rental prices, AI application revenue, customers’ operating cash flow, and the costs of electricity, maintenance, depreciation and financing. In particular, revenue generated by an established market of end users needs to be distinguished from revenue that depends heavily on continued investment within the industry itself. Both create orders in the current period, but they may not be equally sustainable.
Tighter export controls could provide an external shock that tests this cycle. If some overseas data centres expected to serve Chinese customers over the long term, a policy change could directly alter their revenue models. Lower revenue expectations could lead lenders to reassess risk, slow project construction and reduce orders for equipment further up the chain. The change might initially affect only a subset of projects, but capital markets often extrapolate localised risks across an entire industry. There is still insufficient evidence to determine whether this transmission would be enough to interrupt the growth of AI infrastructure investment.
Another factor, at the technical level, is easily overlooked. Recent major IT outages have prompted questions about whether the rapid adoption of AI-generated code, automated development and automated operations has increased risks to software quality and system maintenance. Those concerns deserve investigation, but an incident cannot be attributed to AI simply because it occurred during a period of rapid AI adoption. Large systems have always faced software defects, configuration errors, capacity constraints and complex dependencies. The cause of a particular failure must be established through an incident investigation. The relevant question is whether testing, review, maintenance and accountability have kept pace with the faster development and deployment that AI enables.
This brings the discussion back to economics. If AI can substantially improve productivity, it can certainly create real commercial value. But the additional reliability, security and governance costs it brings must also be counted. A technology that generates code or trains models faster must ultimately demonstrate that it can operate reliably in real conditions and produce returns sufficient to cover all its costs. National security, technical and investment risks begin to intersect here, although the specific relationships between them still require factual examination.
Over the coming weeks and months, the developments worth following therefore extend beyond whether Huang returns or how the share price moves after a particular earnings release. Can the US investigations establish that the companies concerned knew about unlawful transactions? How will the new export licensing requirements operate in practice? How will NVIDIA’s board change its governance arrangements? Will overseas data centres postpone investment? Can AI service providers’ actual revenue support their growing spending on computing capacity? Taken together, these developments will help clarify the direction in which events are moving.
I continue to regard artificial intelligence as an important area of development, and a fall in share prices does not, in my view, invalidate its technological prospects. But promising technology and an economically validated scale, pace and return model for current capital investment are separate matters. The pressure NVIDIA is experiencing may amount to a corporate governance and regulatory crisis, or it may prompt the market to examine the whole AI investment cycle more seriously. It is too early to tell which. More useful than rushing to declare whether a bubble has burst is to ask whether the industry can gradually establish a self-sustaining balance of revenue and costs without depending on ever-expanding financing.
Discover more from Geoffrey Chen
Subscribe to get the latest posts sent to your email.