Following yesterday’s announcement on Washington’s AI Action Plan, we reached out to Professor PS Lee, Head of Mechanical Engineering, National University of Singapore, on what it means in practice.
- Can you briefly explain what the AI Action Plan is about?
Lee: The White House has torn down the “high fence” that once ring-fenced U.S. AI hardware. In its place is a turbo-charged export regime built around bundled “full-stack” AI packages – complete systems that ship GPUs, reference models, data-pipeline tools, security telemetry, and vendor support under a single licence. The strategy is clear: move chips and software to trusted allies faster than China can innovate at home, even if that means relaxing environmental rules and running larger leakage risks.
- How would this impact chip exports to SEA, in particular to Singapore, Malaysia, and Thailand?
Lee: See table below:
Chip Flows to Southeast Asia
| Country |
Upside
|
Watch-points |
|
Singapore |
First-tier access to full-stack exports; strengthens status as the region’s trusted AI gateway. | Rising scrutiny after isolated GPU-smuggling cases and a spike in “compute tourism” by Chinese firms renting local GPUs. |
|
Malaysia |
Greater inflow of high-end chips for Johor & Penang AI parks. | Could face extra U.S. licence caps and audits if diversion persists; higher compliance costs for integrators.
|
|
Thailand |
New opportunities in assembly, cloud hosting, and regional services. | Vulnerable to spill-over restrictions if Malaysian leakages remain; local power and data centre capacity are still catching up. |
Bottom line: Singapore could win big but must tighten vigilance. Malaysia and Thailand can ride the wave only if they seal re-export loopholes quickly.
- What does “full stack AI export packages” mean – does this preclude exports in individual components like AI chips?
Lee: Full-stack means turnkey AI capability plus baked-in traceability. Component-only shipments (e.g., a batch of standalone H100 GPUs) remain legal but would draw heavier paperwork, location-verification telemetry, and spot audits. The package model keeps U.S. firms indispensable and gives Commerce & State a single dashboard for end-use compliance.
- Will this further accelerate the development of AI worldwide but at the same time polarise the “AI-haves” and “have-nots”?
Lee: Turnkey systems slash the adoption barrier for aligned economies, accelerating AI rollout in finance, health, defence, and logistics. Yet countries outside the U.S. orbit, or those lacking data-centre infrastructure, risk falling even further behind, cementing a two-tier global AI order. The digital divide will become wider.
- Will China be able to keep up or even surpass US in AI?
Lee: Export liberalisation buys Washington time, but China is hardly kneecapped – these tactics slow but not stop China. In data-rich verticals where algorithmic ingenuity trumps raw FLOPS, Beijing could still overtake. It can do the following:
- Software efficiency: squeezing more FLOPS from older
- Grey-market chips & domestic fabs: diversifying
- Offshoring compute: renting GPUs in Singapore, Malaysia, and the Gulf to train models, then repatriating the weights.
- Will the environment suffer now that it’s taken a backseat?
Lee: Fast-track permitting and less stringent National Environmental Policy Act (NEPA)/Clean-Water reviews prioritise speed over sustainability just as AI-data-centre power demand is forecast to triple U.S. loads by 2028. Tropical hubs such as Singapore and Johor will face acute energy and water stress unless liquid cooling, low-carbon PPAs, and strict efficiency codes are locked in early.
7. Do you think transshipment will still occur or even increase with more AI chips and other components being exported widely by the US?
Lee: The following will likely happen:
- Hardware diversion: Servers pre-loaded with restricted GPUs continue to leak eastward; each bust triggers tougher end-use audits.
- Data diversion (“compute tourism”): Chinese developers fly terabytes of training data to SEA clouds, lease S. GPUs, and ship model weights home – no chips cross borders.
Expect mandatory telemetry pings, cloud-customer origin vetting, and deeper U.S.- ally inspection pacts; enforcement will remain a whack-a-mole game.
8. In this new regulatory environment, how do you think the US can ensure AI chips won’t fall into the hands of China?
Lee: The US could do the following:
- Location-verification telemetry embedded in high-end
- Enhanced end-user declarations and penalties for false
- Joint audits with trusted hubs (e.g., Singapore Customs + U.S. Commerce). These measures raise the cost of diversion, but spoofing, hardware swaps, and data-offshoring remain viable hacks—constant intelligence sharing is
9. Do you think the Middle-East will leapfrog many other countries due to its alliance with the US?
Lee: The UAE and Saudi Arabia already hold bulk GPU allocations, cheap solar-plus-gas power, and political alignment with Washington. Their capital-driven build-outs could vault them ahead of many Asian and European peers in AI compute per capita, turning the Gulf into a global AI super-node.
10. What is the outlook following this huge game-changing AI strategy?
Lee: I believe the following will happen:
a. Short-term: Allied AI capacity surges; S. vendors reap record orders; regulators scramble to calibrate telemetry controls.
b. Medium-term: Regions pairing compliance with sustainable infrastructure (Singapore, parts of the Gulf) capture the bulk of lawful AI investment.
c. Long-term: The digital map fractures into tightly governed “trusted zones” and a diverse periphery looking to open-source software and non-U.S. silicon to stay in the
Bottom line: Trump’s export push arms friends quickly, but success depends on airtight end-use enforcement and a credible plan to keep the AI-compute surge from smashing through environmental guardrails. Trusted hubs like Singapore can thrive, if they stay one step ahead of diversion schemes and embed sustainability into every new rack.
