Goldman Sachs: AI Consumed $700B in 2025 — the GDP Return Was “Basically Zero”

The headline is stark: Goldman Sachs: AI Consumed $700B in 2025 — the GDP Return Was “Basically Zero.” Chief Economist Jan Hatzius delivered this sobering data point recently. Despite a historic capital investment surge, artificial intelligence added almost nothing to American economic growth last year.

This is not a critique of long-term potential. Instead, it highlights a structural accounting problem. The issue reveals where money actually goes and who truly benefits from current spending trends.

Why Goldman Sachs: AI Consumed $700B in 2025 — the GDP Return Was “Basically Zero”

Hatzius was direct about widespread misreporting regarding economic impact. AI investment contributed only 0.2 percentage points to 2025 growth. However, the real story involves how GDP measures domestic production rather than spending.

When tech companies buy infrastructure, much of that spending flows to imported components. GPUs and specialized hardware often come from overseas manufacturers. Consequently, roughly 75% of data center costs involve imported parts.

Therefore, US investment adds to Taiwanese and Korean GDP instead. Domestic spending and import outflows cancel each other out in calculations. Tech firms spent billions, yet the US economy registered almost none of it as growth.

Inflation Pressures and Rising Costs

The GDP story is only part of the complex picture. Separate research found AI-related price pressures already added to core inflation. Similar pressure is forecast for the next twelve months.

Three channels drive this inflation trend. First, surging demand pushes up prices for digital memory and electronics. Second, tight labor markets create wage pressure for skilled engineers.

Third, data centers consume enormous amounts of electricity. Output rose significantly in 2024 and 2025 due to this demand. As a result, consumer electricity prices are up 4.6% year-over-year.

Analysts call this a macro regime shift. Tech goods prices are now rising faster than wages for the first time in decades. Paradoxically, AI currently acts as a headwind to worker purchasing power.

The Shift Toward the Physical Economy

Goldman argues the next chapter involves steel and industrial automation. The bank says AI’s next frontier is the physical economy. This includes factories, mines, utilities, and oil rigs.

Estimates suggest $7.6 trillion will be invested globally between 2026 and 2031. Software accounts for less than 0.5% of global GDP. Therefore, the remaining economy represents a massive opportunity for deployment.

Urgency in corporate boardrooms is already palpable. Clients fear getting left behind if they do not move quickly. Capital is shifting fast from pure AI plays toward industrial applications.

Furthermore, technology companies are becoming more dependent on non-tech firms. This reverses a decade-long trend of digital self-containment. Convergence is now showing up in deal volumes and blurred industry lines.

Productivity Horizons and Workforce Impact

Economists maintain the real productivity payoff begins around 2027. Widespread adoption should eventually appear in labor data. Past estimates suggested significant annual growth increases over a ten-year period.

However, recent research complicates that optimism. Broad adoption could displace 6-7% of the workforce initially. While characterized as transitory, this framing remains contested among observers.

Gen Z excitement about AI has dropped significantly over the past year. Meanwhile, anger has risen sharply among daily users. Familiarity seems to breed frustration as much as enthusiasm.

Policymakers face timing risks regarding interest rates. Easing prematurely based on unmet optimism could sustain elevated inflation. The gap between expectation and reality cuts both ways.

Implications for Investors and Enterprises

The data center build-out is real and accelerating. Yet economic value accrues disproportionately to Asian manufacturers and utilities. It does not currently boost US productivity metrics.

Infrastructure must generate massive revenue just to achieve modest returns. Current revenues fall far short of capital expenditure projections. This gap is the central financial question of the era.

Enterprise buyers face a double-edged sword. Specific workflows show real gains, but unmanaged consumption burns budgets. Productivity payoffs require deliberate deployment and active measurement.

Additionally, competitive dynamics remain in flux. Chinese providers capture global inference demand at lower costs. The import problem identified at the hardware layer may repeat itself at the model layer.

Real Investment with Deferred Returns

Goldman’s research is structural rather than simply bullish or bearish. The infrastructure being built is genuinely transformative. The 2027 horizon is a reasonable forecast.

Nevertheless, the near-term math is unambiguous. The US transferred manufacturing benefits abroad while introducing new inflationary pressures. Aggregate productivity gains have not yet materialized.

The key question is whether returns will close the gap soon. Capital markets may decide the wait is taking too long. Investors must weigh real investment against deferred economic results.