MediaTek Opens Taiwan Data Center With NVIDIA DGX B200

MediaTek has officially opened the first phase of its new AI research data center in Taiwan's Miaoli Tongluo Science Park, marking the chip designer's biggest infrastructure investment in high-performance computing. The facility is the first in Taiwan to deploy NVIDIA's DGX B200 platform within a full DGX SuperPOD configuration, signaling MediaTek's aggressive push into AI-driven chip development and research.

The data center, which broke ground in 2023, also stands as Taiwan's first large-scale deployment of advanced immersion cooling technology, achieving a remarkably low Power Usage Effectiveness (PUE) of 1.1 — placing it among the most energy-efficient AI compute facilities in Asia.

Key Facts at a Glance

• First in Taiwan to deploy NVIDIA DGX B200-powered DGX SuperPOD systems
• PUE of 1.1, enabled by immersion cooling — far below the industry average of 1.5-1.6
• 235 kW solar panel array installed on the facility's roof, capable of powering 67 households
• Recycled water used as the primary cooling medium for HVAC and partial cooling systems
• 'Fab-grade' power supply infrastructure ensures uninterrupted compute operations
• Phases 2 and 3 planned for flexible, demand-driven expansion in the coming years

NVIDIA DGX B200 SuperPOD Powers MediaTek's AI Ambitions

The centerpiece of MediaTek's new data center is the NVIDIA DGX SuperPOD system built on the DGX B200 platform. NVIDIA's DGX B200 represents one of the company's most powerful AI infrastructure solutions, built around the Blackwell B200 GPU architecture that Jensen Huang unveiled in 2024. Each DGX B200 system packs 8 Blackwell GPUs delivering unprecedented performance for AI training and inference workloads.

The DGX SuperPOD configuration scales these systems into a unified, high-bandwidth cluster capable of handling massive parallel workloads. For MediaTek, this means dramatically accelerating chip design simulations, AI model training for on-device intelligence, and next-generation semiconductor research.

This deployment is particularly significant because MediaTek is not just a consumer of AI — it is one of the world's largest designers of AI-capable mobile processors. The company's Dimensity series of smartphone chips increasingly relies on neural processing units (NPUs) and on-device AI capabilities, which require extensive training infrastructure to develop and optimize.

Immersion Cooling Sets New Efficiency Standard in Taiwan

Traditional air-cooled data centers typically operate with PUE values between 1.4 and 1.6, meaning that for every watt consumed by computing equipment, an additional 0.4 to 0.6 watts goes toward cooling and overhead. MediaTek's facility shatters this benchmark with a PUE of just 1.1, achieved through its pioneering use of immersion cooling technology.

Immersion cooling submerges server components directly in a thermally conductive, non-conductive liquid, which absorbs heat far more efficiently than air. This approach is becoming essential for next-generation AI hardware like the Blackwell B200, which generates substantially more heat than previous-generation GPUs due to its higher transistor density and power consumption.

MediaTek's decision to deploy immersion cooling at scale is notable because the technology, while proven in smaller deployments by companies like Microsoft, Google, and specialized operators like GRC and LiquidCool Solutions, has not yet seen widespread adoption in Taiwan's data center ecosystem. By moving first, MediaTek positions itself as both a technology leader and an environmental steward in the region.

The facility also utilizes recycled water as the primary cooling medium for its air conditioning systems, with partial integration into the broader cooling infrastructure. This reduces the data center's dependence on fresh water — a growing concern as AI infrastructure expands globally.

Sustainability Features Go Beyond Cooling

MediaTek's green credentials for the Miaoli Tongluo facility extend well beyond its cooling architecture. The company has installed a 235 kW rooftop solar panel array that generates enough electricity to meet the annual energy needs of approximately 67 households.

While 235 kW represents a modest fraction of the total power consumed by a DGX SuperPOD deployment — a single DGX B200 system alone can draw upwards of 14.3 kW — the solar installation demonstrates MediaTek's commitment to offsetting its carbon footprint through renewable energy generation. It also aligns with Taiwan's national renewable energy targets, which aim for 20% renewable power by 2025.

The facility's power infrastructure deserves special attention as well. MediaTek describes it as a 'fab-grade' power supply system, borrowing terminology from semiconductor manufacturing. In chip fabrication, even microsecond-level power interruptions can destroy entire wafer batches worth millions of dollars. Applying this same standard to a data center ensures:

• Uninterruptible power supply (UPS) with redundant backup systems
• Multiple independent power feeds to eliminate single points of failure
• Precision voltage regulation to protect sensitive GPU hardware
• Rapid failover capability measured in milliseconds, not seconds

This level of power reliability is critical for AI training workloads, where a single interruption can corrupt days or weeks of model training progress.

Why MediaTek Needs Its Own AI Supercomputer

MediaTek's investment in a dedicated AI data center reflects a broader trend across the semiconductor industry. Chip designers increasingly need massive compute resources not just to design chips, but to develop the AI software stacks that run on those chips.

The company's core business — designing system-on-chip (SoC) processors for smartphones, smart TVs, IoT devices, and automotive platforms — now requires deep AI expertise at every level. Consider the AI-related challenges MediaTek faces:

• On-device AI model optimization: Training and compressing large language models to run efficiently on mobile NPUs
• EDA (Electronic Design Automation) acceleration: Using AI to optimize chip layouts, reduce power consumption, and improve yield
• Computer vision and generative AI: Developing edge AI capabilities for cameras, displays, and multimedia processing
• Automotive AI: Building perception and decision-making models for advanced driver assistance systems (ADAS)

Competitors like Qualcomm, Samsung, and Apple all maintain significant internal AI compute infrastructure. MediaTek's DGX SuperPOD deployment helps close the gap, particularly as the company expands into premium smartphone segments and automotive markets where AI performance is a key differentiator.

How This Fits Into the Global AI Infrastructure Race

MediaTek's data center launch arrives during an unprecedented global buildout of AI compute infrastructure. Hyperscalers like Microsoft, Google, Amazon, and Meta are collectively spending over $200 billion on data center capital expenditure in 2025 alone. Meanwhile, sovereign AI initiatives across Asia, Europe, and the Middle East are driving demand for localized compute capacity.

In Taiwan specifically, the AI infrastructure landscape is evolving rapidly. TSMC, the world's largest contract chip manufacturer and MediaTek's primary foundry partner, has been expanding its own AI and high-performance computing capabilities. The Taiwanese government has also launched initiatives to position the island as a hub for AI development, not just chip manufacturing.

MediaTek's facility represents a different dimension of this trend — a fabless chip designer building its own AI supercomputing infrastructure rather than relying solely on cloud providers. This approach offers several strategic advantages:

• Data sovereignty: Keeping proprietary chip designs and AI models on-premises
• Cost efficiency: Avoiding long-term cloud computing expenses for sustained workloads
• Customization: Tailoring the compute environment to specific EDA and AI training requirements
• Competitive secrecy: Reducing exposure of sensitive R&D data to third-party platforms

Looking Ahead: Phases 2 and 3 on the Horizon

MediaTek has confirmed that phases 2 and 3 of the Miaoli Tongluo data center will be developed according to actual demand, suggesting a flexible, modular expansion strategy. This approach mirrors the practices of major cloud providers who build data center capacity in phases to match workload growth.

Given NVIDIA's rapid product cadence — with the Blackwell Ultra and next-generation Rubin architectures already on the roadmap — future phases of MediaTek's facility could incorporate even more powerful hardware. The company's willingness to adopt cutting-edge platforms like the DGX B200 suggests it will continue to deploy the latest available technology as it scales.

The broader implication is clear: AI infrastructure is no longer optional for major semiconductor companies. Whether designing chips, training models, or optimizing software stacks, the companies that control their own high-performance compute resources will have a decisive advantage in the AI era. MediaTek's Miaoli Tongluo data center is a concrete bet on that future — and with NVIDIA's most powerful platform at its core, it is a bet backed by serious hardware.