Efficiency vs. Growth AI: Let’s not choose the lowest common denominator

The Efficiency Trap: Why Enterprise AI Risks Missing America’s Growth Moment

The first chapter of enterprise AI is being written in the language of cost control, and that should worry us. Boardrooms are fixated on “headcount optimization,” automation pipelines, and margin expansion. But treating AI as a labor-replacement tool isn’t just a strategic misreading of the technology. At the moment when the U.S. most needs a new engine of growth, it risks squandering one.

We often hear comparisons to past technological transformation to rationalize embracing rather than fighting AI’s proliferation. But there’s a nuance. From electrification to the internet, general-purpose technologies have delivered outsized gains not by substitution, but through expansion. Yes, ATMs reduced the need for bank tellers. But their real advantage came from banks’ ability to extend locations, hours, functions, safety, speed, and capabilities. It was companies those that recognized and harnessed the opportunities ATMs provided that saw real adoption gains, expanding network coverage and increasing customer profitability through decreased wait times for higher margin services and improved customer experience and retention.

The Economist has covered this nuance extensively through its Babbage and Boss Class podcasts: past technology waves that concentrated gains in efficiency rather than innovation ultimately produced sluggish aggregate growth and deepened inequality. Productivity, in classical growth accounting, is driven by the creation of new goods arising from technological innovation, not by doing the same work with fewer people.

Unfortunately, today’s enterprise AI strategy is almost entirely focused on shrinking the denominator.

A Capital-Intensive Paradox

Large tech firms are pursuing one of the most capital-intensive investment cycles in modern history (data centers, GPUs, energy infrastructure) while facing investor pressure for near-term returns. That combination creates a bias toward measurable efficiency gains over harder-to-quantify growth bets, and this plays out in corporate adoption.

Empirical evidence suggests this is short-sighted. The Wharton Budget Model projects that AI could increase U.S. GDP by ~1.5% by 2035 under current deployment trajectories, dwindling to negligible long-run annual impact. Other takes, such as from WaPo and Goldman Sachs, are even more negative. The AI Daily Brief’s coverage of enterprise adoption surveys has consistently flagged a gap between AI investment and measurable output with firms racing to integrate AI into workflows without a coherent theory of value creation. The implication is not that AI lacks transformative potential. It’s that the current deployment model is failing to capture it.

The “Jobless Productivity” Problem

There is mounting concern, echoed in recent coverage of macro data, that the U.S. is entering a phase of jobless growth. Productivity gains may be real, even accelerating; but the evidence suggests they’re being achieved through substitution rather than complementarity. Workers face wage scarring and occupational downgrading, rather than opportunity creation. This is a uniquely problematic challenge in a democratic capitalistic system: if productivity gains are decoupled from labor, what is the impact on our expectation that productivity investments translate to worker gains? Can labor be effectively reallocated? AI is not outsourcing, but it is falling into its narrative trap.

This matters because innovation-led growth historically depends on labor reallocation into higher-value activities. If AI compresses labor demand without expanding new domains of work, it erodes the consumption base and dynamism that sustain growth. There is so much opportunity here, and we are seeing it manifest in capacity constrained small business and social enterprises. It’s big business’s turn to investigate its AI strategy.

The Bigger Picture

For those interested in the global landscape, these stakes extend well beyond individual firms. If the US AI ecosystem remains anchored in efficiency, we may see high capital intensity, low growth elasticity, and a widening gap between AI’s hype and its macroeconomic contribution. That would spell bad news for public support for AI innovation and investment, flawed adoption strategies, and a weaker US economy. And while the US fails to determine a near-term solution to our taxation system, which advantages CapEx investments over those in human capital, public negativity is unsurprising.

It also cedes the innovation frontier to those pursuing more expansive and long-term strategies. Chinese AI deployment from companies such as DeepSeek and in robotics is instructive: unconstrained by quarterly returns, willing to pursue long-horizon product bets. American economic leadership has historically been defined not by incremental efficiency, but by the commercialization of innovation.

A Reframed Enterprise AI Agenda

The path forward is not to abandon efficiency, but to shift the strategic balance toward growth and transformation. Electrification is the canonical example for a goal end-state: it delivered limited gains until firms reorganized production around it, moving away from centralized steam power toward distributed electrical systems. AI’s analogy: productivity gains will not come from incremental substitution but from structural reinvention — new products, new business models, new markets.

Enterprises need a new playbook: treating technology not as a tool to do less with less, but as a platform to do more with more. The current efficiency-first approach may be the least efficient path of all, optimizing locally while underperforming globally. The window to get this right is not indefinite. The capital is being deployed now, the organizational habits formed, the macroeconomic narrative, which shapes policy, talent flows, and public legitimacy, being written. Enterprises — and countries — that treat this moment as a cost story will find themselves, a decade from now, having made a very expensive bet on the wrong question.