Semiconductor Engineering frames energy efficiency as the defining constraint for AI computing through 2030

A Semiconductor Engineering analysis published in August 2026 argues that energy efficiency has become the single most consequential variable in AI computing - not just for data center operators, but for every enterprise running inference at scale^[1]. The framing is direct: reducing energy consumption reduces AI cost, because both problems trace back to the same root - AI systems are routinely operated well below maximum efficiency^[1].

The energy numbers are no longer abstract

The International Energy Agency estimates that data centers consumed roughly 460 TWh of electricity globally in 2025, and its base-case projection puts that figure at 945 TWh by 2030 - more than the entire current electricity demand of Japan^[1]. The IEA's updated analysis, published in 2026, confirms that electricity consumption from AI-focused data centers is growing even faster than the overall data center total, and is on track to triple over the same period.

Global data center electricity demand grew 17% in 2025, with AI-focused facilities climbing even faster - surging 50% in that year alone. Capital expenditure from five large technology companies surged to more than $400 billion in 2025 and is set to increase by a further 75% in 2026. The Semiconductor Engineering analysis puts the investment trajectory in starker terms: top players committed roughly $100 billion to data center infrastructure in 2020, a figure projected to reach approximately $1 trillion in 2026^[1].

The grid cannot absorb this at the current pace. Energy supply is not scaling at the pace of AI compute demand - gas turbines, the fastest path to new power capacity, are already booked through 2028, and power availability is emerging as a hard constraint on data center expansion.

Token economics are forcing the cost conversation

Daily token consumption in the US is projected to increase tenfold from roughly 200 trillion units per day today to approximately 2,250 trillion units per day by 2030^[1]. Token prices are falling - Gartner forecast in March 2026 that performing inference on a one-trillion-parameter LLM will cost providers over 90% less by 2030 than it did in 2025 - but usage is expanding faster than prices are dropping. Inference crossed roughly two-thirds of all AI compute in 2026, up from about one-third in 2023.

The result is that CFOs are already rationing AI budgets^[1]. The AI inference cost pressure of 2026 is not a temporary growing pain - it is a structural feature of the AI era that every enterprise deploying AI at scale must plan around.

Physical AI raises the stakes beyond the data center

The third pressure point identified in the analysis is physical AI - robots, drones, autonomous vehicles, and medical wearables^[1]. Unlike a cloud server that can draw from the grid, these devices run on batteries, making energy efficiency a hard constraint rather than a cost optimization^[1].

This shift is accelerating in 2026, with enterprises moving from pilots to scaled deployments in manufacturing, logistics, energy, and beyond. The default assumption that AI inference runs in the cloud is under pressure from two directions: the growth of physical AI applications, and the accelerating energy cost of data center compute - with physical AI expected to be majority edge.

The semiconductor response is already visible across the stack:

• Applied Materials' Subi Kengeri warned at the OCP APAC Summit on 11-12 August 2026 that AI's energy bottleneck cannot be solved at any single layer, and that co-optimization from chip materials to system architecture is required.
• The future of energy-efficient AI will be defined as much by DRAM memory and advanced packaging as by raw compute.
• Applied Materials' EPIC Center, representing a roughly $5 billion investment, is the largest commitment to advanced semiconductor equipment R&D in US history, targeting exactly this materials-to-system efficiency gap.

Whether the industry can bend the energy curve fast enough to keep pace with demand growth - and whether physical AI deployments can be powered efficiently enough to reach commercial scale - will determine which companies lead the next decade of AI.