The market’s reassessment of artificial intelligence did not begin with a collapse in belief. It began with arithmetic. Investors reopened their models, recalculated capital expenditure, and examined the rising cost of memory, electricity, data centres, cooling systems, and financing. The central question changed. AI demand was no longer enough. The market wanted to know how much profit would remain after the infrastructure bill was paid.
That shift became visible in mid-July 2026. On 16 July, Nvidia fell 2.4%, Micron declined 5.6%, Sandisk dropped 12.6%, and Western Digital lost 9.2%. South Korea’s Kospi fell 6.4% as pressure spread to Samsung Electronics and SK Hynix. On 17 July, the Nasdaq lost 1.4%, the S&P 500 fell 1%, and the PHLX Semiconductor Index moved more than 20% below its June peak, entering a technical bear market.
A Correction With a Deeper Message
Part of the sell-off was a normal response to exceptionally strong gains. Several AI-linked stocks remained significantly higher for the year even after the decline. However, the more important development was a separation between AI demand and AI investment returns.
Demand for computing power can remain enormous while shareholder returns weaken. Every additional AI workload requires advanced processors, high-bandwidth memory, storage, networking equipment, buildings, cooling, land, and electricity. If those inputs become more expensive, higher capital expenditure may buy less capacity than expected. Growth in spending can therefore reflect cost inflation rather than stronger future economics.
AI Is Becoming an Industrial Cycle
The current AI boom increasingly resembles an industrial investment cycle, not a traditional software expansion. Software businesses can distribute one product to millions of users with limited additional cost. AI infrastructure requires physical assets at every stage.
This changes who benefits first. Chipmakers, memory producers, equipment suppliers, data-centre operators, utilities, banks, and infrastructure funds can receive revenue before AI services prove their long-term profitability. The buyers of computing capacity carry a different burden. Technology groups and AI start-ups must show that customers will pay enough to cover an increasingly expensive asset base.
Alphabet, Amazon, Microsoft, and Meta can support very large investment programmes. Yet even the strongest balance sheets face investor scrutiny. Upcoming results will be assessed through more than revenue and earnings. Markets will examine contracted spending, installed capacity, utilisation, pricing power, margins, and free cash flow.
Energy and Finance Are Now Core AI Variables
The economics of AI are also becoming more sensitive to energy markets and interest rates. On 20 July, Brent crude traded near $89 per barrel, while average US gasoline prices moved above $4 per gallon. Higher energy prices can reinforce inflation concerns and lift bond yields, reducing the present value of future technology profits.
The direct impact is equally important. Data centres require large, stable electricity supplies. Grid connections, new generation, transmission upgrades, cooling, and water access can delay projects and raise costs. As governments consider how to prevent data-centre expansion from increasing household utility bills, AI is becoming a public-infrastructure issue as well as a technology theme.
Debt adds another layer. Large data-centre projects increasingly rely on project finance, private credit, bonds, and long-term customer contracts. This creates opportunities for lenders, but also connects AI expectations to broader financial markets. If utilisation or pricing disappoints, pressure could spread beyond equities into infrastructure debt and private financing.
Three Possible Market Paths
The first scenario is a soft correction. Major technology companies confirm strong demand, maintain margins, and demonstrate spending discipline. In that case, July’s decline becomes a reset, and companies with scarce technology recover fastest.
The second scenario is a longer repricing. Capital expenditure continues to rise, but revenue and free cash flow fail to keep pace. AI remains strategically important, while valuation multiples fall and capital rotates towards more resilient sectors.
The third scenario is infrastructure stress. Memory, energy, construction, and financing costs rise faster than expected, while customer demand proves less profitable. The impact then extends from chip stocks to data centres, utilities, lenders, and highly leveraged suppliers.
The New Standard for AI Investment
The July sell-off does not prove that the AI cycle is ending. It shows that the market is becoming more selective. Investors are moving from a simple question about demand to a harder question about returns.
The strongest companies will not only deliver faster chips or larger models. They will control costs, secure energy, protect margins, manage debt, and convert infrastructure into measurable cash flow. AI may still define the decade. But the next phase will be determined less by technological promise and more by the economics required to sustain it.