AI infrastructure investment is forecast at $2.5 trillion for 2026 alone. Electricity prices have jumped -- double the 2.9% headline inflation rate. Memory chip prices have doubled. Data centers will account for 40% of electricity demand growth through the end of the decade. Meanwhile, less than 10% of companies use generative AI, and 95% of organizations say they've seen no measurable return on AI. Will AI productivity arrive before the spending boom breaks something?
1. AI Will Crush Prices (Tech and Finance Optimists)
Productivity gains from AI will far outweigh infrastructure costs and solve inflation through supply-side growth.
AI will compress labor costs in knowledge work. Sam Altman says that by the end of 2026, $1,000 spent on AI could enable one person to complete a software project that previously required a team. Legal services, medical diagnostics, financial analysis -- all become "radically cheaper." As these systems scale, "money becomes more valuable" in real terms.
AI will vastly boost GDP and productivity. Goldman Sachs thinks AI could raise global GDP by 7% and lift productivity growth by 1.5 percentage points over a decade. McKinsey quantifies the direct opportunity at $2.6-4.4 trillion annually, with 75% of value in customer operations, marketing, software engineering, and R&D -- all high-labor-cost sectors where AI scales fastest. The BIS found that AI adoption is initially disinflationary through supply expansion: productivity rises, supply outpaces demand, prices fall.
The live examples are already stacking up. Maersk cut vessel idle time 30% through AI. Amazon runs 520,000 AI-powered robots, trimming order-processing costs 20%. Walmart saves $1.5 billion annually with AI inventory management. Federal Reserve Dallas research confirms AI "increases productivity more for less-experienced workers," reducing unit labor costs -- the backbone of services inflation.
2. The Bill Is Due Now, the Payoff Is a Maybe (The Let's-Sees)
AI infrastructure spending is inflating prices right now. Productivity gains are speculative and years away.
AI applies to only about 5% of the economy. Nobel laureate Daron Acemoglu projects 0.53% total productivity gain and roughly 1% GDP growth from AI over the next decade -- a fraction of the tech industry's forecasts. AI's current direction is "wrong," he says -- too much automation, not enough worker augmentation. The hype about doubling GDP growth "lacks theoretical or empirical support."
AI added "basically zero" to US GDP in 2025. That's Goldman Sachs' own chief economist Jan Hatzius. Only 9.3% of companies reported using generative AI in production in recent weeks. Goldman forecasts meaningful impact starting around 2027 -- but that's a forecast, not a fact. Meanwhile, firms are on track to spend $2.9 trillion on AI capex from 2025-2028. If adoption stays slow that's trillions spent on infrastructure that doesn't yet drive productivity. Pure demand-side inflation.
The "J-curve" adoption pattern means years of cost before any return. Manufacturing firms adopting AI show initial performance dips before outperforming peers four or more years later. But most firms lack the complementary assets -- data infrastructure, technical talent, organizational redesign -- to reach that inflection point. A CEO survey of 6,000 executives in February 2026 found the vast majority see "little to no impact" from AI on operations, resurrecting the Solow Paradox: we see AI everywhere except in the productivity statistics.
3. Who Knows (Philosopher-Economists & The Fed)
AI's inflation outcome hinges on expectations, adoption speed, and whether central banks can thread the needle.
AI's inflation dynamics are expectations-driven, not predetermined. --BIS. If firms and households anticipate the productivity boost, they spend ahead of it -- and inflation rises immediately as demand surges before supply catches up. If we don't think productivity gains are coming, AI adoption is initially disinflationary, with moderate inflation emerging gradually. There is no single "AI deflation" or "AI inflation" path. The outcome depends on what the market believes about the future.
The path will probably be non-linear. At least that's what the IMF's "Mind the Gap" working paper models. Short-run: inflation rises modestly as firms invest heavily and households raise consumption on expected future gains. Medium-to-long-run: inflation declines as AI-driven productivity improves supply capacity. Global total factor productivity rises 1.8% in five years, 2.4% in ten -- in the high-growth baseline. But the transition period is the danger zone.
It may be too soon to tell. Fed Vice Chair Jefferson acknowledged both directions in February 2026: "More efficient allocation of resources and supply chain improvements could reduce costs, leading to lower prices," he said. But AI adoption may surge aggregate investment, making prices rise. The Fed is watching, not predicting.
Where This Lands
Altman says prices are about to collapse. Acemoglu says the productivity gains are a rounding error. The BIS and IMF say it depends on whether we believe our own hype. What we know for certain: electricity is up 6.9%, chips have doubled, and trillions are being spent on infrastructure where 95% of organizations report no measurable return on their AI investments. The deflation may come. But the inflation is already here. And the gap between the two is where the real economic risk lives.