There is $667 billion being spent on AI infrastructure in 2026 , more than the entire GDP of Sweden , and Goldman Sachs just delivered the most uncomfortable finding in corporate America: at the economy-wide level, it is not showing up in productivity at all. Not a little. Not yet. Not meaningfully. The number Goldman used was starker than that: the AI boom boosted the US economy by "basically zero" in 2025. And 2026, so far, is telling the same story.
What Actually Happened
During Q1 2026 earnings season, Goldman Sachs analysts systematically tracked every corporate disclosure about AI's measurable operational impact. The findings landed like a cold shower on a market that had priced in transformation. Goldman found "no meaningful relationship between AI adoption and productivity at the economy-wide level." Tech giants had by then revised their 2026 capital expenditure expectations to $667 billion , a 24% increase from the start of earnings season and a 62% jump compared with full-year 2025 capex. Goldman's chief economist was unsparing: AI "boosted the US economy by basically zero in 2025," adding that there had been "a lot of misreporting of the impact that AI investment had on GDP growth."
The data behind that judgment is specific. Only 10% of S&P 500 management teams quantified AI's impact on any specific business function. A mere 1% quantified any impact on earnings. While roughly half the companies in the broader Russell 3000 mentioned AI during earnings calls, fewer than 20% of establishments in the US economy are actually using AI for any business function at all. The gap between narrative and deployment is not a rounding error , it is the central fact of the AI economy in 2026.
Why This Matters More Than People Think
The $667 billion being deployed this year represents the largest coordinated capital allocation bet in corporate history , exceeding the combined IT investment of the entire dot-com boom at its peak. Goldman estimates this spending will contribute roughly 1.5 percentage points to measured capex growth in 2026, but the net impact on overall GDP growth will be just 0.1 to 0.2 percentage points. That ratio , a 62% increase in AI investment generating a 0.1-point GDP contribution , is an investment efficiency number that no CFO in any other context would accept.
What makes this genuinely alarming is the psychology driving the spending. Goldman's own research found that FOMO , fear of missing out , "has proven a stronger incentive than poor stock performance" in driving AI investment decisions. Companies are not investing because they have measured returns. They are investing because they fear being structurally disadvantaged if they do not. That is not a rational investment thesis. It is a collective action problem playing out across the Fortune 500, and Goldman Sachs is the first major institution to document it in earnings data rather than survey responses.
The Competitive Landscape
The 30% productivity gains Goldman did find exist only in two specific domains: software development and customer support. These are not random choices , they are the two areas where AI deployment has been most mature, most measurable, and where workflows were most amenable to AI augmentation. Microsoft Copilot deployments in enterprise development teams have generated consistent 30-40% code throughput improvements. Salesforce Agentforce has documented substantial call deflection rates in enterprise customer support. These gains are real. They simply do not cascade.
The reason is structural: achieving a 30% AI productivity gain requires entire workflow redesigns, not just tool adoption. Companies seeing those gains did not install a chatbot , they restructured engineering organizations, changed code review processes, and rewired support escalation trees. That level of transformation is happening in fewer than 1 in 10 companies. Meanwhile, the 90% that purchased AI tools without restructuring report no measurable productivity change , and their capex is included in the $667 billion figure driving Goldman's "no meaningful relationship" conclusion.
Hidden Insight: The Solow Paradox Is Back, and the Stakes Are an Order of Magnitude Higher
In 1987, Nobel laureate Robert Solow wrote the sentence that defined a generation of technology economics: "You can see the computer age everywhere except in the productivity statistics." It took another full decade before IT investment finally appeared in aggregate GDP numbers , and when it did, it triggered the longest peacetime economic expansion in American history. Goldman Sachs is describing the identical phenomenon in 2026. But there is a critical difference: the capital being deployed is vastly larger.
The dot-com boom generated roughly $500 billion in total IT investment over five years. AI infrastructure is on track to deploy that amount in a single year in 2026 alone. The lag between investment and economic return is being compressed by the sheer scale of deployment , but it is still a lag. Goldman's own long-run models suggest meaningful AI productivity contributions begin appearing in aggregate BLS statistics between 2027 and 2029, with compound effects through the early 2030s. The question is not whether the payoff comes. The question is who is still solvent when it does.
The uncomfortable truth that Goldman's research surfaces is one the AI industry has been careful to avoid discussing publicly: the companies achieving 30% productivity gains in specific functions are not broadcasting how they achieved it. They are treating the methodology as competitive intelligence. The broader market , the 90% seeing no measurable gain , is not failing because AI does not work. It is failing because the organizations that cracked it have no incentive to share the answer. The $667 billion being spent in 2026 is, in large part, paying for lessons that a handful of companies already learned in 2024.
What to Watch Next
The most important leading indicator over the next 90 days is enterprise software renewal rates for AI-powered productivity tools. If companies are not renewing Microsoft Copilot, GitHub Copilot, or Salesforce Agentforce licenses at Q2 renewal cycles, it is a direct signal that the productivity gains are not materializing at scale. Second: watch the Bureau of Labor Statistics quarterly productivity releases for the information services and finance sectors , the two industries with highest AI adoption rates. If sector-level productivity does not begin diverging from the economy-wide average by Q3 2026, the 2027-2029 Goldman timeline for aggregate gains will need to be revised out further.
The 180-day prediction: at least one Fortune 100 company will publicly quantify a 30%+ AI productivity gain in a specific business unit and it will become a case study that accelerates enterprise AI restructuring in that sector. This disclosure , when it comes , will not be altruistic. It will be a recruiting signal, a customer signal, and a competitive signal simultaneously. The companies that have cracked AI productivity will eventually have to show their work, because silence becomes its own kind of competitive liability when the market starts asking harder questions about what $667 billion actually bought.
The 30% productivity gains from AI are real , the problem is that only 1 in 10 companies knows how to find them, and $667 billion is not buying that knowledge.
Key Takeaways
- No economy-wide impact , Goldman Sachs found no meaningful relationship between AI adoption and productivity at the macro level in its Q1 2026 earnings analysis.
- $667 billion in 2026 AI capex , a 62% jump over 2025 , yet Goldman estimates only a 0.1 0.2 percentage point contribution to GDP growth.
- 30% productivity gains are real but narrow , they exist only in software development and customer support, the two most mature AI deployment domains.
- 1 in 10 companies quantified anything , only 10% of S&P 500 management teams quantified AI's impact on specific operations; 1% quantified earnings impact.
- Goldman chief economist: AI boosted the US economy by "basically zero" in 2025 , calling existing GDP impact reporting "a lot of misreporting."
Questions Worth Asking
- If only 1% of S&P 500 companies can quantify AI's earnings impact, what does that say about the quality of board-level AI strategy at the other 99%?
- The Solow Paradox resolved into the longest expansion in US history , but that resolution took 10 years. Can the current AI investment cycle sustain itself for a decade before the productivity payoff arrives?
- If the companies achieving 30% AI productivity gains are treating their methodology as competitive intelligence, what is your organization doing to find those lessons before your competitors do?