It’s the Economy, Stupid: The Real Drivers of the Blue Wave
AI’s Next Bottleneck May Be Public Consent, Not Compute
We measure AI progress in parameters, tokens and benchmark scores. Citizens, however, may experience it through something far less glamorous: a higher electricity bill.
A recent CleanTechnica commentary attributes a major electoral shift primarily to inflation and household economics, while also identifying AI data centres and rising power costs as sources of public frustration. Leaving aside its political conclusions, the article surfaces a strategic issue technology leaders cannot ignore-the economics surrounding an AI system can become as important as the system itself.
Architecture Does Not End at the Data-Centre Fence
The industry often treats compute as an abstract, infinitely elastic resource. In reality, every training run and inference request sits atop a physical stack: generation capacity, transmission lines, cooling systems, water, land and fibre.
Data centres are not the sole cause of electricity inflation; grid economics are considerably more complex. But dismissing local impacts because the causes are distributed would be strategically naïve. If communities perceive that infrastructure costs are being socialised while AI’s financial gains remain concentrated, the result will not merely be public resentment. It will become permitting delays, stricter regulation, power-allocation constraints and reputational risk.
In other words, political economy is now part of systems architecture. An application that scales technically while creating unpredictable costs for its host ecosystem is not truly scalable.
From Model Performance to System Affordability
For CTOs, this demands a broader definition of performance. Accuracy and latency remain essential, but they must be evaluated alongside energy intensity, infrastructure volatility and cost per useful outcome.
I believe the next phase of enterprise AI will reward architectural restraint as much as model ambition. Not every customer query requires a frontier model. Model routing, smaller domain-specific models, retrieval, caching, batching and well-designed deterministic workflows can often deliver better economics without materially degrading user experience.
Workload placement also becomes a multi-dimensional decision. A region that appears attractive on GPU price may carry higher egress costs, weaker power availability, data-residency complications or greater exposure to regulatory change. Conversely, edge inference may reduce latency and data movement while introducing device-management and observability challenges.
The correct response is not a simplistic “cloud versus edge” debate. It is to design portable workloads, explicit service-level objectives and FinOps practices that incorporate energy and infrastructure risk. The cheapest token is useful; the cheapest successfully completed business process is far more valuable.
India’s AI Ambition Needs an Affordability Layer
This lesson is particularly relevant to India. Our Digital Public Infrastructure demonstrates the power of shared, interoperable rails that reduce transaction costs at population scale. AI should extend that philosophy-not quietly make essential digital services more expensive to operate.
A citizen-facing AI service with volatile inference costs is an unfinished public system. Likewise, data-centre expansion must be assessed alongside grid readiness, water stress, renewable availability and the question of who pays for upstream upgrades.
For the Northeast, the answer need not be a race to build hyperscale campuses everywhere. A more resilient approach could combine shared regional compute, efficient local inference, stronger connectivity and low-power systems designed for real operating conditions. Local-language services and agricultural or governance applications should be judged by public value delivered per unit of infrastructure consumed.
Four Questions for the Boardroom
- Unit economics: Do we know the fully loaded cost per successful AI-enabled outcome, including network, energy and operational overhead?
- Right-sized intelligence: Can simpler models or conventional software handle part of this workload more reliably?
- Infrastructure accountability: Have we assessed power, water, residency and community impacts-not merely GPU availability?
- Portability and resilience: Can workloads move when energy prices, regulations or supply conditions change?
AI will not fail only because models hallucinate. It can also fail because its surrounding infrastructure becomes unaffordable or socially untenable.
The next generation of great architects will be judged not just by how intelligently their systems think, but by how lightly-and fairly-they allow society to carry them.
About the Author: Sanjeev Sarma is the Founder Director and Chief Software Architect at Webx Technologies. With a core focus on Generative AI integration, Cloud-Native Scalability, and Enterprise Software Architecture, he has spent over two decades driving digital transformation across Northeast India and beyond. Beyond his corporate leadership, Sanjeev is deeply invested in shaping the future of the IT industry. He serves as an Industry Expert on the Board of Studies for Assam Don Bosco University’s School of Technology, advises state technology committees, and actively mentors emerging tech startups at STPI. He brings a unique, dual perspective of high-level enterprise execution and future-ready academic curriculum development.