Asian Players Buying Chips and AI: Who Gets Past the Bouncers?

Buying advanced AI chips is no longer simply a contest of price, technical need and delivery schedules. Suppliers increasingly want to know who owns the buyer, who will actually use the equipment, where the computing power will be deployed and whether a Chinese-controlled entity sits somewhere behind the transaction.

Recent US guidance has clarified that restrictions covering sensitive American technology may apply not only to named Chinese companies, but also to entities they ultimately own. Nvidia has reportedly responded by sharply reducing the number of approved Asian customers across Singapore, Malaysia and Japan after intensive checks on ownership and end users. Access to its most sought-after processors increasingly depends on making it onto a heavily vetted “white list.”

In this case, neo-cloud operators were the most affected. But these matter equally to Asian data-centre operators, semiconductor manufacturers, telecom groups building AI infrastructure, industrial companies developing proprietary AI, sovereign-backed technology platforms and regional businesses whose investors, customers or affiliates span China and the West.

In light of enhanced checks by suppliers, customers should expect deeper diligence and more intrusive questioning: reviews of shipping manifests, beneficial ownership, holding-company chains, joint ventures, nominee structures, financing and the identities of ultimate users.

Many companies are not prepared for that scrutiny. As was the case when anti-money laundering rules first rolled out. Building the necessary systems may take months. Building an organizational culture capable of adapting as the rules change usually takes much longer.

A New Ring of Gatekeepers

Three proposed US measures targeting semiconductors and AI, if enacted, would bring more actors into decisions once handled largely through supplier diligence and Commerce Department licensing.

The AI OVERWATCH Act would reclassify certain exports of advanced AI chips capable of training frontier models as Foreign Military Sales. That would give Congress direct oversight of major semiconductor deals and reduce the degree of discretion concentrated in the Department of Commerce.

The Chip Security Act would require location-verification to be embedded in advanced hardware. Its purpose is to help detect and prevent chips from being diverted through third countries to specific destinations.

The MATCH Act would restrict access to specialised deep-ultraviolet immersion lithography tools and other semiconductor-manufacturing equipment. It would also give the Netherlands, Japan and other equipment-producing allies a deadline to bring their controls into line with Washington’s. If they do not, the US could extend its own restrictions to foreign-made tools that rely on American technology, components, software or servicing. The choice is therefore not simply whether allies wish to coordinate: their companies may become subject to US controls even if their governments do not follow.

That matters most immediately to the Netherlands and Japan, whose companies supply some of the most advanced lithography and semiconductor-manufacturing equipment in the world. That equipment does and may still depend on US-origin technology, components or software—and therefore remain exposed to Washington’s export-control reach.

Beijing Raises Gates of Its Own

China is hardly new to export controls. But since 2023, it has been tightening the screws—first around critical minerals and materials, then through a more Western-style machinery of licences, end-user restrictions and targeted entity controls.

Now Beijing is reportedly considering a further turn of the dial: tighter controls on the export of advanced AI models, chips and other frontier technologies.

Chinese regulators have reportedly consulted companies including Alibaba, ByteDance and Z.ai about possible restrictions on the overseas transfer of model weights, important training data and Chinese-developed semiconductor technologies. The ideas remain under consideration.

Why Sellers May Walk away, Leaving You Without a Plan B

The concentric rings of gatekeepers growing each day create immediate and long-term impacts on Asian companies, who rely on supplies from one or the other to drive their own technologies. While some are straddling both systems, hoping to keep options open, they find themselves wedged between doors that can close in both directions.

While one can expect companies to screen customers beyond the formal export-control lists (eg the buyer, the end user, the proposed use and signs that an item might be redirected), the intensity of today’s political exposure, shifting rules, and above all – scarcity – mean that the runway of options can come to an abrupt end.

Nvidia will not be alone in having to be extremely selective about who they do business with. Nvidia, TSMC, ASML, Applied Materials, Lam Research and KLA sit at different points in the technology stack, but they share one unusual characteristic: customers cannot easily replace them without changing the underlying technical plan.

Their products and services are difficult—and in some cases practically impossible—to substitute. When one of these firms declines to sell, the customer is not merely facing a higher price from the next vendor. It may face a multi-year redesign, a capability gap with no near-term substitute or an outright dead end.

The suppliers, meanwhile, are not short of buyers.

What causes a supplier hesitation usually begins with a story that is difficult to verify quickly. In the face of tight due diligence actions, buyers may struggle with a clear and evidenced narrative: for instance, ownership takes several diagrams to trace back to a real owner; user lists that do not seem fully consistent. Then there could be financing that arrives from sources with no obvious connection to the buyer. Or, a cloud arrangement in which no one can show with confidence who will actually rent the computing power once it’s live.

Indeed, each of these can be investigated, negotiated around, and eventually explained to the supplier’s satisfaction — but that takes weeks, sometimes months, against a customer next in line whose story is legible in five minutes. Even a transaction that clears every present test can still go wrong later: a parent company might be added to a list, remote access provisions may be spotlighted, tracking obligations might require the manufacturer to account for a processor’s whereabouts long after delivery.

The commercial calculation may become as simple as: Why carry one customer into the club with a collection of issues you must resolve yourself when ten less complicated people are waiting at the door?

Two Regional Worries

1. Grey market — where is the hardware ending up?

Sophisticated regional trading networks are creating logistical challenges for tracking advanced technology hardware, as highlighted by a case involving U.S. server suppliers and shipments destined for Malaysia via Singapore. While initial reports speculated on the final destination of powerful, potentially NVIDIA-equipped servers, the case focuses on alleged misrepresentations rather than proven illicit diversion to specific, unconfirmed end-users. This episode underscores the complexities of managing tech supply chains and compliance in Southeast Asia, where invoice addresses may not reflect the final location of sensitive equipment.

2. Cloud and data hubs — who is using the computing power?

Hardware is only part of the puzzle. Advanced computing capability can cross borders even when the chip does not. A processor may remain installed in a Malaysian data centre while a foreign customer uploads data, trains a model and downloads the results. From one perspective, that can look like an ordinary export of cloud-computing services. From Washington’s perspective, it may allow a Chinese company to obtain the benefit of advanced US technology that it could not purchase directly for use at home.

That ambiguity has placed Malaysia under pressure to show that its rapidly growing data-centre industry will not become a back door into restricted computing capacity. In July 2025, Malaysia required permits for the export, transit and transshipment of high-performance US-origin AI chips. The measure primarily governs the movement of hardware; but the trickier question of how operators identify the true customer when computing capacity is rented remotely or through intermediaries is an ongoing issue.

Singapore presents a different problem: the purchaser named in the transaction may not be the ultimate user. In a case launched in 2025, prosecutors alleged that individuals falsely told Dell and Super Micro that Singapore-based companies would be the end users of servers that were subsequently shipped to Malaysia. The final destination was not publicly established. The case shows why a supplier would be cautious about relying on the company named on the purchase order; it may also need to understand the intermediaries, onward shipment arrangements and real beneficiary behind the transaction.

The Standing Obligation

When the first cycle is satisfactorily complete, the temptation is to treat gatekeeping as a hurdle cleared once: pass the check, get the stamp, move on to the actual business of building. But expectations of the world of sensitive technology is shaping out to be just the opposite. Every licence or “trusted” designation, is turning into a standing obligation that must be underwritten continuously.

The same instinct is stirring on the other side of the rivalry. Beijing’s own gatekeeping has so far centred on what leaves China physically like critical minerals. But the reported consultations with Alibaba, ByteDance and Z.ai point toward the next front: gating what leaves digitally — model weights, training data, the semiconductor know-how China has developed on its own terms.

If that hardens into policy or formal rules, companies hoping to draw on cheaper Chinese models or mature-node manufacturing capacity as a hedge against Western tightening may find that door subject to its own vetting, its own list, its own annual reapproval.

The bouncers are multiplying on both sides of the corridor. The companies most likely to get through will not necessarily be the richest or technically strongest. They will be the ones whose ownership, users and intentions can be understood quickly—and trusted continuously.