Barclays Survey Confirms AI Is Now Mission-Critical for Institutions as Andreessen Warns of Energy Bottlenecks

A comprehensive survey commissioned by Barclays has revealed that artificial intelligence has firmly embedded itself into the daily operations of institutional investors worldwide. The findings arrive alongside a pointed observation from venture capital heavyweight Marc Andreessen, who argues that power infrastructure — not software — will ultimately determine how far AI can advance.

The Barclays study gathered responses from 410 fixed-income investment professionals spanning North America, Europe, the Middle East, and Asia. Its central conclusion: AI has graduated from the pilot phase and is now a routine part of institutional workflows, even as human judgment continues to govern final decisions.

**Research Dominates AI Adoption Across Investor Types**

When it comes to specific applications, research and data analysis top the list. Approximately 52% of long-only managers and asset owners identified research as their primary AI use case. Among hedge funds, around 44% rely on AI chiefly to process and interpret market data.

Hedge funds lead the pack in terms of usage intensity. A striking 72% of hedge fund respondents reported using AI on a daily basis. That figure stands well above the 49% recorded among long-only managers and the 38% seen among asset owners — a gap that reflects the broader trend of aggressive AI integration in more actively managed strategies.

Despite this momentum, trading and execution functions have seen little disruption. Most survey participants described AI's influence on those areas as minimal at best. Data security emerged as the single biggest obstacle preventing more widespread adoption, outranking concerns about accuracy or regulatory compliance.

On the workforce front, the survey paints a reassuring picture. A mere 7% of respondents anticipated meaningful reductions in headcount as a result of AI adoption. The majority instead foresee improved productivity with staffing levels largely unchanged.

**Andreessen Links AI's Ceiling to Physical Infrastructure**

Marc Andreessen, co-founder of venture firm Andreessen Horowitz, recently offered a provocative framing of AI's long-term constraints. In a widely circulated post, he introduced what he called the AI:AC Hypothesis: "In the future, in each country, the amount of AI will be proportional to the amount of AC. And vice versa."

The metaphor cuts to the heart of a growing infrastructure problem. AI data centers consume enormous quantities of electricity, and the cooling systems required to prevent hardware from overheating add another significant layer of energy demand. This dual burden is driving a rapid and steep increase in global electricity consumption tied directly to AI operations.

The International Energy Agency projects that data center power demand will more than double by 2030, reaching approximately 945 terawatt-hours annually — a figure comparable to Japan's entire current national electricity consumption. The United States faces particular pressure. According to IEA estimates, American data centers could soon consume more power than the country's aluminum, steel, and cement industries combined. Geographic regions offering affordable, dependable power are expected to attract a disproportionate share of AI infrastructure investment.

**What This Means for Capital Allocation**

Taken together, the Barclays survey data and Andreessen's infrastructure argument converge on a single investment thesis. Institutional demand for AI is not theoretical — it is already here and growing. But the energy and cooling constraints Andreessen highlights will determine which companies and regions emerge as long-term winners.

The largest technology companies are already responding with unprecedented capital commitments. Microsoft, Amazon, Alphabet, and Meta have collectively outlined approximately $725 billion in capital expenditure guidance for 2026, representing a 77% increase compared to current-year spending. These hyperscalers are effectively betting that building out physical AI capacity now will pay dividends as demand scales.

Whether electrical grids can keep pace with that ambition remains an open and consequential question — one that may shape not only the trajectory of AI development, but also the investment returns that follow.