By Business & Finance 
Jack McCauley, Board Trustee and Innovator in Residence at the University of California at Berkley, spoke at the Dublin Tech Summit 2025. His session topic focused on exploring the journey of AI.
Since my talk at DTS a year ago, one shift has risen above all others in transforming how neural networks are developed and deployed: energy infrastructure. We often think of AI progress in terms of model size, training speed, or accuracy gains. But behind every large language model, vision system, or recommendation engine lies a voracious appetite for power.
The massive data centres that train and deploy these neural networks are not just technological marvels; they’re energy hogs. And as models grow in scale, so too does their electricity consumption. This has turned energy availability – its source, cost, and reliability – into one of the most critical factors in AI development.
The data centre boom and the search for power
Enter locations like Henderson, Nevada, and the greater Las Vegas area. These regions are becoming prime targets for data centre development because of their proximity to Lake Mead and regional hydroelectric resources. The Hoover Dam, fed by the Colorado River, provides a stable source of renewable hydroelectric power, exactly the kind of clean energy that AI infrastructure needs.
Nevada’s relatively open land, favourable regulatory environment, and access to water for cooling make it an ideal hub for AI scalability. Meanwhile, states like California, despite being home to many of the world’s leading AI companies, are facing a stark reality: they simply don’t have the grid capacity to meet growing AI energy demands.
California’s energy paradox
California has long been a leader in climate policy, mandating a transition to all-electric vehicles and setting aggressive clean energy targets. Yet, the state’s power grid is strained. During heat waves, rolling blackouts are not uncommon. The grid struggles to keep up with residential and commercial demand, let alone the exponential load of AI data centres.
Imagine trying to power a fleet of electric cars while simultaneously running data centres that consume as much electricity as small cities. It’s a recipe for grid instability.
The gap between ambition and infrastructure is pushing AI developers beyond Silicon Valley, suggesting that the future of AI may not be built in California but powered in Nevada.
Has the rise of modern tensor processors made past AI techniques obsolete?
The field of artificial intelligence and high-performance computing has undergone transformative shifts in recent years, none more impactful than the rapid evolution of tensor processing technology.
In 2022, many AI practitioners and organisations were still optimising models using conventional GPU-accelerated computing frameworks. However, the accelerated development of specialised Tensor Processing Units (TPUs), led by industry pioneers like NVIDIA, has redefined what’s possible in terms of computational efficiency, scalability, and cost.
From GPUs to tensor-centric architectures
While modern tensor processors are rooted in the architecture of Graphics Processing Units (GPUs), originally designed for computer-aided design (CAD) and gaming, their role has been completely reimagined.
What began as parallel processing engines for rendering pixels is now a purpose-built powerhouse optimised for the massive matrix operations that underpin deep learning. NVIDIA’s advancements in this space, most notably with the introduction of Ada Lovelace and Hopper architectures, have pushed tensor cores to new performance heights.
These innovations deliver significantly higher throughput for mixed-precision computations (FP16, BF16, INT8, and even INT4), enabling faster training and inference at lower power and cost per operation.
Looking ahead: the future of neural networks and the workforce
As neural networks continue to evolve at a rapid pace, their transformative impact on industries and the global job market is impossible to ignore.
Over the next 1–2 years, one of the most significant developments and challenges we can anticipate is the large-scale integration of AI into mid-level white-collar roles. This shift promises increased efficiency and scalability but also raises important questions about workforce adaptation and economic equity.
Neural networks are poised to automate a wide range of cognitive tasks previously thought to require human judgment and expertise. Customer support, sales and market research roles relying heavily on data analysis, pattern recognition and routine decision-making are particularly vulnerable. AI-driven systems can handle customer inquiries with near-human responsiveness, generate personalised sales strategies from vast datasets, and deliver real-time market insights with unprecedented speed and accuracy.
About the author: Jack McCauley is a Board Trustee and Innovator in Residence at the University of California at Berkley.
Read more:
Dublin Tech Summit 2026 announces keynote speakers
AI goes rogue: Managing autonomous threats in the real world
DEI Catch-22: Business vs. Belief Debate at Dublin Tech Summit
