Billions in GPUs Flood Asia Raising Questions About Their Destination

Across Asia, countries are racing to bolster their technological infrastructure, driven by the promise of artificial intelligence and cloud computing. The region has become a hotspot for data center construction, with nations like China, Singapore, and Malaysia investing billions to meet soaring demand. Yet, a less visible trend is raising eyebrows: the massive procurement of graphics processing units (GPUs), the powerful chips that fuel AI and high-performance computing. Some observers question whether the volume of GPUs being purchased aligns with the region's available power capacity.

The Asia-Pacific region is projected to be the fastest-growing market for GPUs, with Japan and South Korea anticipating over 32% annual growth in demand through 2035. China alone spent over $16 billion on Nvidia GPUs - H20 data center GPUs in early 2025, a sharp rise from the previous year. This surge, led by tech giants like Alibaba and Tencent, reflects both the region's AI ambitions and fears of impending U.S. export restrictions. But the sheer scale of these purchases has sparked speculation about their ultimate destination and use.

Adding to the complexity, countries like Singapore have seen GPU sales skyrocket, with Nvidia reporting a tenfold increase in revenue from Singapore entities, from $2.29 billion in 2023 to $23.68 billion in 2025. This has led some to suspect that GPUs sold to intermediaries in Southeast Asia may be rerouted to restricted markets, raising questions about the effectiveness of global trade controls. The mismatch between GPU purchases and available power infrastructure only deepens the mystery.

The rapid expansion of data centers and GPU use is straining Asia's energy infrastructure. Data centers are projected to consume 536 terawatt-hours globally in 2025, with AI-specific facilities expected to account for 90 terawatt-hours by 2026. In Singapore, data centers already consume 7% of the nation's electricity, a figure set to climb to 12% by 2030. Modern AI GPUs, which can draw up to 1,200 watts per chip, are pushing server rack power densities to new heights, from 36 kW in 2023 to a projected 50 kW by 2027.

This energy challenge is particularly acute in Southeast Asia, where land and power constraints limit growth. Singapore, despite hosting 60% of the region's data center capacity, has faced occupancy rates above 85%, prompting a temporary halt on new builds until 2022. Emerging hubs like Malaysia and Indonesia offer cheaper energy and land, but their grids struggle to keep pace with demand. Efforts to adopt renewable energy and advanced cooling systems are underway, yet the gap between power supply and GPU-driven demand remains a critical bottleneck.

The energy question ties directly to concerns about GPU procurement. If countries are buying more GPUs than their grids can support, it suggests either overstocking in anticipation of future capacity or diversion to other markets. This discrepancy has fueled speculation that some purchases may be funneled through intermediaries to bypass export controls, particularly to China, where domestic chip development lags behind global leaders.

The GPU trade has become a flashpoint in the broader U.S.-China technological rivalry. U.S. export controls, tightened in April 2025, now require licenses for advanced AI chips like Nvidia's H20 to over 40 countries, aiming to curb China's AI capabilities. These measures, coupled with tariffs as high as 145% on Chinese tech goods, have disrupted global supply chains and spurred China to retaliate with 125% tariffs on U.S. imports. The result is a fragmented market, with both nations racing toward self-sufficiency in chip production.

Intermediaries, including shell companies in places like Hong Kong and Singapore, have emerged as key players in navigating these restrictions. By obscuring end-users, these entities enable the rerouting of GPUs to restricted markets, a practice that U.S. and European regulators are struggling to curb. Recent U.S. legislation, effective January 2025, mandates stricter ownership reporting to combat such tactics, but enforcement across borders remains patchy. Meanwhile, countries like South Korea, exempt from some U.S. restrictions, are scaling up their own GPU acquisitions, with plans to deploy 10,000 units by 2025.

The stakes extend beyond economics. GPUs are critical to AI applications in defense, healthcare, and industry, making access to these chips a matter of national security. As tensions rise, Southeast Asia's role as a trade hub complicates efforts to enforce controls, while China's push for domestic chip production signals a long-term shift in the global tech landscape.

The GPU trade underscores the intricate interplay of technology, energy, and geopolitics shaping Asia's digital future. As countries pour billions into data centers and AI infrastructure, the gap between ambition and capacity - both in power and regulation - will test their ability to sustain growth. For now, the region's role as a tech powerhouse is undeniable, but questions about where GPUs end up and how they're used linger unresolved.

Balancing innovation with accountability remains the challenge. Policymakers, tech firms, and energy providers must navigate a landscape where every chip purchase carries strategic weight. As Asia's tech boom accelerates, the world watches closely, aware that the outcomes will ripple far beyond the region's borders.