Global AI Competition: The Overlooked Power of Non-Semiconductor Infrastructure

The global race for AI dominance has long fixated on semiconductors, with companies like Nvidia dominating headlines. Yet, beneath this silicon-centric narrative lies a critical, underappreciated layer of infrastructure that will determine the winners and losers in the AI era. From 2023 to 2025, the non-semiconductor components of AI infrastructure-ranging from data center cooling to cybersecurity frameworks-have emerged as both a bottleneck and an opportunity. These elements, though less glamorous, are essential for scaling AI workloads sustainably, securely, and efficiently.

The Energy and Cooling Challenge

AI's insatiable appetite for computational power is straining global energy grids. According to a report by Gartner, data center electricity consumption rose 80% from 2020 to 2025 and is projected to double by 2030, per the Global AI Race analysis (see visual below). This surge is driven by AI's reliance on high-density computing, which generates unprecedented heat. Over 35% of AI-centric data centers now deploy liquid cooling solutions to manage thermal loads, a trend accelerated by chips like Nvidia's H100, which demand advanced thermal management due to their power consumption, according to Flexential's 2025 report.

Meanwhile, the U.S. power grid, already strained by decarbonization efforts, faces a dual challenge: meeting AI's energy demands while integrating renewable sources. AI tools are being leveraged to optimize grid operations, from forecasting supply-demand imbalances to streamlining renewable interconnections, as reported by MIT Technology Review (https://www.technologyreview.com/2025/09/09/1123404/ai-grid-help/#:~:text=The%20rising%20popularity%20of%20AI%20is%20driving%20an). However, as one expert cautions, infrastructure growth to support AI has outpaced the technology's ability to deliver transformative grid benefits, the piece argues.

Networking and Power Distribution: The Hidden Bottlenecks

Ethernet is rapidly overtaking InfiniBand as the networking backbone for AI, thanks to innovations like RDMA over Converged Ethernet (RoCEv2) and Broadcom's Scale-Up Ethernet framework. These advancements enable low-latency, high-throughput connections across thousands of GPUs, a necessity for training large models, according to NextGenInfra (https://nextgeninfra.io/2025-dc-network-ai/). Yet, 59% of organizations report bandwidth issues, and 53% face latency challenges, underscoring the urgency of upgrading networking infrastructure, Deloitte's report finds (https://www.deloitte.com/us/en/services/consulting/blogs/ai-adoption-challenges-ai-trends.html).

Power distribution is another critical frontier. As data centers shift toward renewable energy, construction firms are deploying physical AI to automate tasks like land leveling and solar panel installation. For instance, Xpanner's AI-based autonomy kits have streamlined the development of 400 MW solar farms, addressing labor shortages and accelerating deployment, as covered in a Forbes article (https://www.forbes.com/sites/sabbirrangwala/2025/10/01/physical-aibuilding-the-energy-infrastructure-for-data-centers/). Such innovations are vital for reducing reliance on traditional grids and ensuring AI's energy needs are met sustainably.

Cybersecurity: The Unseen Frontline

AI's expansion has also amplified cybersecurity risks. A Flexential survey reveals that 55% of organizations report increased exposure to cyber threats due to AI's data intensity. Frameworks like NIST's AI Risk Management Framework and Microsoft's AI Security Framework are gaining traction, offering structured approaches to safeguarding models and training data, according to Practical DevSecOps (https://www.practical-devsecops.com/best-ai-security-frameworks-for-enterprises/). However, 86% of enterprises struggle to acquire talent for implementing these frameworks, creating a gap between policy and practice, Deloitte reports.

The EU AI Act and DORA further complicate the landscape, mandating stringent cybersecurity standards for critical infrastructure. Companies that invest in adaptive defenses-such as AI-driven threat detection-will gain a competitive edge. For example, Databricks' AI Security Framework (DASF) integrates 64 controls tailored to diverse models, offering a blueprint for secure AI deployment.

Investment Opportunities Beyond the Chip

While semiconductors remain central, investors should look beyond silicon. Infrastructure software vendors like Broadcom and Cadence Design Systems are poised to benefit from custom chip demand, with AI projected to account for 40% of Broadcom's revenue by 2026, according to CNBC (https://www.cnbc.com/2025/07/10/buy-nvidia-broadcom-and-these-other-underappreciated-ai-plays-goldman-sachs-says.html). Similarly, utilities and cooling technology firms-though currently undervalued-are essential for sustaining AI's growth.

The shift from cloud reliance to in-house AI solutions also opens opportunities. Edge AI processors, which enable real-time decision-making in power-constrained environments, are gaining traction. Goldman Sachs highlights these "underappreciated plays" as key to the next phase of AI infrastructure investment (https://www.goldmansachs.com/insights/articles/ai-infrastructure-stocks-poised-to-be-next-phase).

Conclusion

The global AI competition is no longer just about who can design the fastest chip. It is about who can build the resilient, scalable, and secure infrastructure to support AI's exponential demands. From liquid cooling systems to adaptive cybersecurity frameworks, the non-semiconductor components of AI infrastructure are the unsung heroes of this race. For investors, these overlooked sectors represent both a challenge and a golden opportunity-one that will define the next decade of technological progress.