The Material Turn

Why Wall Street Is Pricing AI as Infrastructure

As the generative hype cools, a new industrial reality emerges. From BlackRock’s energy bets to Berkshire’s quiet pivot, Wall Street is remapping AI from a Silicon Valley software story to a multi-trillion-dollar infrastructure cycle defined by megawatts and physical constraints.

For much of the past decade, artificial intelligence was narrated as a triumph of software — an exponential curve of smarter models, larger datasets and frictionless digital scale. Intelligence appeared to detach from geography, material limits and even from the industrial logic that governed earlier technological revolutions. The cloud, in this telling, was weightless.

Early 2026 is dissolving that illusion. Warehouses of processors draw as much electricity as small cities. Cooling systems consume rivers of water. Transmission bottlenecks delay deployments more effectively than any shortage of code. The constraint on artificial intelligence is no longer conceptual sophistication but physical capacity — how much power can be generated, moved and dissipated without destabilising the systems that support it.

Markets are beginning to reflect this reality. Valuations once driven by narrative momentum are being recalibrated around capital expenditure, grid access, thermal efficiency and land availability. Intelligence, it turns out, scales physically, not digitally.

“We believe there will be trillions of dollars of investment needed in infrastructure related to power grids and AI. Infrastructure is at the beginning of a golden age.”
— Larry Fink, Chairman & CEO, BlackRock

Fink’s assessment is less a prediction than a recognition of a structural shift already underway. The world’s largest asset manager is not repositioning toward electricity infrastructure because it is fashionable; it is doing so because electricity has become the binding constraint on the most celebrated technology of the era.

From Models to Megawatts

The first phase of the AI boom was defined by logical efficiency — improvements in algorithms, architectures and training techniques that made models more capable per unit of computation. The emerging phase is dominated by thermal efficiency: how effectively physical systems convert electricity into computation without overheating, failure or prohibitive operating costs.

This distinction is not academic. Logical efficiency can be improved through research; thermal efficiency requires steel, copper, cooling towers, substations and permitting processes. It transforms AI from a software industry into a heavy industrial ecosystem.

Utilities, transformer manufacturers, cooling specialists and grid operators — historically stable, low-growth sectors — now sit at the centre of strategic capital allocation. The speed at which new generation capacity can be brought online increasingly determines the pace at which AI capabilities can expand.

“The bottleneck for scaling AI has shifted from model capacity to speed to power. Intelligence is becoming a utility.”
— Strategic report, The Physical Bottleneck, Financial Content Markets (Feb 2026)

The phrase “speed to power” captures a new metric of competitiveness. In the early cloud era, speed to market mattered. In the AI infrastructure era, speed to electricity may matter more.

The Repricing of the Picks and Shovels

Technological revolutions rarely reward only the companies that produce the most visible innovations. Railroads enriched steel producers, land developers and financiers as much as locomotive manufacturers. Electrification transformed copper mining and utilities before it transformed consumer appliances.

AI appears to be following the same pattern. Investors are rotating from companies selling applications to those controlling the physical layer: semiconductors, data-centre construction, power equipment and energy supply chains. The market is rediscovering the logic of “picks and shovels” — the suppliers of essential inputs to a boom rather than its most glamorous participants.

This shift also reflects a deeper reassessment of risk. Software valuations depend heavily on expectations of future adoption and pricing power. Infrastructure assets, by contrast, generate predictable cash flows once operational, often backed by long-term contracts or regulated returns. In an environment of uncertainty, predictability itself becomes a premium asset.

“Investors are no longer willing to reward all AI exposure equally. Capital is rotating toward the physical layer where operational profit growth is visible.”
— Ryan Hammond, Senior Analyst, Goldman Sachs Research

Such repricing does not signal the end of AI innovation. It signals the maturation of the market’s understanding of where value will ultimately accumulate.

Capital at Scale: The Berkshire Signal

If BlackRock represents the world’s largest allocator of capital, Berkshire Hathaway represents something different: the archetype of patient, conservative capital. Its decisions are often interpreted less as tactical moves than as indicators of long-term structural confidence.

Greg Abel’s succession to Warren Buffett in January 2026 marks the beginning of a new era for the conglomerate. Early signals suggest not a departure from Berkshire’s cautious ethos but a recalibration of where durable value is likely to reside. A multibillion-dollar position in Alphabet reflects exposure not merely to advertising revenue but to cloud infrastructure, data centres and AI platforms embedded in the physical economy.

Within Berkshire’s operating companies, AI is being deployed less as a product than as an efficiency multiplier for large-scale systems — insurance underwriting, energy distribution and freight logistics.

“There is no question that AI will be a game-changer. It will affect how we manage risk and pay claims.”
— Ajit Jain, Vice Chairman, Insurance Operations, Berkshire Hathaway

Even more revealing is Berkshire’s exposure to rail and energy infrastructure. Optimising train networks through predictive analytics does not create a new digital service; it moves physical goods more efficiently across a continent. Algorithms become tools for improving throughput in systems built from steel and fuel.

Abel himself has emphasised the scale of investment required to support rising demand from data centres.

“The challenge for utilities is the enormous capital investment needed to modernize the grid and meet demand.”
— Greg Abel, CEO, Berkshire Hathaway

When capital that has historically favoured stability begins preparing for massive infrastructure spending, it suggests that the AI boom is entering a phase defined less by experimentation and more by construction.

Volatility and the End of Pure Hype

Financial markets in early 2026 exhibit a distinctive pattern: enthusiasm for AI remains strong, but tolerance for disappointment has evaporated. Companies that exceed expectations are rewarded disproportionately, while those that fall even slightly short experience sharp corrections.

This asymmetry reflects a transition from narrative-driven valuation to evidence-driven valuation. During the initial hype cycle, the mere association with AI could sustain elevated prices. As capital expenditures mount and timelines extend, investors demand proof of profitability.

At the macro level, some institutions remain bullish, arguing that AI-driven productivity gains could justify historically high market multiples. Yet even optimistic forecasts increasingly assume massive investment in energy systems, manufacturing capacity and supply chains — conditions more typical of industrial expansions than software booms.

AI, Trust and Systemic Risk

Beyond economics, AI introduces new vulnerabilities into financial systems themselves. The capacity to generate convincing synthetic voices, images and documents challenges the mechanisms through which markets verify information.

Warnings from major institutions about impersonation and fraud underscore that AI can erode trust as easily as it can create efficiency. In markets built on confidence, uncertainty about authenticity can propagate rapidly, amplifying volatility.

The issue is not merely criminal misuse but epistemic instability — difficulty determining what is real. In such an environment, authority figures and trusted institutions become targets for manipulation precisely because their credibility carries economic weight.

The Industrialization of Intelligence

Taken together, these developments point to a fundamental reframing of artificial intelligence. Rather than a discrete sector, AI is becoming an infrastructure layer embedded across the economy — comparable to electricity, rail networks or telecommunications.

This transformation has geopolitical implications. Regions with abundant, reliable energy supplies and favourable regulatory environments are positioned to attract data centres and the capital that accompanies them. Jurisdictions constrained by power shortages or permitting delays risk technological marginalisation regardless of their software talent.

AI capacity may thus evolve into a measure of national power, analogous to oil reserves in the twentieth century. The competition is not only between companies but between energy systems.

“AI investments will continue, but more capital will flow into energy infrastructure than into AI companies themselves.”
— BlackRock Investment Directions 2026

The concept of “sovereign compute” — the ability of a nation or region to host and sustain large-scale AI operations — is emerging as a strategic priority. In this context, megawatts become as consequential as microchips.

Conclusion: Intelligence Meets Physics

The narrative of artificial intelligence as an ethereal software phenomenon is giving way to a more prosaic but more consequential reality. Intelligence at scale requires enormous physical support systems: power plants, transmission lines, cooling infrastructure, specialised facilities and vast quantities of capital.

Wall Street’s recalibration reflects a recognition that technological revolutions ultimately obey the laws of physics and economics. The companies and regions that can supply energy reliably and afford sustained investment may capture disproportionate benefits, regardless of who produces the most elegant algorithms.

The decisive question of the AI age may therefore not be who designs the smartest systems, but who can finance, power and maintain them.

In the end, intelligence is not escaping the material world. It is becoming one of its most energy-intensive expressions — constrained by thermodynamics, enabled by infrastructure and priced by capital markets accordingly.

Photo credit:
AI-generated illustration (DALL·E / OpenAI)

Caption:
Stylized depiction of Wall Street’s iconic Charging Bull in the colors of the American flag — symbolizing the fusion of finance, national power and industrial-scale capital in the emerging AI infrastructure era.