The AI Economy

Image on Unsplash

When tech companies announce AI breakthroughs, they show you the models, the capability gains, the benchmark scores and the roadmap to AGI. What they do not show you is the $7 trillion that goes behind it. 

That is the estimated cost to build out the data center infrastructure needed to support AI through 2030 (McKinsey, 2025). Alphabet alone is planning to spend $91-93 billion in 2025 on infrastructure, primarily for AI. Microsoft spent $34.9 billion in capital expenditures last quarter (AI Media House, 2025). Meta is planning to spend $600 billion on US infrastructure through 2028 (TechCrunch, 2025). These are not research budgets or talent acquisition costs. These are just the costs for hardware, real estate, and electricity.

The physical reality of AI is hitting hard. Data centers are projected to consume nearly 260 terawatt-hours of electricity by 2026, up from 200 terawatt-hours in 2022, using roughly 4-5% of total US national power demand (AI Media House, 2025). That is not an afterthought or an implementation detail, but a transformation of the energy grid itself. Utilities and regulators are already rethinking expansion plans to accommodate the surge.

Goldman Sachs projects annual AI infrastructure investment will reach $200 billion by 2027, with about half going toward cloud hardware and the rest toward energy, land, and chips (AI Media House, 2025).

Here is what makes this interesting from a labor perspective. While infrastructure spending is scaling vertically, salaries in some markets are moving the opposite direction. Median compensation for engineering and data roles in India dropped 40% in 2025, from $36,000 to $22,000 annually, according to analysis from Deel and Carta. In the US, median compensation for similar roles surged 23% to over $150,000 (Times of India, 2025). That divergence is a structural realignment of the global labor market driven by AI adoption.

The reason is straightforward. AI automation displaced routine work, and the cost-arbitrage model that sustained India's tech economy for decades is evaporating. When companies can use AI to automate parts of development that junior engineers once handled, they no longer need as many junior roles, and they no longer need to offshore that work to save money. 

Simultaneously, companies in the US are bidding aggressively for scarce talent in AI, machine learning, and cloud architecture, driving compensation up. The result is a widening chasm between two geographies that marched in step during the previous era.

So, what is the actual employment impact? Research is still going on, but the early signs are concerning. Goldman Sachs estimates AI could displace 6-7% of the US workforce if widely adopted, though that impact is expected to be transitory as new opportunities emerge elsewhere. However, younger tech workers are already bearing the brunt. Unemployment among 20 to 30-year-olds in tech-exposed occupations has risen by almost 3 percent since the start of 2025, notably higher than for same-aged counterparts in other trades (Goldman Sachs, 2025). 

A Stanford paper found early-career workers in the most AI-exposed occupations have experienced a 13% decline in employment relative to less exposed occupations (Econofact, 2025). 

The occupations at highest risk are computer programmers, accountants and auditors, legal and administrative assistants, customer service representatives, telemarketers, proofreaders and copy editors, and credit analysts. These are not jobs that require retraining into entirely different fields. They are jobs that exist in a grey zone where AI can do part of the work convincingly but not all of it. That means companies can either cut headcount or restructure roles to supervise and audit AI output instead of generating it from scratch (Goldman Sachs, 2025). The second option sounds better until you realize it is usually more efficient to just cut headcount.

Meanwhile, the jobs created by AI so far are concentrated at the top. New roles include AI ethics officers, prompt engineers, human-AI collaboration specialists. These positions require advanced degrees, and 77% of new AI jobs require master's degrees according to the research (SSRN, 2025). That creates a specific kind of labor market problem. You have displacement in middle-income knowledge work and creation in senior, specialized roles. The transition path for someone laid off from a data entry position is not obvious at all.

What makes infrastructure spending relevant to this conversation is that the capital being poured into AI systems is not flowing back into hiring. Alphabet spent $91-93 billion on infrastructure in 2025 and laid off thousands of workers. Meta spent tens of billions on data centers while conducting multiple rounds of workforce reductions (AI Media House, 2025). The business model seems to be building capacity first, automating labor second, and discovering what to do with the excess compute power third. That is a backwards order for an economy dependent on wage income.

The other detail nobody talks about is the energy constraint. If data center electricity consumption jumps from 200 to 260 terawatt-hours by 2026 and keeps climbing, at some point you hit a physical limit to how much compute you can deploy (AI Media House, 2025). Utilities cannot build out power plants as fast as tech companies want to build data centers. That creates pricing pressure on electricity, which flows into infrastructure costs, which flows into the return-on-investment calculations that justify all this spending (McKinsey, 2025). Nobody is talking about this ceiling because it sounds like a constraint, and the industry prefers to talk about possibilities rather than limitations.

The honest framing would acknowledge that the AI boom is now primarily an infrastructure boom, not a software boom. It is capital intensive, labor intensive on the specialized end, and labor-disruptive on the routine end. It has created a two-speed labor market where some people are pulling away dramatically while others are being pushed out. The companies doing the building are spending more money on hardware than on human beings. And the systems being built are consuming electricity at a rate that is starting to collide with real-world infrastructure.

Whether policymakers start treating this as a labor market transition that needs management or whether they wait until the disruption is obvious and then scramble to respond, is yet to be seen. India's tech workers are already living in the scramble version. That experience suggests it is not pretty.