Taiwan’s AI Future May Hinge on a New Metric: ‘Useful Compute’

TAIPEI, Taiwan – July 07, 2026 – The global AI industry is running on an explosive, almost unquenchable, thirst for computing power. With demand for AI services having skyrocketed—by some measures over 300-fold—the world’s largest technology firms are pouring hundreds of billions into building out data center capacity. Yet, a critical question is emerging from the heart of the world's semiconductor supply chain: what if we are measuring this new gold rush all wrong?

At a pivotal seminar in Taipei, AI computing firm Zettabyte, in collaboration with Taiwan’s legendary Industrial Technology Research Institute (ITRI), made a compelling case that the industry’s obsession with raw hardware specifications and hourly rental costs is dangerously misguided. The argument is simple but profound: the true value of AI infrastructure lies not in how much hardware you have, but in how much useful work that hardware actually produces. It’s a shift in perspective that could unlock billions in savings and form the bedrock of Taiwan's ambition to build a sovereign AI ecosystem.

The Currency of Compute: A Crisis of Measurement

The AI industry has long spoken a language of brute force metrics. Success is often quantified by the number of GPUs in a cluster, the theoretical processing power measured in floating-point operations per second (FLOPs), and the hourly cost to rent that power. However, these metrics are proving to be a poor proxy for value.

“The AI industry has measured infrastructure primarily through two metrics: GPU count and hourly rental cost. Yet neither metric answers the core question that matters most to customers,” said Jeff Lin, Executive Director of Zettabyte, at the AI Server Supply Chain Seminar. “Paying customers want to know how much ‘useful’ AI is being produced by the GPUs, and to be able to reliably measure this.”

This disconnect is more than academic. Zettabyte contends that reliability and efficiency gaps—periods where GPUs are spinning but not producing valuable output due to software bottlenecks, data pipeline issues, or failed jobs—can drive up to 15 percent of the total cost of ownership. In an industry where major cloud providers like AWS are planning capital expenditures of $200 billion in a single year, predominantly for AI, that 15 percent represents a colossal waste of capital and energy.

Zettabyte’s proposed solution is its zSUITE platform, which introduces the concept of “goodput” to AI infrastructure. Borrowed from networking, the term distinguishes useful data from mere traffic. Applied here, it measures the actual, valuable computational output that contributes to an AI model’s goal. This allows operators to see beyond a simple “100% utilization” metric—which can often hide profound inefficiencies—and pinpoint the true cost and productivity of their AI investments. This focus on efficiency arrives as analysts note the industry's bottleneck is shifting from building enough compute to effectively utilizing it for complex AI systems.

Forging Sovereignty from Silicon and Software

Zettabyte’s campaign for a new standard is not happening in a vacuum. Its choice of partner and location—ITRI in Taipei—is deeply strategic. Founded in 1973, ITRI is the government-backed institution that effectively birthed Taiwan’s semiconductor dominance, incubating titans like TSMC and UMC. Its pivot toward shaping AI standards signals a new national priority.

“Taiwan's strength in semiconductors and AI hardware is well established. Developing rigorous standards for how compute is measured is a logical next step, and the kind of work ITRI is positioned to support,” stated Stephen Su, Senior Vice President at ITRI.

This collaboration is a cornerstone of Taiwan’s burgeoning “sovereign AI” strategy. In an era of intense geopolitical competition over technology, nations are increasingly seeking to build and control their own critical AI infrastructure to ensure economic competitiveness and national security. For Taiwan, which produces the vast majority of the world’s advanced logic chips, developing an independent and highly efficient software and systems layer for AI is a crucial move to secure its leadership position.

Zettabyte is embedding itself directly into this national project. The company is advancing software licensing for zSUITE for sovereign use and partnering with ITRI and top Taiwanese universities. The goal is to cultivate a new generation of local AI infrastructure talent that understands how to build, measure, and optimize these complex systems from the ground up. By putting its measurement tools into the hands of researchers, Zettabyte is helping build the local expertise on which a truly sovereign AI capability depends.

The Trillion-Dollar Efficiency Gap

The financial implications of measuring useful compute are staggering. Global spending on AI is forecast by IDC to hit $2.52 trillion in 2026, with AI infrastructure alone contributing over $400 billion. If the 15% efficiency gap Zettabyte identifies holds true at that scale, the industry could be wasting over $60 billion annually on underperforming infrastructure.

Beyond the direct financial cost, this inefficiency has a significant environmental footprint. AI’s power consumption is a growing crisis, with some hyperscale data centers now consuming as much electricity as a small city. Optimizing for “goodput” means less wasted compute, which translates directly into less wasted energy and a more sustainable growth trajectory for the entire AI sector.

As the AI arms race intensifies, the competitive landscape is evolving. The initial scramble was for hardware, leading to year-long lead times for NVIDIA’s top-tier GPUs. Now, a new battleground is emerging in the software layer that manages and optimizes that hardware, a space populated by MLOps and AI observability platforms. Zettabyte’s focus on a verifiable “useful work” metric offers a compelling differentiator, moving the conversation from process monitoring to outcome measurement.

By anchoring its strategy in Taiwan’s sovereign AI ambitions, Zettabyte is making a long-term bet that the future of AI will be defined not just by those who build the most powerful chips, but by those who can wield them with the greatest efficiency.